s Anyone Home?
An Investigation Into Occupancy Detection for Home Automation and Why Capacitive Sensing Could be the Solution.
With conclusive evidence indicating buildings account for 40% of CO2 emissions, ensuing studies have also revealed that energy consumption of buildings can be reduced by up to a third through home automation. However, with cameras being intrusive and motion sensors being unable to detect motionless occupants, the potential of home automation remains largely untapped. This paper set out with the aim of identifying a solution to detect static occupants non-intrusively in real-time.
A review of current sensing modalities was conducted to ultimately propose capacitive sensing, in all its crudeness, as a novel but promising avenue. Stipulations from desk based research served as initial evidence and subsequent empirical observations of a demonstration using a model confirmed a capacity to detect static presence non-intrusively in real-time. To further progress prior work, attention was then drawn towards addressing the key unanswered questions. If the sensor detects a change in the environment based on a difference in baseline charge time, how would the sensor infer occupancy if new objects such as furniture enter the observed space? Furthermore, how would the sensor be integrated into households and at what cost? The investigation was divided accordingly by initially performing an analysis to induce a relationship between charge time and distance to develop a localisation system to determine size of the object. A subsequent examination of the potential methods of integration of the sensor into households treated the latter question. The findings of the analysis yielded a relationship between charge time and distance to successfully localise and follow an object in a three-dimensional space. This localisation method cannot be applied to infer size as it requires a calibration step of a known an object at a given distance. However, at low frequencies a capacitive sensor is not sensitive to furniture and common electronic devices such as computers. It is therefore concluded that capacitive sensing demonstrates not only a capacity to detect static presence non-intrusively but also to track occupants, which provides compartmentalisation capabilities and effective automation of spaces. The examination of potential methods of integration led to the conclusion that a conductive paint surfacing or modular panels is the optimal approach, with a key benefit of having few dead spots in contrast to off-the shelf sensors. In relation to cost however, it is concluded that the technology is at an early stage and therefore, identifying a cost bracket would be misleading due to an extensive number of uncertainties.
Key Words: Home automation, Smart home, Capacitive sensing, Occupancy detection, Static presence, Non-intrusive sensor, Real-time occupancy information, Occupant localisation, Conductive paint, Modular panels
The student would like to thank Dr. Sukumar Natarajan for his invaluable guidance, insight and support. Particularly in imparting me with his wisdom to treat failure and success as equal in the scientific world, where each provide necessary lessons for all to benefit from. I would equally like to thank Dr. Hugo Jenks for his patience and generosity of time in providing technical support and guidance. Finally, I would like to thank my Uncle, Dr. David Stokoe, who’s unique background as a Doctor stimulated original thinking over conversations on the topic.
Figure 1.1 Diffusion of innovations (Stanford Web, 2003)…………………………….
Figure 2.1 Occupancy information levels (Teixeira, Dublon and Savvides, 2010)…………….
Figure 2.2 The abandoned Presidio Modelo in Cuba – Photographed in 1995 (McMullan, 2015)….
Figure 2.3 Activity recognition using built-in features of the smartphone (Lara and Labrador, 2013).
Figure 2.4 Illustration of network representing a complex non-linear system……………….
Figure 2.5 How proximity sensing functions for occupancy detection (Basic Electronics Tutorials, sd)
Figure 2.6 Nodal overview of complexity of network for capacitive sensing………………..
Figure 3.1 Schematic of electronics for charge time analysis – Schematic has been drawn using Fritzing software to justify the generic Arduino
Figure 3.2 Sequence of set-up for connection to the sensor plate………………………
Figure 3.3 Testing methodology for one direction.
Figure 3.4 Relationship of charge time to distance…………………………………
Figure 3.5 Distance against 1/√τ………………………………………………
Figure 3.6 Schematic of electronics for 3D location
Figure 3.7 Connection to microcontroller for 3 sensor plates…………………………..
Figure 3.8 Full assembly of model for demonstration………………………………..
Figure 3.9 Visual display of localisation capacity in 3D………………………………
Unless otherwise cited in the caption of thefigure, student is author of the content.
|HVAC||Heat, Ventilation and Air-Conditioning|
|LIDAR||Light Detection and Ranging|
|USB||Universal Serial Bus|
|SAR||Specific Absorption Rate|
|CSV||Comma Separated Variables|
|τ||Time constant in seconds|
|R||Resistance in ohms|
|C||Capacitance in Farads|
|d||Distance between two charged particles|
|q||Charge of particle|
|F||Force of attraction or repulsion of charged particles|
As the technological wheel continues to accelerate, this past decade has seen the birth of the smart home and all its promises. The house of tomorrow assures the consumer of an enhanced experience where turning on lights or locking the door would be a thing of the past. 78% of global users have expressed interest in the smart home and adopting the technology to improve their home security, comfort and energy consumption (GFK, 2016).
One of the key motivations to adopt the smart home lies in home automation, where studies have indicated that the energy consumption of buildings can be reduced by up to a third (Jiang, et al., 2016). Moreover, according to the international energy agency, buildings account for up to 39% of emissions (International Energy Agency, 2016). These figures combined provide an opportunity to address the pressing climate change agenda; one of the key 20-20-20 targets to keep energy CO2 emissions under control is to generate a 20% improvement in energy efficiency by 2020 (European Commission, 2016). The increase in efficiency of energy consumption can provide the user with the enticement of a financial reward as well as the moral satisfaction of helping abate the issue of global warming. In the spirit of innovation, independent of the financial or moral motivations, there is equally the stimulus to develop the technology for the sake of evolution, or as Nikola Tesla put it simply “The progressive development of man is vitally dependent on invention” (Tesla, 2011).
At the heart of home automation is the ability to sense a human’s presence, where occupancy information can commonly be applied to automate indoor lighting systems and Heat, Ventilation and Air-Conditioning (HVAC) systems (Arens et al., 2005). More advanced human sensing in the future may make a more comprehensive analysis of the human. A medical application could ask itself “Who is in this room? Is Alfie in the room? If so, what is his body temperature and heart rate?”. If Alfie is found to be ill, the sensor may also be able to answer the question “Who has he been in contact with in the past 24 hours?”. This technology is in fact not as distant as it may seem, for instance (Medvedev et al., 2016) have been able to successfully demonstrate that a chemosensor could be used to analyse blood glucose levels as well as airway inflammation in relation to asthma through the chemical composition of the breath.
There are further exciting applications to human-sensing but for the scope of this paper, an investigation of current sensor modalities will be performed to highlight the key challenges and limitations of current technology to subsequently identify capacitive sensing as a potential solution. This will be achieved through a desk-based study and will be proceeded by a deeper study into capacitive sensing with aim of concentrating on previously unexplored areas of the technology to help answer key questions. Namely, if the sensor detects a change in the environment based on a difference from baseline charge time, how does the sensor differentiate an occupant to another conductive object in the observed space? Furthermore, how would the sensor be integrated into households and at what cost? Before jumping directly into the investigation, it is helpful to provide context by establishing what is envisaged by home automation.
At times referred to as energy intelligent buildings or building energy and comfort management systems, it refers to a system that has as objective to create and maintain comfort requirements of an internal space for the occupant whilst increasing efficiency of energy consumption during building operations (Zipperer et al., 2013). This capacity to control a building, or in fact groups of buildings is achieved by technological means whereby the system can be broken down into its micro and macro components.
The micro system relates to the individual units of equipment coupled with microprocessors for data storage, monitoring and control/ communication. These microprocessors can be connected to the building services equipment, commonly HVAC systems, lighting and domestic water services (Nguyen & Aiello, 2012).
This distributed network of microprocessors can then be connected to a centralised computer to form the macro system, whereby the centralised computer can communicate with the microprocessors, receiving as well as outputting information in a form of task delegation. The ability to control the macro system through the computer is fundamentally where home automation emerges and the potential for improvements in efficiency emerge. An experiment performed by a company in Minnesota, 3M, at their own offices involved turning off all office devices and lights that were not in use. The results demonstrated a 26% saving of electrical energy in 15 minutes (Arens et al., 2005). This example of course required the manual control of the services not in use, but by application of the automation principle, (Nguyen and Aiello, 2012) have demonstrated that occupancy based control can result in up to 40% in energy saving for both HVAC and lighting systems. Henceforth, home automation will be perceived under this light.
Despite the technological progress of the past decade and the quantifiable benefits of adopting a home automation system, the potential remains largely untapped. Why does it remain so elusive to the 21st century household and more to the point, how is occupancy detection limiting it?
