Smart Carpet using differential piezoresistive pressure sensors for elderly fall detection Kabalan Chaccour, Student Member, IEEE, Rony Darazi, Member, IEEE, Amir Hajjam el Hassani, Emmanuel Andres Ticket Lab, Antonine University, 40016 Hadat-Baabda, Lebanon IRTES-SeT, Université de Technologie Belfort-Montbéliard, 90400 Sevenans, Belfort, France Université de Strasbourg, Centre Hospitalier Universitaire, 67000 Strasbourg, France Email: kabalan.chaccour@upa.edu.lb, rony.darazi@upa.edu.lb, amir.hajjam-el-hassani@utbm.fr, emmanuel.andres@chru-strasbourg.fr Abstract Falls are events that affect almost every aging human being above the age of 65. These incidents can have major consequences on the physiological, psychological and socioeconomical levels. In this paper, a simple smart carpet design is developed to detect falls using a novel sensing technique. Conventional sensing methods use either inertial measurement sensors (accelerometers, gyroscopes) or environmental sensors (infrared, force, vibration, acoustic, etc.). The proposed technique employs differential piezoresistive pressure sensors. The prototype of the system is implemented and tested using statistical methods. Experimental results show the sensitivity and the specificity of the system to be 88.8% and 94.9% respectively. The system could be deployed in home care environment as a final product. Our proposed sensing technique can be also integrated in beds to alert patients from falling during sleep. Keywords Aging; fall detection; smart home; intelligent floor; differential pressure sensors. I. INTRODUCTION Aging is a heterogeneous natural process that can affect all kind of living species particularly humans. This process is characterized by a decrease in functional abilities of the person. One of the major consequences of aging in humans is fragility. As a result, the aging person is subject to many incidents particularly falls. Falls are common incidents that accompany aging. They are very critical if they happen and in some cases they are fatal. Statistics show that 1 out of 3 persons over the age of 65 fall each year [1]. In addition, these events have many impacts on different levels. As a matter of fact, falls do not only affect the person himself by increasing his dependencies, but also impacts his environment where the cost of hospitalization and elderly care management become more significant. In this context, the need for fall detection became important. Many devices and systems were developed to detect falls. These devices use different sensor technologies. These are divided in two categories: wearable sensors systems and non-wearable sensor systems. The formers commonly use inertial sensors such as accelerometers, gyroscopes and Inertial Measurement Units (IMU), while the latter exploit environmental sensors such as pressure, motion, vibration, acoustic, and infrared sensors, etc. There are also other environmental sensors that use vision techniques through camera or radar deployment. The objective of this article is to develop a smart carpet to detect falls of elderly persons in a home care environment. The smart carpet uses embedded differential piezoresistive pressure sensors that to our knowledge have not been yet deployed in such applications. Moreover, a threshold based algorithm is also designed to detect falls using the technique. The rest of the paper is structured as follows: In section II, we will give an overview of the common sensing techniques previously deployed in such applications. Our focus resides on existing solutions that use environmental sensors since our system belongs only to this category. Section III will provide a detailed description of the developed system. Experimental results and discussions on the proposed technique are provided in section IV and V respectively. We finally conclude the research in section VI. II. RELATED WORK The fall detection problem has been widely studied in the literature. Many systems have been developed using various sensor technologies. These systems may use wearable and non-wearable sensors to detect falls. Non-wearable sensing techniques are the focus of our research. C. Franco et al. in [2], proposed a multisensory Telehealth- Care system using PIR infrared motion sensors that monitor the daily activities of the elderly as well as his environment (temperature, illumination, etc.). The system also uses wearable sensors to monitor the nycthemeral rhythms of the patient that may indicate a possible nightfall only. The proposed system architecture does not take into consideration falls during normal ADL (Activities of Daily Living). Besides, PIR (Passive Infrared Sensor) motion sensors alone are not used for fall detection. Controversially in [3], Yazar et al., use the same PIR sensors with floor vibration sensors for fall detection. The system uses a novel signal processing algorithm to predict falls. The approach is simple, reliable and has a low cost but it requires complex installation and cabling to be implemented in large scale environments. Recognizing the person s activities and identity was proposed by Cheng et al. in [4]. They demonstrated using a pressure sensor matrix that posture affects weight distribution on the ground. The sensing technique uses pressure sensors made of Electrostatic Discharge (ESD) protection foam mixed with carbon fiber. The density of carbon particles grows when pressure is exerted reducing resistance. Resistance is therefore measured by inserting 2 electrodes into the foam. The system has a spatial resolution of 3cm 2 and uses an array of 32x32 978-1-4673-7701-0/15/$31.00 2015 IEEE 233
