Compressive gait biometric with wireless distributed pyroelectric sensors
Human tracking and recognition are desirable yet challenging for many applications including surveillance, computer vision, robotics, virtual reality, etc. Many biometric modalities have been used on these applications. Compared to other biometric modalities, such as fingerprints, face, and iris, gait biometrics are advantageous in their capability of recognition at a distance under changing environmental and cosmetic conditions. Despite having many limitations, from clothing changes to gait variation due to different physical and emotional conditions, the discrimination power of gait can still serve as a unique and useful component in human tracking and recognition systems. The work presented in this thesis aims at developing a distributed wireless sensor human recognition and tracking system, in order to improve the performance of previously established centralized pyroelectric sensor system. Our final goal is to provide wireless distributed pyroelectric sensor nodes as an alternative to the centralized infrared video sensors, with lower cost, lower detectability, lower power consumption and computation, and less privacy infringement. In previous related study, the system was able to succeed in identifying individuals walking along the same path, or just randomly inside a room, with an identification rate higher than 80% for around 10 subjects. For the human recognition system, innovations and adaptations are developed in: (1) sampling structure, multiple modified two-column sensor nodes are engaged to leverage the ability of effective acquisition of both the shape and dynamic gait attribution. (2) sensing protocols, different compressive measurement functions are provided for accomplishing the central task of compressive sensing protocol - choosing a proper scheme of the random projection encoding. (3) processing architecture, different levels of fusion schemes performed at data level, feature level, score level, and decision level constitute the processing architecture. Along with the advent of several new digital features, a higher recognition rate for both path dependent human recognition and path independent human recognition is achieved. For the human tracking system, a distributed tracking method was proposed to replace the previous centralized algorithm. Both recognition and tracking system will eventually be combined together and work cooperatively to form the human tracking and identification system. Real time implementation results are presented in the thesis. Moreover, experimental work and the related results are also discussed.