This question can be initially explored through the lens of diffusion of innovations, a theory that was originally articulated and developed by professor of communication studies Everett Rogers in his book Diffusion of Innovations. This process of diffusion relies primarily on the technology being adopted to self-sustain and outlines the process of implementation in society. The adopters in this case are individuals for which the technology is targeted at and are categorised into innovators, early adopters, early majority, late majority and laggards. As the technology moves through the consumer types, its market share and hence diffusion increases (Stanford Web, 2003).
Full diffusion of a technology would imply a 100% market share on the cumulative distribution curve shown in yellow in Figure 1.1, but it has been shown that reaching the point of inflection or the transition of early majority to late majority carries a momentum for continued growth. Placing a product into the market place highlights that a technology equally needs to be able to survive commercially. As Seth Godin explains cleverly with sliced bread, it was only 15 years after the slicing technology was originally created that it gained popularity following a change in marketing strategy, enabling a movement through the consumer types and for the expression “the best thing since sliced bread” (Seth Godin: How to get your ideas to spread, 2003). Is occupancy detection then just a case of sliced bread, where the technology is ready but it is a marketing strategy that could trigger widespread adoption? As (Zenezini, et al., 2016) discovered, the key barriers faced by occupancy detection at present relate to;
- Usability, functionality and reliability
- Difficulty of integration in existing households
Emphasis was drawn to how the human sensor element of home automation at present is incapable of overcoming these barriers and finding a balance between reliability of information, integration and cost, hinting that this is not a case of poor marketing as with sliced bread. To clarify between point i. and ii., the first barrier relates to the performance capacity of providing the occupant information whereas integration works on a human level. It considers the ease of installation and adaptability; is it unappealing to see in the home or does it make the homeowner uncomfortable? By examining the current human sensing technology and related prior work with consideration of these barriers, it is anticipated that a sense of confidence will be developed in underlining why capacitive sensing is a promising solution.
To begin the survey of current sensing technology, Table 2.1 attempts to qualitatively list an extensive choice of sensor modalities with various capacities.
Table 2.1 Overview of Sensor Modalities – Adapted from Table 1 in (Teixeira, Dublon and Savvides, 2010)
|Image based sensors||Camera, thermal imager||Nest camera, Flir One, Melexis, Irisys Gazelle|
|Instrumented sensors||Device-to-device ranging, environmental recognition and ID sensors||Wifi fingerprinting, smartphone recognition, radio pairs|
|Motion sensors||Binary, pressure, vibration and inertial sensors||PIR, breakbeam, accelerometers, piezo-resistors, magnetometers|
|Doppler-shift radar sensor||Radios, ultrasound transducers||Doppler radar|
|Chemosensors||CO2, temperature, humidity||K30 CO2 Sensor|
|Electric field sensors||Capacitive proximity sensors||Capacitive floor tiles, capacitive seat detection|
To narrow the range of potential sensor modalities worth studying, it is necessary to outline what information the sensor is expected to provide. Figure 2.1 details a range of information into five levels; presence, count, location, track and identity.
In consideration of home automation, despite using cameras to achieve 40% energy savings for the HVAC and lighting systems, (Nguyen and Aiello, 2012) recognised that occupancy information relating to presence and count was sufficient as beyond that level there was no noticeable difference in efficiency. In addition, this level of information allows for zoning of internal spaces to enable different thermal requirements to be compartmentalised when required. Would it not then make sense to just adopt a camera as the sensing modality?
220.127.116.11 Cameras and Intrusiveness
With the emergence of the driverless car, cameras have been at the core of the development of the technology and consequently the most sophisticated human sensing has been born in this field of research. In order to build an accurate image of an environment that is both complex and dynamic, these cars are fitted with numerous sensors of high intelligence. Sensors such as Light Detection and Radar (LIDAR) help detect the lines on a road and radars monitor nearby cars for instance. Cameras are at a stage where they can not only detect a pedestrian, but even understand the mechanics of the legs to infer type of motion (walking or running), relative speed and direction of motion to ultimately determine if the pedestrian is likely to cross the road (Surden, 2016) (Armstrong, 2017). This technology has now trickled through to the home, where companies like Nest have adopted the camera and complimented it with Nest Aware software, to not only monitor the home but also successfully detect intruders by facial recognition (Nest, 2016). Despite the brilliant sophistication of these technologies, it is the very fact they are capable of building such accurate images that limits the camera.
A cameras true limitation lies in the psychological impact of its presence rather than performance, simply explained through the concepts of Panopticon and Cyber-security. The concept of Panopticon was originally developed by philosopher and social theorist Jeremy Bentham for implementation in prisons. His design of prisons enabled inmates to be observed at all times by a prison guard (Foucault, 1975).
As can be seen in Figure 2.2, the watchtower at the centre of the building had a view into all cells, where the guard can shine a light into the cells without the prisoner ever being able to know if he/ she is being observed. Otherwise put in the words of Michel Foucault, “He is seen, but he does not see; he is an object of information, never subject in communication.” (Foucault, 1975). As a result, the inmate’s uncertainty to being observed leads to self-policing.
The relevance of the Panoptican concept becomes clearer following a consideration of the inherent risk of cyber-security in home automation. A recent risk analysis of a smart home automation system at Malmö University identified cyber-security as an obstacle in the “vision of smart, energy-efficient homes and buildings” (Jacobsson, Boldt and Carlsson, 2016). Still, this risk is not exclusive to sensors; in November 2016, 1.3 million android smartphones were hacked (Pagliery, 2017) yet in the same year smartphones hit record sales of 1,495.36 million units (Statista, 2017). This highlights that despite the inherited risk of any cyber connected device, consumers will accept the hazards in view of the benefits. As previously discussed, home automation provides clear financial and moral motivations and provided the system adopts acceptable security measures, it is reasonable to assume consumers will make the same the trade-off. However, cameras possess a unique danger independent of the inherent threat of any web-connected electronic system. A recent article by Energy and Technology magazine broached the topic of a hack of millions of web cameras worldwide made by botnets, resulting in major manufacturers such as China’s Hangzhou Xiongmai having to recall 10,000 devices (Pultarova, 2016). More precisely, it also explained why cameras are deliberately targeted by hackers due to the information they provide; an image of a personal or intimate space. In April 2014, a couple from Ohio installed a Foscam web-connected camera in their child’s bedroom and were woken by the sound of a man shouting “Wake up, baby!” (The Economist, 2014). Stories like these are unique and further examples exist, but the detrimental impact of these stories brings back to light the psychological factor.
Placing the concept of panopticon in the age of digital technology and being mindful of the risks of the webcam, a user could feel a deep sense of uncertainty if they were to place a dead-eye lens in their home. Does this proposed psychological turmoil of a Panoptic system exist if people are not aware that they may be watched? A recent study performed by CamPatch academy of 250 computer users found that 49% were aware of webcam hacking (Rouse, 2012). Under this light, it can be argued that a portion of the general public can experience this uncertainty and hence the panopticon effect can apply in the digital era.
Whether a user will be willing to place a dead-eye lens in their intimate spaces for the benefit of home automation is subjective and therefore has an unclear outcome. The user in this case will have to be persuaded that the benefit outweighs the uncertainty of their privacy. To comprehensively reduce emissions, home automation needs to be maximised in both number of adopters but equally number of spaces in the home in which the sensor can be placed, including those most intimate.
Based on this argument of panopticon and cyber-security proposed above, it can be concluded that cameras are unable to overcome the barrier of integration due to intrusiveness. Furthermore, in relation to barrier iii. Cost (refer to section 1.2.1), using the Nest camera as an indicator of price range, at the current time of writing it has a retail value of £159 according to their online store. Installing a network of cameras throughout a home is expensive and hence has a limited audience. Yet Nest as a company was acquired by Google for $3.2 Billion back in 2014 (BBC, 2014), what is the industry’s take on occupancy detection and the issue of intrusiveness?
18.104.22.168 Thermal Imagers – Interview with Delta Dore
To investigate this question, an interview with Delta Dore founder Mr. Joël Renault helped provided an insight into the industry’s perspective (Renault, 2016). Delta Dore was originally founded in 1970 in Rennes, France as an electronics and communications subcontractor and has since grown into a multinational business offering a wealth of electronic technologies related to the smart home. As explained by Mr. Renault, in the past their main occupancy detection technology has relied on variations of the classic motion sensor and they have recently complimented this with a home camera technology for security surveillance. However, they are currently developing a new sensor, although it must be stated that there is an element of sensitivity in the topic, limited information could be provided due to the company itself currently developing a commercial static occupancy detector. What can be said is the sensor modality under consideration at Delta Dore is a thermal imager by means of infra-red. The sensor looks to measure the surface colour temperature of its view and infer occupancy from the number of pixels at a certain temperature. Still, there are two key limitations to this approach; i. cost ii. imaging of the space.
