|Institution:||University of Cincinnati|
|Department:||Engineering and Applied Science: Mechanical Engineering|
|Keywords:||Mechanical Engineering; Human Motion tracking; Kalman filtering; Inverse Kinematics|
|Full text PDF:||http://rave.ohiolink.edu/etdc/view?acc_num=ucin1406809906|
The advent of portable video and image sensing devices such as Microsoft Kinect have provided computer scientists and engineers an opportunity to exploit data obtained from these devices to help with monitoring different aspects of human activities. However, human motion tracking has always been a challenging area of research. The tracking process has an additional dimension of complexity when the results are solely dependent on the vision based sensors and when the tracked user is not wearing any contact sensors such as accelerometers, gyroscopes, etc. This research addresses the problem of motion tracking of human skeleton system using non-wearable vision based sensors. The application areas of such a system will be in patient care and nursing homes where the focus is on having passive sensors that do not affect the day to day activities of the individuals living in such homes.The proposed approach combines synergistic paradigms of image processing, kinematics of rigid bodies and Extended Kalman Filtering scheme to estimate the motion of a human skeletal system which may be studied as a system of kinematic linkages. The unique aspect of this research is that, this approach solely depends on the measurement obtained from the vision sensors without involving any wearable or inertial sensors to measure the motion parameters. In this research we propose fusion of two filtering schemes- the optical flow equations applied to raw images obtained from the Microsoft Kinect and extended Kalman filter for human skeleton considered as a system of kinematic linkages. The novel approach proposed in this research presents a an effective estimation algorithm which is tested with the help of experiments performed using the Microsoft Kinect sensor and compared using accurate tracks obtained from 24-Camera Opti-track motion capture system.