sensors. The system identifies persons by performing static activities on the carpet. Falls are not studied in this research, however gait analysis may be considered if using higher spatial resolution. Miguel et al. in [5], proposed a low power resistive pressure sensor array for smart floors made of conductive foam mixed with carbon particles. The concept is very similar the one proposed by Cheng et al.; however the latter investigates more the geometric shape and the material of the sensor. The research also focuses on the power consumption of the conditioning electronics used. The system is simulated on a tile with 8x8 sensors. Fall profile can easily be predicted using this system in order to categorize the severity of the fall. Indoor people localization and tracking is also studied by Contigiani et al. in [6]. The research uses capacitive sensors installed inside floor woods. Specific shoes were designed to work with the floor. This condition narrows the application of this system in real life and restricts the freedom of the elderly to wear his favorite shoes. More research is done with Al-Naimi et al. [7] with their advanced approach for indoor identification and tracking. Their smart environment use Force Sensitive Resistor sensor with a customized PIR motion sensor. The system has the ability to track and identify footsteps. Falls are also not considered in this context however; gait parameters may be extracted and analyzed. Anh ho et al. in [8] developed a promising multimodal sensing technique. The sensor itself is made of double yarn layers made of a mixture of polyester and stainless steel fibers. The layers are separated by a non-conductive basal yarn layer making it a resistive pressure sensor. The sensor has three modalities: proximity detection and light touch, tactile perception and tensile. Each modality may be used for a specific application. Despite its simplicity, the author didn t discuss the deployment in a large scale environment such the interfacing of the sensor with conditioning and processing electronics. In addition, the resistance of the sensor to heavy falls is not studied. In addition, to the above mentioned sensing techniques, successful products are currently being deployed. The system denoted SensFloor, by Steinhage et al. in [9], uses also capacitive pressure in rather moderate spatial resolution. Tekscan also develops products with resistive pressure sensors for clinical and research purposes. Plastic Optical Fibers (POF) were also used by Cantoral-ceballos et al [10]. They have measured the deformation magnitude induced by exerting pressure on the substrate of the fiber. This method can monitor the subject balance as well as his gait. In this paper, we are proposing a novel sensing technique that has not yet been exploited in existing fall detection systems. The technique is very promising to be manufactured as a final product. III. A. General system description SMART CARPET SYSTEM Our approach to the problem of fall detection consists of using differential piezoresistive pressure sensors that to our knowledge have not yet been deployed in such applications. The system is divided into 2 parts: The smart carpet where pressure is exerted and picked up using sensors and the signal conditioning and processing electronics where signals are acquired and processed. The objective of the system is Fig. 1. Smart Carpet system architecture. to detect fall events. The electronics utilizes a threshold-based algorithm to trigger an SMS alarm notification in case of a fall event. Figure 1 shows the general architecture of the system. B. Differential pressure sensor carpet The smart carpet uses differential piezoresistive pressure sensors generally found in industrial applications. It appears that the commonly known and effective way to quantify pressure is to measure its difference with respect to a common reference. Our sensor model uses the atmospheric pressure as a common reference to quantify the exerted pressure. In this case, the structure of the sensor is made of 2 chambers CH1 and CH2 as shown in Figure 2a. CH1 is the reference chamber where the atmospheric pressure Pr is applied, whereas CH2 is the measurement chamber where an