It is anticipated that this sensor will be sold as a high-end product due to the cost implications during the manufacturing process. The high cost stems from the company having invested the last year in developing and patenting a novel production method for the glass based on mechanosynthesis to ensure the thermal image of the space is as realistic as a camera.
- Imaging of the space
The high quality thermal image however, brings back to light the topic of intrusiveness. The projected use for the sensor is directed for commercial buildings rather than domestic, therefore the argument proposed by Mr. Renault is that in this scenario the privacy element diminishes in a professional environment and the benefits justify the means.
It must be considered that solving half of the problem is better than no solution at all. Re-iterating the problem under consideration, the combined global energy consumption of domestic and commercial buildings in developed countries is reaching 20 – 40% (Pérez-Lombard, Ortiz and Pout, 2008). Hence, reducing a portion of this remains preferable to neglecting the problem altogether, but what if a solution can be found for both domestic and commercial buildings without needing to impinge on the topic of privacy entirely?
22.214.171.124 Pixelated Thermal Imagers
Pixelating the thermal image is the theme explored by Beltran, Erickson and Cerpa in attempting to estimate occupancy non-intrusively (Beltran, et al., 2013). The Grid-Eye 8×8 thermal array is trialled whereby the computational aspect works off the temperature of the pixels similarly to Delta Dore; if a sufficient area is at the right temperature, occupancy is inferred. Tyndall, Cardell-Oliver and Keating produced a similar occupancy estimation sensor system but in this case opted for the Melexis MLX 90620 16×4 thermal array (Tyndall, et al., 2016). Both were capable of estimating occupancy with an error of
~0.35 persons but independent of functional capacity, thermal imaging remains expensive. The Panasonic Grid-Eye is available for roughly £45 yet covers a 6.25 m2 floor area at a height of 3m. To put this area into perspective, a standard tennis court covers a 195 m2, meaning 31 sensors would be needed to avoid any dead spots. A home equating to the floor area of a tennis court is modest but £1,395 to automate the liveable space is perhaps not. In addition, this would bear the inherited functional limitations of a pixelated thermal imager and whether the image produced in this case is too intrusive again remains a subjective matter. What if the sensor was not required to detect a human but rather a device?
This is the premise to instrumented sensors, which from the outset have the benefit of not treading on the issue of intrusiveness. As the name indicates, these sensor modalities require the occupant to be carrying a device that can communicate with the sensor to infer occupancy. These typically include Device-to-device ranging, Environmental recognition sensors and ID sensors. (de Moes et al., 2016) utilised the recent development in computational power of smartphones to demonstrate that occupancy information can be obtained to automate a lighting system. Basic Wi-Fi communication between smartphone and sensor provide presence and identity information. The built-in accelerometer and magnetometer also allowe localisation to an accuracy of 2-4 meters. By processing this information and controlling the Philips Hue bulbs, it was revealed that the lighting energy consumption in this scenario reduced by 67%. A recent IDC research report also demonstrated that 79% of people between the ages of 18-44 carry their smartphone for up to 22 hours of a day, suggesting that a smartphone based wearable sensor may not be unreasonable approach (Stadd, 2013). Furthermore, (Lara and Labrador, 2013) surveyed 28 systems that used wearable sensors to assess activity recognition, coincidentally revealing that similar built-in features, such as the accelerometer can be used to great effect as shown in Figure 2.3.
Although initially sounding promising, once active and passive deception enter the picture, it becomes clear instrumented sensors can be unreliable. Active deception recognises adversarial scenarios whereby the occupant would attempt to deceive the sensor; simple scenario would be an intruder entering the space in the absence of the wearable device to not be detected. This touches on the topic of security and that any instrumented based occupancy detection would need to be complemented by an additional intruder detection system. It can be argued that most home owners carry smartphones to begin with, meaning cost would be primarily centred about the complimentary detection system. This would seem rather nonsensical however, if the system can detect intruders, then it would make sense to apply that same technology to detect the homeowner. In addition, each smartphone would need to be added to the system and if the buildings begin to grow in size and number of occupants, the additional effort of having to manual program devices into the system could serve as a deterrent.
Passive deception plays on how an occupant can involuntarily deceive the sensor. If the user enters a space without the device, then there would quite simply be no detection. Despite the statistic that smartphone owners carry it for 22 hours in the day, for those 2 hours the human sensing device will be incapable of providing the smart home with useful occupancy information. Even if a person comes home with their device, under a simple scenario where the person walks into the kitchen but makes their way to another space without it, the system can quickly be undermined unintentionally. A further scenario could be if the user leaves their home without their instrument, then the system will continue to function as though the user were still present.
Such a system will struggle from a fundamental standpoint, it is only responsive to communication and not attempting to detect disturbances in the observed environment to infer occupancy. With cameras, thermal imagers and instrumented sensors seemingly cast aside, would the solution then be to just adopt the classic motion sensor?
Motion sensors attempt to observe an environment, with passive infrared sensors (PIR) being a common example, a moving person creates a change in radiation temperature from the background. These are not new to the home and have known widespread applications such as alarms, lighting systems and hand dryers. In contrast to instrumented based sensing devices, a motion sensor applies a basic physics approach of measuring a disturbance a person makes in the space. These PIR sensors have the advantage of being non-intrusive and cheap (
~£5), but they rely on motion, and hence stationary people cannot be detected (Raykov, et al., sd).
Despite the limitations of PIR, there have been attempts to optimise the algorithm to undermine these limitations. (Raykov, et al., sd) endeavoured to stretch the boundaries of what was achievable with a single PIR sensor. The algorithm recognised motion patterns and consequently use this to estimate occupancy. The system was successful to a certain extent in estimating occupancy to an accuracy of
±1 in rooms of up to 14 occupants over timeframes of 30 seconds. However, it was unable to overcome the inherent limitation of stationary occupants and with increased number of occupants, occlusion would lead to certain occupants not being detected as the human body emits infrared shielding other occupants.
To put the implications into context, one could ask how does the system respond for the overnight period where the occupant is sleeping? The limitation of a motion sensor extends itself to binary, pressure, vibration and inertial sensors which all depend on a detected movement in the environment. Are there other sensor modalities that could observe a disturbance in the environment only a human could create in both a static and moving state?
(Yavari, Lubecke and Lubecke, 2013) use an original approach by using a Doppler radar occupancy sensor to extract respiratory and heart signals, ultimately enabling presence detection. The sensor works to detect electromagnetic signals rather than infrared or ultrasonic signals, allowing the limitations of typical motion sensors to be avoided. The radar was capable of accurately monitoring respiratory and heart signals providing accurate information to infer occupancy and more. Unfortunately, the system relies on sound waves at a frequency range of 0.03-30Hz, meaning that in an urban environment for instance, background noise can infiltrate the home at similar frequency ranges. As they both exist in the same medium (internal air), interference occurs meaning the detected wave will not have amplitudes and periods accurately representative of the occupant. Admittedly, homes meeting the Passivhaus standard with high acoustic insulation could be a potential target to investigate, however only 30,000 Passivhaus standard buildings exist worldwide, creating a limited target from the outset (Passivhaus, 2017). Despite the shortcomings of this modality, the concept of detecting a human trait that is observable whilst the occupant is static could lead to a solution. Detecting signs of human breathing is clever as it works for static occupants, what if the sensor observed the exhaled CO2 rather than the sound wave?
The use of a chemosensor to detect CO2 exhibits an element of biomimicry. Like a hidden trail behind any human, Mosquitoes’ use this plume of CO2 to sense a human at distances of over 50 meters. Once they hone in on their target, they complement the tracking with their thermal sight at smaller distances of 5-10 meters to then use thermal and moisture sensors at 20 centimetres (The California Institute of Technology, 2015). The first identifiable benefit of this sensor modality is they are cheaper than cameras (K30 CO2 Sensor
~£50) and non-intrusive. In (Jiang, Masood and Soh, 2016), human emission of CO2 is used to infer occupancy by analysis of concentration levels. A statistical approach is applied where the CO2 concentration level are discretised over time intervals of 5 minutes. The discretisation of the time index is necessary due to latency between a person entering a space and sufficient accumulation of CO2 for a measurable variation to infer occupancy. The latency emphasises poor sensitivity to rapid changes in occupancy, meaning no real-time occupancy information can be provided. This implies that occupants may vacate and the system would continue to behave as though it were occupied. Jiang, Masood and Soh also consider CO2 concentration to have a Markov property in the statistical process, that is the probability of occupancy in ten minutes is influenced by whether the space is presently occupied.
