Control Strategies and Particle Filter for RGB-D Based Human Subject Tracking and Behavior Recognition by a Bio-monitoring Mobile Robot

• Published in: Intelligent Robotics and Applications pp. 318 - 329
• Main Authors: Imamoglu, Nevrez, Nergui, Myagmarbayar, Yoshida, Yuki, Gonzalez, Jose, Yu, Wenwei
• Format: Book Chapter
• Language: English; Japanese
• Published: Berlin, Heidelberg Springer Berlin Heidelberg 2013
• Series: Lecture Notes in Computer Science
• Subjects: contour data; fuzzy control; home healthcare; human behavior recognition; Mobile robot; tracking strategies
• ISBN: 9783642408519; 3642408516
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-642-40852-6_33

Our ultimate goal is to develop autonomous mobile home healthcare robots which closely monitor and evaluate the patients’ motor function, and their at-home training therapy process, providing automatically calling for medical personnel in emergency situations. In our previous study, we developed basic algorithms for tracking, measuring, and behavior recognition of human subjects by a mobile robot, thus, demonstrated the feasibility of the idea of bio-monitoring home healthcare mobile robots. In this study, in order to realize effective bio-monitoring robots, we investigated 1) color based particle filter subject tracking with proposed depth likelihood integration to control the weights of particles; 2) control schemes for acquiring stable image sources for further human motion analysis, especially, the algorithms for reducing the camera vibration due to the acceleration and deceleration of the robot; 3) human activity recognition using contour data of the tracked human subjects extracted from depth images. Results showed that, depending on depth data can be quite useful as an observation by simplifying state space in 2D rather than 3D state space, and, a fuzzy control algorithm could decrease the vibration due to the acceleration and deceleration. Finally, the human activity recognition could be achieved with a high correct rate, by using geometric parameters extracted from contour data.