analogue pressure Px is exerted. Sensing elements are fixed on the membrane separating both chambers. The sensor membrane has a stress strain gauge design type which allows a high precision differential pressure measurement. Therefore, the output signal of the sensor is the differential pressure defined by: Px - Pr. The Freescale Semiconductor MPXV53GC7U differential pressure sensor is used in our system design (Figure 2b). In addition to its low cost, the sensor provides a very precise linear voltage output. The output voltage is proportional to the applied pressure. The signal is measured on the output pins V out+ and V out (Figure 2c). The top nozzle of the sensor as shown in Figure 2b above is connected to a manual air pump via an air tube (Figure 3a). By applying force on the air pump, a pressure is exerted on the sensor and the differential pressure with respect to the atmospheric is transduced. The maintenance of this sensor is easy since it is located outside the carpet, whereas other types of sensors (force resistive or capacitive) are implanted inside and difficult to maintain in case of failure. The carpet itself is made of a thin rectangular wooden plate. The plate is covered by a sponge and a thick tissue. The carpet has an area of 1m 2 with a thickness of 1.5 cm. off-the shelf manual air pumps are placed in a square like shape at the center of the carpet. The distance separating 2 air pumps has the size of one foot (32 cm) on the same side. This length is chosen to avoid having the foot on 2 air pumps at the same time (Figure 3b). C. Signal conditioning and processing electronics The signal conditioning and processing electronic board acquires the differential output voltage from the sensors. This 234
(a) Structure concept of the sensor (b) Motorola MPX series differential pressure sensor Fig. 2. (c) MPX pressure sensor schematic Motorola pressure sensor working principle. Fig. 4. Threshold based fall detection algorithm. The threshold value is obtained after extensive fall trials on the carpet. It is around the digital value 600. The system automatically resets after the person is lifted from the floor. The flow chart of the fall detection algorithm is illustrated in Figure 4. (a) Air pump to pressure sensor connec-(btion pumps Smart carpet dimension and air distribution Fig. 3. Pressure sensor connection and distribution. signal is relatively weak (ranging from 20 to 100 mv); therefore, the first step is to amplify the signal. Microchip MCP602 operational amplifier is used for this purpose. The planned circuit has two stages of amplification and the output voltage gain is controlled through input and feedback resistors. The output voltage V out of the amplifier circuit is afterwards injected to the A to D (Analogue to Digital) converter of the Microchip PIC16F876A microcontroller. The latter has a 10- bit A to D converter, a large program memory and runs on an 8MHz crystal clock. Microchip microcontrollers are low cost and satisfy the needs of a wide range of applications. D. Fall detection algorithm The fall detection algorithm uses a threshold based method to trigger the alarm in case of a fall event. Sensor analogue signals are sampled at 21 KHz. They are constantly read and digital values are stored. Referring to the above sensor distribution, it shows that 3 or more sensors are triggered when a person falls on the floor. In this case, values of the triggered sensors are each compared to a specific threshold T H fall. If a combination of at least three sensors is triggered than a potential fall may have occurred. In this case, the system sends and SMS notification message person on the ground. IV. EXPERIMENTAL RESULTS At this stage a prototype of the Smart Carpet system has been developed using the previous conceptual design strategies. The fall detection algorithm was also implemented and programmed onto the microcontroller. To validate the reliability of the system, we have considered various scenarios. Three students at the faculty of engineering volunteered to perform the tests. The average height and weight of the students are 171 cm and 75.3 Kg. They were instructed accordingly to perform the proposed scenarios. Each scenario was repeated 3 times which makes a total number of 9 tests. The following scenarios were considered: Standing, walking, running, jumping, sitting and falling. To broaden our scope of study, many cases were also considered in each scenario. The falling scenario is of a particular attention. During this scenario, we have only considered the case when the person falls on the carpet, since the prototype has a small area (1m 2 ). For this test, the person is originally standing outside the carpet area. The person is instructed to fall on the carpet and lie on his either sides ending with different shapes to cover at least 3 sensors. Elders were not considered in this experiments to avoid injuries during repetitive tests. The values of the sensors shown in the table below are raw digital values obtained from the A to D converters of the microcontroller. Values are coded in 10 bits binary resolution. These values are displayed in decimal on LCD (Liquid Crystal Display) added in the electronic board. This board has also a serial RS-232 interface to communicate to a host computer for future analysis and testing. Table I shows the testing results of the suggested scenarios. Similarly, our experiments must take into account the presence of pets, since the carpet can be 235