When solely analysing CO2 concentration, accuracy is less than 75% for a maximum of four occupants. When temperature is included as an additional parameter however, accuracy is increased to 88.8% relative to the ground truth. This accuracy is exclusive to the model of the space however, in this case an office of plan area 9x20m. As the space size and additional parameters change in new model spaces, these figures will no longer apply. In addition to occupancy detection being a statistical process, an increase in number of occupants was found to not linearly increase CO2 concentration making count complex to infer (Gruber, Truschel and Dalenback, 2014). (Ekwevugbe et al., 2013) and (Lam et al., 2009) are further attempts to estimate number of indoor occupants from environmental parameters. However, most research has only been able to deal with a small number of occupants and are only reliable when an accurate but highly complex model of the environment is constructed.
126.96.36.199 Complex Non-Linear Systems and Networks
Due to the probabilistic approach, there remains an uncertainty to occupancy estimation and to increase accuracy, the detail of the model of the space for CO2 must increase leading to greater complexity. To visualise the hindrance created by complexity in modelling the environment, it makes sense to enter the concept of a network as can be seen in Figure 2.4.
It illustrates how CO2 concentration at any one time is linked to multiple parameters and the computational constituent to the detector needs to be able to account for these parameters. The greater the number of nodes, the more accurate the representation of the space but consequently at a cost of higher complexity. The CO2 sensor provides information for the CO2 node and if information regarding all nodes is known, then the occupant node can be singled out. Evidently lower complexity implies fewer nodes and hence singling out occupancy becomes easier. However, the network in Figure 2.4 is in fact a simplified representation and could include additional parameters such as a fireplace, pets or even degrees of freedom of the occupant node by considering size, activity and count.
In addition to the complexity, as shown in Table 2.2, the change in emitted CO2 based on activity does not vary linearly (non-linear).
Table 2.2 Variation in CO2 Emission per person based on activity (Engineering Toolbox, sd).
|Resting or low activity work||0.02|
|Normal work||0.08 – 0.13|
|Hard work||0.33 – 0.38|
188.8.131.52 Adaptability of Complex Systems
How would this complex system adapt to millions of homes? Size of space, location, daylight hours for instance would need to be reconfigured for each individual application. Some of the nodes could reconfigure themselves, but other variations such as size of space require manual input and certain nodes such as plants would require additional sensors to provide a feed of information. Is there a need for such a complex system however?
In view of 88.8% accuracy when used in conjunction with temperature, then potentially complexity can be overlooked. However, as previously mentioned this is exclusive to the test space conducted in the experiment. Furthermore, this modality is reliant on poorly ventilated spaces and the detrimental effect ventilation plays in the bandwidth over which the concentration may fluctuate. In fact, part of the benefit of a home automation is to ensure a healthy environment by using occupancy information to maintain steady CO2 levels, as studies have demonstrated the detrimental effect of increased CO2 levels on our cognitive abilities (The Center for Health and the Global Environment, 2015). Out of this arises a conflict of interest, the disturbance of the space which is being used to infer occupancy is equally what is being restricted to ensure occupancy comfort. This ultimately renders the modality unreliable in providing occupancy information. What if instead of attempting to monitor a physical attribute of the space that is equally influenced by a multitude of other parameters, a sensor observing a physical value of a space with fewer influencing parameters could yield occupancy information with greater certainty?
The technology to date has been limited to short spans, with the longest span being seen in the automotive industry where it has been explored for seat detection to enhance airbag deployment (George et al., 2009). However, (Lindahl et al., 2016) are challenging the bounds of capacitive sensing by using an operational-amplifier to generate a higher frequency to create an electromagnetic field at 50 Hertz (Hz). It was successful in achieving a range of 3.5m and outlined its intention of creating such a device as a platform for research and development. (Cooley et al., 2011) and (Avestruz et al., 2012) use the inherent electrical systems in a home to apply the concept of an electric field to detect occupancy. They adjust LED lighting ballasts with their high-frequency electrical interchanges to generate an electric field, achieving detection ranges of 3m.
In contrast to the chemosensor, there is a reduction in complexity as fewer parameters can influence the capacitance of the sensor. Furthermore, unlike the camera based sensor modality, capacitive sensing remains non-intrusive as there is no image of the space, information is limited to measurements of charge time and capacitance suggesting any panopticon effect is avoided. However, how does all this work and can it feasibly provide reliable occupant information? To answer this question, it is necessary to qualitatively explore the physics involved.
184.108.40.206 Principles of Capacitive Sensing
When a current is passed through a conductive material, the current generates a charge in the material that produces an electromagnetic field within its proximity. This material becomes known as the capacitor and the time taken for the material to hold a capacitance is a function of its size and conductivity (All About Circuits, 2016). This charge time is also known as the resistor-capacitor (RC) time constant because electromagnetic proximity sensing is formed by an RC circuit. It follows that when the space is unoccupied, the capacitor couples with the empty space denoting a baseline charge time, but when a conductive object enters the observed space, a new RC time constant is generated as illustrated in Figure 2.5.
220.127.116.11 Functionality and Reliability of Occupant Information
Given (Lindahl et al., 2016) provided evidence of a capacity to reach distances of 3.5m, what does the application of the physics of capacitive sensing indicate about the functionality and reliability of information. To begin with, the time index is significantly shorter than a CO2 sensor, in this case an order of magnitude of seconds.
Equation (1) defines the time constant in relation to resistance and capacitance, where the exact value in seconds is unique to the size of the plate, voltage and current as these will invariably affect the time it takes a certain voltage to charge and discharge the plate. The charging and discharging form a necessary cyclic process as a capacitor can only hold its charge for an instance before it begins ionising the air around it, losing the build-up of electrons and hence charge (All About Circuits, 2016).
Similarly to the CO2 sensor, various parameters are influencing the value of charge time creating uncertainty. As illustrated in Figure 2.6, the network necessary to represent the parameters is less complex than the CO2 sensor. In addition, the sensors primary objective is to monitor changes from its unoccupied state, therefore the influence of furniture or certain parasitic background capacitance will be accounted for through the unoccupied (baseline) charge time. If the occupant were to re-arrange furniture in the space however, would how would the sensor conclude that there is a new baseline charge time instead of an occupant? Research performed by (Chabalko, et al., 2017) provides evidence that using a low frequency (50Hz is considered low frequency) range means the electric field will not interact with common every day electronics such as phones or lamps and office furniture will not strongly couple. In addition, due to a human’s unique size and more importantly dielectric properties, it is stipulated that this significantly larger capacity to absorb charge in the electric field heightens the detectors sensitivity to people over objects.
The node labelled parasitic background capacitance refers to the fact that electronics in a home can generate their own electromagnetic fields that could distort the charge time. For instance, the high concentration of electrons in the electrical wires circulating through a house will quickly begin to ionise surrounding air creating an electromagnetic field. These current carrying wires are insulated to mitigate against this, but there will inevitably be residue pollution in the form of low frequency electromagnetic fields. On a theoretical level, this should not influence the working principle, but (Lindahl et al., 2016) used a ground plate to couple with the background parasitic electromagnetic fields to measure and filter the data.
Based on the evidence proposed in relation to capacitive sensing, it can be concluded that there is no further need to pursue the survey of current sensor modalities. Before pursuing a targeted investigation into this sensor modality however, a summary will help recapitulate what has been explored to subsequently underline what key points can be drawn from the survey.
Table 2.3 Summary of Sensor Modalities and Responses to Barriers
|Camera||Reliability (functionality)||High quality occupancy information|
|Thermal Imager||Reliability (functionality)||Dependant on level of pixilation|
|Instrumented Sensors||Reliability (functionality)||Passive and active deception|
|Integration||Additional security system needed|
|Cost||Dependant on type of instrumented sensor and secondary system|
|Motion Sensors||Reliability (functionality)||Incapable of detecting static occupants|
|Integration||Easy to install|
|Doppler Shift Radar||Reliability (functionality)||Interference from background noise|
|Integration||Off-the shelf installation comparable to PIR motion sensor|
|Chemosensors – CO2||Reliability (functionality)||No real-time occupancy information, not applicable to ventilated spaces|
|Cost||Acceptable cost bracket|
|Capacitive Sensing||Reliability (functionality)||Real-time occupancy information, can detect static occupants|
|Cost||Not yet known|
The desk-based study has aided in clarifying the key limitations to specific sensor modalities and how they relate to the identified barriers of functionality, integration and cost for the subsequent adoption of home automation systems in the future. As can be seen in Table 2.3, the key challenges faced by a camera is intrusiveness, motion sensor is its inability to detect static occupants and CO2 to provide real-time occupancy information amongst others. The evidence from research equally indicates that capacitive sensing is capable of overcoming the shortcomings of other sensor modalities. However, due to the fact the sensor modality is relatively new in the domain of home automation, some of its limitations as well as benefits have not yet been identified as further development needs to take place to reveal them.
