Study on Resident's Behavior Awareness with Diffusive Sensors in Home

Yukitoshi Kashimoto ( 1461001 )


Thanks to the astonishing progress of the ubiquitous computing technology, resident's behavior awareness with sensors in home attracts the attention. With the technique, we can realize various kinds of applications such as energy-saving home appliance control. The resident's behavior awareness system such as the indoor positioning or activity recognition system by utilizing camera has been gradually realized. However, those information systems can be applied to limited applications. Especially, for quick diffusion of those techniques to ordinary home, accurate recognition of activities with low-cost commodity sensors must be realized. Objective of this study is to solve several problems which disturbs the propagation of resident's behavior awareness, and to expand applicability of it. To investigate the reason for limited use, we categorized the resident's behavior awareness systems into two types in terms of system characteristics: the indoor positioning system and the activity recognition system in home. The indoor positioning system includes the floorplan creation tool. The indoor position system enables the resident's behavior awareness application to recognize the surrounding environment of the user. First problem is that there is no floorplan creation tool that adopts diffusive sensors and achieve accurate floorplan creation. Then, we utilize the prevailed device: smartphone, and easy-to-use measurement method: the user completes a lap along the walls of all rooms, and the tool estimates the accurate shape and size of the room. To realize the accurate measurement, we attach the ultrasonic distance measurement sensor to the phone and develop the technique to handle the noise effect from the object attached to wall. The evaluation result showed that the subject created the accurate floorplan. Next problem is that there is no indoor positiong system that adpots diffusive sensors and achieve accurate floorplan creation. Then, we work on the development of the vibration type estimation technique toward indoor positioning system. The proposed system estimates the position of the user by distinguishing the vibrations that occur when the user touches furniture. To design the diffusive and low-cost system, we utilize an easy-to-be-concealed and low-cost piezo component attached on the floor in home. To improve the recognition accuracy, we utilize MFCC feature to estimate the vibration type. Through the evaluation, the system estimated the vibration type with F-measure: 93.9%. The activity recognition system in home enables the resident's behavior awareness applications to recognize the state of the user. However, there is no system that utilizes diffusive sensors and performs accurate activity recognition. Then, we develop the activity recognition system to track resident's behavior in home. We develop the system by utilizing PIR, door, and watt sensors installed in home. To design the diffusive and low-cost system, we adopt the device-free and low-cost energy-harvesting PIR sensor and watt meter attached to each appliance, which does not demand the battery replacement to the user. To improve the recognition performance, we utilize Random Forest classifier and supplemental technique for PIR sensor's dead zone. The evaluation result showed that the system estimated the user's activity with F-measure: 68.6%.