TABLE I. EXPERIMENTAL RESULTS OF DIFFERENT STUDY SCENARIOS ON HUMANS. Study scenarios on humans # Scenario Study case Sensor value FP TP System status 1 2 3 4 5 6 Standing: One or more persons are standing on the carpet. Normal walking: The person normally walks on the carpet. Running: One or more persons run on the carpet. Jumping: One person jumps on the carpet. Sitting position: One person is sitting on the carpet. Falling: The person falls on the carpet. Case 1: One person steps on 2 sensors while standing. Case 2: Two or more persons are standing on the carpet. Case 1: The person steps on 1 sensor while walking. Case 2: The person steps on 2 sensors while walking. Case 3: Two persons walk normally on the carpet. Case 1: The person steps on 1 or 2 sensors while running. Case 2: Two persons step on the carpet while running. Case 1: The person jumps on the carpet and his feet hit only 1 sensor. Case 2: The person jumps on the carpet and his feet hit 2 sensors. Case 1: The person sits on one sensor. Case 2: The person sits on one sensor with one hand on another sensor. Case of fall: In this case the person falls on 3 or 4 sensors at the same time. At this point, values of the 3 or 4 sensors are above the T H fall 1018 2/9 0 500 < V < 1018 2/9 0 1018 1/9 0 Values of S i>=3 > T H fall = 600 0/9 8/9 Alarm SMS sent TABLE II. EXPERIMENTAL RESULTS ON ANIMALS. Study scenarios on animals # Scenario Study case Sensor value System status 1 2 Walking: Animals walk on the carpet. Lying: Animals lie on the carpet. Case 1: The animal legs step on 1 or 2 sensors. Case 1: The animal body may lie on more than 2 sensors. 800 Values of S i>=3 < T H fall deployed for indoor environments. We have considered in this scope two possible scenarios: the walking of animals on the carpet and the situation where the pet lies on the carpet. The results of these two cases are shown in table II. Table I and II show the experimental results performed on humans and pets respectively. Table I supplements 2 additional parameters that further investigate the sensitivity and specificity of the system. These parameters are: FP (False Positive): The event when the system triggers an alarm but no fall has occurred. TP (True Positive): The event when the system triggers an alarm when a fall has occurred. The sensitivity and the specificity of the system are therefore calculated using the statistical equations 1 and 2 below: T P Sensitivity = (1) T P + F N with FN(False Negative) = N-TP, whenre N is the total number of tests conducted on humans. T N Specif icity = (2) T N + F P with TN the total number of non-fall tests with negative result. V. DISCUSSIONS The data displayed in table I deliver a sensitivity of 88.8% and a specificity of 94.9%. These results indicate the reliability of the proposed sensing technique as well as the proposed fall detection algorithm. As a matter of fact, one of the main characteristics of the utilized differential pressure sensor is that its output voltage drops rapidly after the pressure is exerted. This will dramatically reduce the possible false alarms when more than one person walk or run on the carpet. If we take for example case 2 in scenario 1, the first person may step on 1 or 2 sensors following by another person who may also step on both sensors. In this case the system will not trigger an alarm since the voltage of the sensors triggered by the first person has dropped before the advent of the second. Controversially, when two people step on 3 or more sensors at the same time, the system may trigger a false alarm. Moreover the distance between sensors may also help reduce false positive events. If we take for example case 3 in scenario 2, false alarms may happen when more than 2 sensors are triggered. In this case the foot can either fall on one sensor or in between. The same situation applies to the running scenario. In the latter, the step time on the sensor is also very small. The experimental data on pets show 0% sensitivity and 100% specificity. However, these values are ideal and may 236