(Lindahl et al., 2016) have proposed a novel and innovative system, however there is an assumption that the space remains static and any change can solely stem from occupants. However, what if a conductive object is placed into the observed space, how does the sensor accurately deduce that it is an object and not a human? There lies a need to be able to not only detect change but equally determine size of the object. Since the observed charge time of an object is a function of its distance from the sensor, localising the object would in theory allow distance and charge time to be equated to a known value of a human to compare to accurately infer occupancy. In order to devise a method for localising an object, a relationship between charge time and distance must be empirically induced. From this relationship, it is anticipated that a localisation capacity can be developed. This examination into further functional capabilities of the capacitive sensor will be followed by an investigation to ultimately answer the question of whether there are feasible methods of integration into the household at an acceptable cost bracket?
- One Directional Charge Time Analysis
This analysis aims to demonstrate and confirm real-time static detection and allow the relationship of the time constant to distance to be observed by testing a conductive object at incremental distances. Charge time is expected to follow a power law relative to distance, implying that the relationship will take the form;
- Verification of Relationship
Run a test to verify relationship has not been misinterpreted and is holding true in one direction by predicting distance relative to a measured distance. The predicted distance will follow the relationship;
- Localisation Algorithm
Step three will aim to develop a localization algorithm based on the observed relationship from the testing in step one by considering three-dimensional (3D) localisation.
- Integration into the Household and Cost
This last component to the investigation will explore the feasibility of the capacitive sensing in relation to the latter two barriers outlined by (Zenezini, et al., 2016) to explore commercial viability.
From Coulomb’s law, it is known that the strength of an electric field, whereby the strength is a measure of the force of attraction acting between two particles is directly proportional to the scalar product of the magnitude of charge and inversely proportional to the square of the distance between them (All About Circuits, 2016).
This relationship follows what is known as a power law relative to distance, whereby as distance increases, the electric force value is inversely proportional to the distance squared. Since the strength of the electric field is proportional to capacitance and the time constant, then this relationship holds true for the RC time constant. Therefore, it is predicted that the relationship can be defined by Equation (2);
For the testing of change of the charge time with distance, the environment has been simplified to clarify the identifiable behaviour and relationships. For this reason, the environment is being set up to ensure there is no background pollution from other elements, such as the cables, that could hold a charge and hence influence charge time of the plates. The plate will work over a short range (
~30cm) and at present it will initially consider one conductive object.
- Microcontroller to communicate intelligible tasks to the plate and feed the sensed data back to the computer – Elegoo Uno R3 (identical functionality to Arduino Uno)
- Coaxial cable – behaves as a shielded wire
- Alligator clip – to connect coaxial cable to plate
- 220kΩ Resistor
- 10kΩ Resistor
- Aluminium foil – material to act as sensor plate
- Computer (Macbook Pro) – Running the Arduino Ide interface
Figure 3.1 details the schematic of the hardware set-up for the capacitive plate, where the purpose to the connections is explained in the next section. It must be noted that although not visible in the schematic, the microcontroller is connected to the computers Universal Serial Bus (USB). This connection has a threefold purpose; communication, power and grounding.
18.104.22.168 Setting up the Electronics and Sensor
- Connection to the Sensor Plates
The secondary wire that contours the inner wire in the coaxial cable carries a charge and hence acts as a shield. As a result, the main wire is prevented from acting as an antenna which could influence the plate readings. At the connection to the plate, this secondary wire has been trimmed back and covered with gaffer tape to insulate at the connection to the crocodile clip. These crocodile clips were then clipped on to the plate for the charging/ discharging process to generate the capacitance. This sequence is visible in Figure 3.2 where the plate in this case was 30x30cm.
- Connection to the Microcontroller
The essential of this end was to connect the main wire to pin 8 (pin connected to sensor plate) to measure the charge time. As shown in the schematic in Figure 3.1, the main wire was soldered to a 10kΩ resistor before being connected to pin 8 on the microcontroller. To ensure the secondary wire serves as a shield, it was twisted to the side at a distance from the exposed main wire to avoid interference and then soldered to a jumper wire connecting to the 5V pin. The current drawn from the 5V pin, as previously mentioned, originates from the USB connection of the microcontroller to the computer.
22.214.171.124 Logic to the Arduino Script
The computational component began with a loop that initially set pin 8 to output mode and then proceeded to write a digital low. The digital low function triggered a grounding process; the pin registers that it is an output and the digital low command removes any residual charge that would be emitted from the pin. Consequently, any charge at the pin or in any connected material to the pin was grounded through the computer and discharges the sensor plate. For this reason, it was necessary to ensure the computer being used was plugged in to a socket with a grounding point. With the plate discharged, the pin was subsequently set to input mode and a timer function was called to measure the charge time it takes for the pin mode to go to ‘high’. The charging of the plate came from the 5V pin connection to the plate. Once the pin mode went to ‘high’, the timer was prompted to stop meaning a raw charge time was obtained. This process was then repeated continuously to measure the charge time for all the distances.
126.96.36.199 Testing Methodology
To obtain data, a hand was used as the object to test, starting at 3cm and increasing in increments of 3cm to a final distance of 30cm.
For each increment of 3cm, three charge times were recorded and averaged. The raw charge times were printed as coma separated variables (CSV) to be plotted in excel.
Following the testing, the obtained results were plotted validating a power law relative, whereby from Figure 3.4, it can be seen that charge time is inversely proportional.
By manipulating the results, the hypothesis of Equation (2) is supported by Figure 3.5, where plotting Equation (3)
d= 1τhas yielded a linearly proportional relationship.
The demonstrated results visually support the hypothesis Equation (2) and the superposition of the baseline charge time helps underline that any change in the environment will be noticeable regardless of whether the detected object is in motion or at rest. This confirms and demonstrates the proof of concept that capacitive sensing can statically detect occupants non-intrusively. In addition, the baseline charge time was found to be 21
μs (0.000021 of a second) equally confirming real-time information. Due to this minute latency, the code calls a delay function of 100 milliseconds to ‘slow’ the system down, as working faster than a tenth of a second is not necessary.
It is worth noting however, these results suffer from certain technical limitations in the methodology. Aluminium has been used as a material but a material of higher conductivity such as copper would yield more sensitive results. In addition, manipulating the value of the resistors will impact the charge time, smaller resistance results in a faster charge time. However, a baseline resistance is needed otherwise current could be drawn too quickly from the computer which could result in lack of reliability of power. This could be mitigated against by directly connecting the Arduino to the wall, providing more reliable power source and a direct grounding method. Disconnecting the microcontroller from the computer would be at the expense of the direct communication method via the serial ports, however a Wi-Fi communication alternative will not impact latency and reliability of data.
Finally, as shown in Figure 3.3, the measured distances suffer from accuracy in both the measuring method from the ruler and due to the fact that the hand does not remain perfectly stable at the same distance for each measured charge time. As a result, the measurements will invariably be to a certain extent inaccurate in the order of magnitude of cm’s. Despite the limitations in the methodology, the demonstration has exhibited with sufficient clarity the relationship between charge time and distance of object. Furthermore, the error in accuracy of the charge time relative to the distance is not of concern in consideration of the application. The difference in accuracy of mm’s to cm’s will not inhibit presence detection and the localisation capacity will not be severely limited.
It would be informative to understand the accuracy and verify there is no misinterpretation by comparing the predicted distance using the assumed relationship to the measured distance (ground truth).
It is anticipated that the estimation of the distance of the object will be achievable to a certain degree of accuracy with an error in an order of magnitude of cm’s. As can be seen in Figure 3.3 and discussed in section 3.2.5, the predicted distance and measured distance will inherently have errors of accuracy due to limitations in methodology; simple changes in hand height or precision of measurement of hand height from the ruler for instance. Therefore, based on limitations in methodology, any margin of error >10cm will be indicative of a misinterpretation of the relationship rather than a limitation of testing methodology.
For the sensor to be able to compute a distance, a calibration process was performed. Without the calibration process, the sensor could only see a charge time but the value was meaningless without a reference point. For this reason, an initial measurement at a distance of 3cm for the test object provided a reference charge time, subsequently the system could independently compute the distance of the object based on this ‘training data’.
188.8.131.52 Methodology of testing
Following in the same methodology of testing as outlined in section 184.108.40.206 for the measuring of distance of the object, the predicted distance will stem from the empirically induced relationship using Equation (3).
Table 3.1 displays the predicted distance based on Equation (3) and tabulates the error between measured and predicted.
Table 3.1 Comparison of Predicted vs. Measured Distance
|DistanceofObject (cm)||PredictedDistanceofObject (cm)||Error (cm)|
As anticipated, the comparison of the results of the predicted distance of the object relative to the measured distance differ by an error in the order of magnitude of cm’s, therefore no misinterpretation has occurred. Ultimately, adjustments to refine testing may potentially lead to more accurate results but the same identified relationship will still be evident.
It is important to remain critical, consequently the implications of the calibration must be discussed. The need to ‘train’ the sensor by feeding it data highlights that this method will be unsuccessful in inferring size based on equating distance and charge time. This calibration informs the sensor that at a given distance, the charge time for the object to compute distance for is a respective value. However, any subsequent measured charge time will use the same sized object as a reference point to infer distance. If a smaller object such as a bag yields a smaller charge time, based on the training data the sensor will compute a greater distance and ultimately the true distance of new objects will remain unknown.
Despite this functionality related shortcoming, the observed relationships still allow for a real-time tracking system to be developed. This is worthwhile investigating as this could reveal a beneficial capacity to compartmentalise spaces for different thermal requirements and comprehensively automate the lighting system of the home, even if it is just calibrated to one known person.
To develop the established relationship between distance and the charge time of Equation (3), the addition of two plates is expected to provide an ability to localise an object in a 3D space.
Similarly to the one directional analysis, a simplified model of an environment was necessary to demonstrate a proof of concept (refer to section 3.2.2).
- Microcontroller – Elegoo Uno R3
- 3 Coaxial cables
- 3 Alligator clips
- 3 220kΩ Resistors
- 3 10kΩ Resistors
- Aluminium foil
- Macbook Pro – Running the Arduino Ide interface and Processing 3
220.127.116.11 Setting up Electronics and Sensors
- Connection to the Sensor Plates
The shield wire was prepared in the same manner as the one directional charge analysis and the main wire connected to the crocodile clip just repeated for the three connections as outlined in more detail in section 18.104.22.168.
- Connection to the Microcontroller
As can be seen in Figure 3.7, the shield wire was twisted to the side at a distance from the exposed main wire and subsequently soldered to a jumper wire connecting on the microcontroller at the 5V pin. Figure 3.7 also shows how the main wire’s resistors were soldered together before joining to the same 5V connection as the shield wire.
22.214.171.124 Displaying Capacities of Processing
Processing 3 was used to visually demonstrate the location of the object with a delay of 100 milliseconds giving the effect of a dynamic model of the location (Figure 3.9). A 3D reference field was generated where the measured charge time and subsequently calculated distance in each plate served as coordinates.
126.96.36.199 Full Assembly
Figure 3.8 shows the full assembly of the small-scale model. Note that a gap was left in-between each plate to avoid contact and ensure the system retains three directions.
Figure 3.9 visually displays in Processing the localised object in 3D using capacitive sensing. The findings will be deliberated in further depth on a holistic level to underline key limitations and benefits in section 4.
If such a sensor modality were to be adopted, how would it be fitted into a home and at what cost? From the outset, cost has high variability due to its dependency on installation and development of the technology, therefore the integration into the household will be initially considered. Two potential avenues can be explored by considering i. a conventional off-the shelf style sensor and ii. the sensor being built into the fabric of the home.
- Off-the Shelf Sensor
Cameras or PIR motion sensors are simple examples that have adopted the concept of off-the shelf sensors. It is in other words a sensor that comes in units; buy a unit of the sensor in a shop and then install it in a room. Both the number of operational rooms in a home and the price will vary linearly depending on the number of sensors bought. In this scenario, if only one sensor is used then evidently occupancy can only be detected for one room. This brings to light a key limitation; efficacy of the home automation system is not only dependant on functionality but also how dead spots limit its observable view.
The need to limit number of dead spots does not necessarily imply a high cost, individual units can be produced cheaply as is seen with the CO2 sensor and PIR motion sensors. The ideal view that a unit can just be bought off the shelf and then fitted into a space is however not necessarily true. In reality, there will be additional installation considerations but equally a demand for power as it is expected to run continuously.
In relation to capacitive sensing, the greater the electric field implies the greater the monitored area. In the form of units, these sensors could end up being large, unappealing to the eye and needing additional structural support in contrast to motion sensors, which can be hidden in a corner. The implications of this approach could dictate that people feel deterred to adopt a sensor that is cumbersome to install and not aesthetically appealing. If the sensor is kept to a minimal size, this will have the likely effect of limiting the observable view.
- Built into the Fabric of the Home
The main benefits to this approach is it gives a comprehensive detection system with few dead spots and sidesteps the risk of developing large unit sensors with poor aesthetic qualities. The sensor could be integrated using either the insulation or retrofit the walls with modular panels or conductive paint. If incorporated during construction, the sensor could use a material that would exist in the building regardless. In contrast, retrofitting the sensor into the fabric approach will struggle to be integrated as easily as an off-the shelf method. To utilise the insulation, place modular panels or paint all the walls with a conductive finishing will require additional installation costs and potentially a specialist. However, cost in this context can only remain qualitative.
It is unwise to explore cost quantitatively as the technology to occupancy detection by capacitive sensing is new with a high number of unknowns about how it will develop to a finished product. Therefore, extracting a figure for either method of integration for comparison to other sensors would be both speculative and potentially misleading.
Before discussing the implications of the findings, it is vital to remain both objective and critical of the validity of what has been observed. The need for this is best articulated through (Ioannidis, 2005), where John P.A. Ioannidis investigates why findings from published research is mostly false. The two key conclusions to be drawn from the paper are that findings are misinterpreted due to limitations in demonstrating only part of a picture and hence induced relationships are not sensitive to the larger picture. Secondly, the human element of bias leads to reporting being manipulated by distorting findings. Therefore, this section will initially review the methodology to assess whether this induced relationship is at risk of producing false findings. Subsequently, an analysis of the findings by division of benefits and limitations of capacitive sensing will be performed with attention to bias. Potential implications of certain findings will equally be discussed where applicable and lastly future research recommendations will be suggested.
(Ioannidis, 2005) argues that a key factor influencing whether a research finding is true or false is the prior probability of it being true. Following this logic, the empirically observed findings and subsequent induced relationship of charge time to distance grounds itself in fundamental theories of physics deduced by Charles Augustin de Coulomb in 1784 and then further developed by James Clerk Maxwell in the 19th century (MIT Open Courseware, 2004). The theories have since been empirically observed and confirmed repeatedly providing confidence that they accurately represent the true nature of capacitive behaviour. This conversely provides a well-established bigger picture and therefore, despite technical limitations (as outlined in section 3.2.5), it can be concluded that the relationship between charge time and distance is true. This conclusion is made with sensitivity to the induction fallacy proposed by David Hume, which argues knowledge of an event based off empirical observation does not necessarily hold true for future events (Morris and Brown, 2017). In other words, just because the induced relationship of charge time to distance has been observed countless times in the past, it cannot be said with absolute certainty that the same will hold true for a future event. Despite this scepticism, it is necessary to take a stance and ascertain that with high probability this relationship is true.
Based on the conclusion that the findings of the relationship between charge time and distance are true, then it is worthwhile examining what these have revealed. Figure 3.9 (section 3.4.4) displays not only non-intrusive real-time detection of static occupants but equally a proof of concept of tracking capabilities. Unmistakeably, the sensor will need to be designed to be resistant to cyber hacking, but in contrast to the image provided by a camera, it is not placed at a heightened risk based on the information it provides.
The preferred method of integrating capacitive sensing into buildings would be retrofitting it into the fabric of the walls. As mentioned, many factors play into evaluating the cost (surface area of walls, modular panels or conductive paint etc.) and therefore it has been concluded that is ultimately misleading rather than informative. However, it is anticipated that the technology will initially be expensive, a potential method to mitigate against this would be to adopt the Tesla business model; when the technology is new and expensive to produce, develop the technology as a high-end product. As the technology gains popularity, a wider audience is willing to adopt it and with the revenues for the high-end product, economies of scale can be applied to design cheaper to produce versions of the technology as a lower-end product, whilst not compromising the authenticity and integrity of the actual technology (Musk, 2006).