vary. They are largely dependent on the weight and the length of the animal. If the animal has a large weight, it may trigger a false alarm when it lies on the carpet. In this case, the value of threshold T H fall may be adjusted to remove false positive situations. Experiments on pets were limited to one test only. The system response is also very fast. The fall detection algorithm runs on a microcontroller clocked at 8MHz which makes the readings of the sensors and the decision almost simultaneous. Performance of the decision making tree were tested on 4 sensors only. Future work will investigate the behavior of this method when more than 4 sensors are deployed. In this case, clustering of sensors must be taken into consideration in the decision of fall flow. The smart carpet prototype was designed with 4 sensors on a 1m 2 area; however, it can be deployed on a large scale for indoor use in home care environments. In this case, design for manufacturing parameters must be taken into consideration. Parameters such as the thickness, size of the air pumps, air tubes routing, conditioning electronics and power supply are critical. Our present prototype has a thickness of 1.5 cm. The air pumps used were of the shelf devices taken from blood pressure measurement units, and the electronic board is powered with a 9V battery. Moreover, our carpet has also a low cost which makes it a potential commercial product. The sensing technique used in our smart carpet was designed for fall detection; however, it can be integrated in mats to alert patients from fall during sleep. As a matter of fact, air pressure pumps are more suitable for this kind of application. They can be easily embedded in mats where thickness is not an issue. These pumps will react to the pressure of the body when it reaches the edge of the bed. A different processing algorithm must be considered to alert the patient while maintaining his comfort during his sleep. [2] C. Franco, J. Demongeot, Y. Fouquet, C. Villemazet, N. Vuillerme, Perspectives in home TeleHealthCare system : daily routine nycthemeral rhythm monitoring from location data, In Proc. of Int. Conf. on Complex, Intelligent and Software Intensive Systems, pp. 611-617, February 2010. [3] A. Yazar, F. Erden, E. Cetin, Multi-Sensor Ambient assisted living system for fall detection, [4] J. Cheng, M. Sandholm, B. Zhou, M. Kreil, P. Lukowicz, Recognizing subtle user activities and person identity with cheap resistive pressure sensing carpet, In Proc. of Int. Conf. on Intelligent Environments, pp. 148-153, June, July 2014. [5] J. Miguel, A. Morgado, A. Konig, Low-Power concept and prototype of distributed resistive pressure sensor array for smart floor and surfaces in intelligent environments, In Proc. of Int. Conf. on systems, signals and devices, pp. 1-6, March 2012. [6] M. Contigiani, E. Frontoni, M. Adriano, A. Gatto, Indoor peaple localisation and tracking using and energy harvesting smart floor, In Proc. of Int. Conf. Mechatronic and Embedded Systems and Applications (MESA), pp. 1-5, September 2014. [7] I. Al-Naimi, C. Biu Wong, Ph. Moore, C. Chen, Advanced Approach for indoor identification and tracking using smart floor and Pyro electric Infrared Sensors, In Proc. of Int. Conf. in Information and Communication Systems (ICICS), pp.1-6, April 2014. [8] V. Anh Ho, S. Imai, S. Hirai, Multimodal flexible sensor for healthcare system, In Proc. of Int. Conf. IEEE Conference on Engineering in medicine and biology society (EMBS), pp.5976-5979, August 2014. [9] A. Steinhage, C. Lauterbach, SensFloor and NaviFloor: Large-area Sensor system beneath your feet, In: N. Chong, F. Mastrogiovanni, Handbook of Research on Ambient Intelligence and Smart environments: trends and perspectives. IGI global Hershey [10] J. Cantoral-Ceballos, J. Vaughan, P. Wright, C. Brown-Wilson, P. Scully, B. Krikor, Intelligent carpet system based on photonic guided-path tomography for gait and balance monitoring in home environments, IEEE sensors Journal, Vol 15, No. 1, pp.279-289, January 2015. VI. CONCLUSION Falls represent a major public health problem in elderly society. Thus, the development of technologies to prevent falls of old people and ensure their protection is essential. A new sensing technique using differential piezoresitive pressure sensors has been designed and developed in this paper. This technique was integrated on a carpet. The system uses a threshold based fall detection algorithm that can trigger an alarm by sending an SMS message. After conducting several tests and experiments, the system ensures the detection of falls by a sensitivity of 89% and a specificity of 95%. Experiments were also conducted on pets as well. The developed prototype is a potential commercial product that can be deployed for indoor use in home care environments, however, more testing must be conducted to increase the sensitivity and the specificity of the system. Finally, the developed sensing technique is also suitable to alert patients from falling off their beds while sleeping. REFERENCES [1] Tromp, A. M., et al. Fall-risk screening test: a prospective study on predictors for falls in community-dwelling elderly. Journal of clinical epidemiology, vol 54, No. 8, pp. 837-844 (2001). 237