As previously discussed, the method integration could yield fewer dead spots in contrast to motion sensors for instance, which are typically present in corridors, stairways, behind furniture, large rooms. This feature coupled with the localisation capability provides a heightened efficiency in use of spaces, particularly in large spaces. For example, in libraries the ability to track an occupant means not an entire floor has to be lit if an occupant is present, conversely the person does not need to perform jumping jacks every 10 minutes once the motion sensor’s timer elapses. Finally, this also allows for effective compartmentalisation of spaces for different thermal requirements and automated lighting.
Another beneficial implication of integrating plate-like capacitive sensors into the fabric of the home is it coincidentally aligns with the wireless charging technology that has recently been developed by Disney Research. The basic premise of the research is to seamlessly charge electrical devices in a space and to achieve this, an electromagnetic field has been generated by lining the walls with a conductive material (Chabalko, et al., 2017). Should we then just paint our walls in electrically conductive paint to adopt home automation?
Answering this question highlights the need to balance the discussion with the limitations. As has been established, the adopted method of localisation is unsuccessful in inferring size. It is important to realise that the identified relationship between charge time and distance is merely a relation but that relationship is meaningless to the sensor without a reference point. As a result, the sensor at present can only localise a known object as its computation of distance is based on a known charge time. In reality however, there are other conductive objects that will enter a space aside from furniture. For instance, in a scenario such as a garage where a car, with highly conductive material, enters the garage, the sensor will detect the object but will not be able to differentiate between car or occupant. This poses a real problem, if not solved then the system in this scenario would light and heat the space wastefully.
Through the desk-based study, it has been concluded that an electric field at a low frequency range means the capacitive plate would not couple with furniture and common everyday electronics such as phones or computers. However, this is not entirely true as it assumes contents of the liveable space (furniture, sofa, bed etc.) will be made of a material with low absorption capacity. Chairs and book-cases are common examples of furniture that is found with metallic material which the sensor will be more sensitive to. This hints at potentially problematic implications; if objects such as chairs are moved closer to the wall, the charge time could deviate from its baseline value. A threshold could be considered but the sensors sensitivity to this change remains unknown.
Another important limitation is that of deducing number of occupants in the space. This dimension of functionality has not been considered in the scope of the investigation of capacitive sensing, therefore it remains a limitation of the technology at present. On a theoretical level, increased number of occupants results in an increased charge time, however it has been observed that distance influences the perceived charge time of an object. A simple visualisation of a scenario helps illustrate the challenge of determining number of occupants. Imagine three people enter a space that has a one-directional sensor similarly to the analysis in section 3.2. If all three people line up at equidistance from the sensor, how does the sensor know three people are present? Remember, the localisation system compares the measured charge time to a calibrated value to subsequently calculate distance. In this scenario, it will see a large charge time and may conclude that one person is extremely close to the sensor. Alternatively, if the charge time is too big it could conclude two people are at a medium distance. However, it remains incapable of accurately determining with high certainty that three people are present. Furthermore, if these people were at different distances and considering that charge time follows a power law relative as distance increases, how does the sensor then divide the charge times up? A potential avenue by which to explore this topic will be discussed in the future research section.
With regards to integration, retrofitting the sensor may not only prove to be costly depending on development, but equally the increase in effort in contrast to an off-the shelf sensor could consequently serve as a deterrent to potential adopters. Additionally, there is a heavy reliance on walls, but what if the architect has opted for a more transparent glass walled design? A solution could be to place the sensor into the floor, providing a means to detect a change in the space. Furthermore, an ambitious assumption in the computational element could follow the logic that since the plate is in the floor, the object would always be on the floor and hence have a fixed distance. In this case, there would be a known distance for the sensor to then compare distance and charge time to a trained value for the charge time of the occupant.
There are also indirect implications of generating an electromagnetic field in a space that could lead to questions surrounding health and safety. This topic was addressed in a Specific Absorption Rate (SAR) analysis performed by Disney Research in accordance with safety guidelines, which demonstrated that no long term health damage would occur if the electric field frequency is below 1.34 MHz (>50 Hz from (Lindahl, et al., 2016)) (Chabalko, Shahmohammadi and Sample, 2017).
Based on the evidence prospered from the research and the findings observed in the investigation, capacitive sensing is worthy of further research to address the current unknowns.
The first area of further development should focus around solving the limitation of not being able to determine the size of an object in a space. The key limitation to the current methodology revolves around the fact the relationship is purely relative and there are too many unknowns, meaning the equations will not predict an accurate value. When referring to the calibration process, training or calibrating the sensor with a measured distance and charge time for the occupant means a reference value can be computed for all locations by the sensor. If an object at a given distance and charge time is observed by the sensor, then it could compare the observed value to a calibrated or trained value. Under this basis, if the calibrated value for an occupant matched the observed value, then occupancy could be inferred. However, the measured charge time remains meaningless unless a known distance is provided, but the method applied during the investigation provided no accurate way of obtaining distance. Despite having deliberated over the issue extensively to develop a computational method of overcoming this, to the authors best knowledge the only way forwards would be to compliment the system with an additional sensor to detect distance. A basic ultrasonic distance sensor such as the HC SR04 could provide a cheap means to measure distance, however in view of developing the technology to infer number of occupants, a radar could provide a larger view to detect multiple objects distances as well as an estimate of size. This will enable a measured charge time to be divided up into relevant distances and according size to accurately infer occupancy.
Construction of a full-scale model would allow for the investigation of the sensors sensitivity to furniture. This could serve as validation of the research-based evidence that the sensors would not be influenced by furniture or its movement. Conducting tests with variations in amount of furniture, placement and material will yield results for empirical observation. It is hoped that this will shed light on a necessary threshold to implement into the system and whether this threshold is at risk of overlapping with the threshold for occupancy detection.
In the foreseeable future, the development of capacitive sensing holds a lot of promise but in the distant future it is expected that quantum sensing will pave the way for a ‘quantum’ leap in sensing technology. The basic premise behind the technology is that at a quantum level, atoms can store information on the state of atoms in another space which if better understood could allow a wealth of information to be extracted. This is based on modern physics that requires extensive development, which is nevertheless a field worthy of investigation and development.
One of the key challenges in the concept of sensing is the ability to accurately represent the environment to extract the desired information. Physics provides the grounding that through an equation that accurately describes the nature of the physical world, obtaining the necessary values to the variables of the equation allows the unknown to be resolved. The sensor has an ability to provide the value for a certain variable in the equation, however there is no explicit equation for occupancy detection. This equation needs to serve as a link between the sensed physical variable and a variable to represent an occupant. Certain sensor modalities, such as CO2, may create this link between observed physical variable and an occupant, but account for numerous additional variables in the process. The equation itself can also fluctuate between two extremes, if a purely physics based extreme is applied, then the equation verges towards accurately describing the true nature of the observed physics. In the case of capacitive sensing, this could entail additional consideration of SAR and size, information that it is unrealistic to obtain. Therefore, a balance needs to be drawn between the extreme of physics that verges towards an absolute truth and the necessary abstraction of the equation to account for the fact a sensor only provides information for one variable. This balance needs to provide a sufficient level of confidence that despite the abstraction, and hence ignored variables, it remains to an extent sufficiently accurate to estimate occupancy. Evidently, the equation proposed in the investigation following the empirical induction of a relationship has three unknowns and one equation if the reference point is the third unknown. If future research were to further investigate capacitive sensing in conjunction with a distance sensor, it is hoped that this will provide the information for the missing variable to provide the system with a relative indication of the actual size of the object to subsequently infer occupancy.
On a holistic level, the research based survey comprehensively establishes the field of occupancy detection at present to articulate capacitive sensing as a valid solution. Furthermore, the state of the technology considered will undoubtedly experience development before evolving into a finished final product to be placed in homes. The limitations and further research attempt draw attention to areas of research that will be critical to the evolutionary path that capacitive sensing will undergo before finding its place in homes. Nevertheless, the findings successfully yielded key conclusive findings to be drawn from the paper. Specifically, a means to detect static occupants non-intrusively in real-time and, with calibration, track an occupant. This indicates a potential to successfully track an occupant to allow for effective automation of lighting systems and compartmentalisation of spaces for different thermal demands for the HVAC systems.
All About Circuits, 2016. Series Resistor-Capacitor Circuits. [Online]
Available at: https://www.allaboutcircuits.com/textbook/alternating-current/chpt-4/series-resistor-capacitor-circuits/
[Accessed 30 November 2016].
Arens, E., Federspiel, C., Wang, D. & Huizenga, C., 2005. How ambient intelligence will improve habitability. Ambient Intelligence, pp. 63-80.
Armstrong, J., 2016. How do driverless cars work?. [Online]
Available at: http://www.telegraph.co.uk/cars/features/how-do-driverless-cars-work
[Accessed 2016 5 2016].
Basic Electronics Tutorials, n.d. Basic Electronics Tutorials. [Online]
Available at: http://www.electronics-tutorials.ws/rc/rc_1.html
[Accessed 21 November 2016].
BBC, 2014. Google to buy Nest Labs for $3.2bn. [Online]
Available at: http://www.bbc.co.uk/news/business-25722666
[Accessed 18 November 2016].
Beltran, A., Erickson, V. & Cerpa, A., 2013. ThermoSense: Occupancy Thermal Based Sensing for HVAC Control. Buildsys.
Chabalko, M., Shahmohammadi, M. & Sample, A., 2017. A Quasistatic Cavity Resonance for Ubiquitous Wireless Power Transfer. PLOS One.
Cooley, J. et al., 2012. Dimmable Solid State Ballast with Integral Capacitive Occupancy Sensor. IEEE Transactions on Industrial Electronics, 59(4).
de Moes, A., Pedersen, J., Sarkar, C. & Prasad, V., 2016. Fine-tuned lighting control leveraging smartphone-based occupancy detection. EWSN.
Ekwevugbe, T., Brown, N., Pakka, V. & Fan, D., 2013. Real-time building occupancy sensing using neural-network based sensor. IEEE International Conference on Digital Ecosystems and Technologies, pp. 114-119.
Engineering Toolbox, n.d. Carbon Dioxide Emission from Peopl. [Online]
Available at: http://www.engineeringtoolbox.com/co2-persons-d_691.
[Accessed 10 March 2017].
European Commission, 2016. Climate Action – European Commission. [Online]
Available at: https://ec.europa.eu/clima/policies/strategies/2020_en
[Accessed 20 December 2016].
Foucault, M., 1975. Discipline and Punish – The Birth of Prison. New York: Penguin Group.
George, B., Zangl, H., Bretterklieber, T. & Brasseur, G., 2009. Seat Occupancy Detection Based on Capacitive Sensing. IEEE Transactions on Instrumentation and Measurement, 58(5), pp. 1487-1495.
GFK, 2016. UK Smart Home Statistics in the UK from GFK. [Online]
Available at: http://www.dataselect.co.uk/uk-smart-home-statistics/
[Accessed 26 February 2017].
Gruber, M., Truschel, A. & Dalenback, J., 2014. CO2 Sensors for Occupancy Estimations: Potential in building automation applications. Energy and Buildings, Volume 84, pp. 548-556.
How driverless cars see the road. 2015. [Film] Directed by Ted Talk. s.l.: s.n.
International Energy Agency, 2016. Key World Energy Statistics. International Energy Agency.
Ioannidis, J., 2005. Why Most Published Research Findings Are False. PLoS Medecine, 2(8), p. 124.
Jacobsson, A., Boldt, M. & Carlsson, B., 2016. A risk analysis of a smart home automation system. Future Generation Computer Systems, Volume 56, pp. 719-733.
Jiang, C., Masood, M. & Soh, Y., 2016. Indoor Occupancy estimation from carbon dioxide concentration. Energy and Buildings, Volume 131, pp. 132-141.
Lam, K. et al., 2009. Occupancy detection through an extensive environmental sensor network in an open-plan office. IBPSA Build. Simul., Volume 145, pp. 1452-1459.
Lara, O. & Labrador, M., 2013. A Survey on Human Activity Recognition using Wearable Sensors. IEEE, 15(3).
Lindahl, P., Avestruz, A., Thompson, W. & George, E., 2016. A Transmitter-Receiver System for Long-Range Capacitive Sensing Applications. IEEE Transactions on Instrumentation and Measurement, 65(10).
McMullan, T., 2015. What does the panopticon mean in the age of digital surveillance?. [Online]
Available at: https://www.theguardian.com/technology/2015/jul/23/panopticon-digital-surveillance-jeremy-bentham
[Accessed 18 November 2016].
Medvedev, I., Schueler, R. & Thomas, J., 2016. Analysis of Exhaled Human Breath via Terahertz Molecular Spectroscopy. IEEE.
MIT Open Courseware, 2004. Electromagnetic Theory, s.l.: MIT.
Morris, W. & Brown, C., 2017. David Hume. [Online]
Available at: https://plato.stanford.edu/cgi-bin/encyclopedia/archinfo.cgi?entry=hume
[Accessed 4 April 2017].
Musk, E., 2006. The Secret Tesla Motors Master Plan (just between you and me). [Online]
Available at: https://www.tesla.com/en_GB/blog/secret-tesla-motors-master-plan-just-between-you-and-me
[Accessed 4 April 2017].
Nest, 2016. Meet the Nest Cam Indoor security Camera. [Online]
Available at: https://nest.com/uk/camera/meet-nest-cam
[Accessed 23 December 2016].
Nguyen, T. & Aiello, M., 2012. Energy Intelligent Buildings based on user activity: A Survey. Energy and Buildings.
Pagliery, J., 2016. Nearly one million Android phones infected by hackers. [Online]
Available at: http://money.cnn.com/2016/11/30/technology/android-phones-infected/
[Accessed 12 February 2017].
Passivhaus, 2017. Passivhaus: The Passivhaus Standard. [Online]
Available at: http://www.passivhaus.org.uk/standard.jsp?id=
[Accessed 11 March 2017].
Pérez-Lombar, L., Ortiz, J. & Pout, C., 2008. A review on buildings energy consumption information. Energy and Buildings, 40(3), pp. 394-398.
Pultarova, T., 2016. Webcam hack shows vulnerability of connected devices. [Online]
Available at: https://eandt.theiet.org/content/articles/2016/11/webcam-hack-shows-vulnerability-of-connected-devices/
[Accessed 22 February 2017].
Raykov, Y. P. et al., n.d. Predicting room occupancy witha single passive infrared sensor through behavior extraction.
Renault, J., 2016. Interview with Founder of Delta Dore [Interview] (November 2016).
Rouse, R. A., 2012. Is Someone Watching You Through Your Webcam. Campatch Academy.
Seth Godin: How to get your ideas to spread. 2003. [Film] Directed by TED Talk. s.l.: s.n.
Stadd, A., 2013. 79% of People 18-44 Have Their Smartphones With Them 22 Hours A Day. [Online]
Available at: http://www.adweek.com/digital/smartphones/
[Accessed 2016 November 28].
Stanford Web, 2003. Diffusion of Innovations by Everett Rogers (1995). [Online]
Available at: https://web.stanford.edu/class/symbsys205/Diffusion%20of%20Innovations.htm
[Accessed 5 March 2017].
Statista, 2017. Smartphone sales worldwide 2007-2015. [Online]
Available at: https://www.statista.com/statistics/263437/global-smartphone-sales-to-end-users-since-2007/
[Accessed 21 February 2017].
Surden, H., 2016. How Self-Driving Cars Work. SSRN.
Teixeira, T., Dublon, G. & Savvides, A., 2010. A Survey of Human-Sensing: Methods for Detecting Presence, Count, Location, Track and Identity. ENALAB Technical Report, Volume 1.
Tesla, N., 2011. My Inventions and Other Writings. New York: Penguin Group.
The California Institute of Technology, 2015. Mosquitoes Use Smell to See Their Hosts. [Online]
Available at: https://www.caltech.edu/news/mosquitoes-use-smell-see-their-hosts-47338
[Accessed 19 October 2016].
The Center for Health and the Global Environment, 2015. The Impact of Green Buildings on Cognitive Functions. [Online]
Available at: http://www.chgeharvard.org/resource/impact-green-buildings-cognitive-
[Accessed 14 October 2016].
The Economist, 2014. A special report on cyber-security: The internet of thigns. [Online]
Available at: http://www.economist.com/news/special-report/21606420-perils-connected-devices-home-hacked-home
[Accessed 25 February 2017].
Tyndall, A., Cardell-Oliver, R. & Keating, A., 2016. Occupancy Estimation Using a Low-Pixel Count Thermal Imager. IEEE Sensors Journal, Volume 16, pp. 3784-3791.
Yavari, E., Lubecke, V. & Lubecke, O., 2013. Doppler Radar Sensor for Occupancy Monitoring. IEEE.
Zenezini, G., Ghajargar, M., Fjore, E. & De Marco, A., 2016. The Smart Home Services Diffusion Process: A System Dynamics Model. Politecnico di Torino.
Zipperer, A., Roche, R., Earle, L. & Bauleo, P., 2013. Electric Energy Management in the Smart Home: Perspectives on Enabling Technologies and Consumer Behavior. National Renewable Energy Laboratory, p. 12.