Model-based human motion analysis in single view video

Abstract

Tracking human motion in monocular video is a challenging problem in computer vision. It has found a wide range of applications such as visual surveillance, virtual reality, sports science, etc. This project aims to develop a model-based human motion analysis system that can track human movement in monocular image sequence with minimum constraint. No markers or sensors are attached to the subject. There is no need for the subject to wear tight clothing and occlusion will not seriously affect the tracking process. Monocular 3D human motion tracking is challenging, because the system needs to estimate a large number of degrees of freedom (DOFs) given the minimal amount of motion information. In this project, the number of DOFs is reduced to 12 under the assumption that the subject is walking parallel to the image plane. Given a clip of the video, the first step is to manually fit the 3D human model to the subject in the first frame of the video. Then background subtraction is used to extract the human silhouette. The background modeling and subtraction algorithm uses both color and edge information. A confidence map is used to fuse intermediate results and represent the results of background subtraction. Therefore, we propose the silhouette chamfer as the main matching feature. Chamfer distance measure is carried out on the extracted subject silhouette. The silhouette chamfer contains both the chamfer distance and region information. The chamfer algorithm searches for the best fit of edge points from two different images. Finally, we use discrete Kalman filter to predict the pose of the subject in each image frame. The update step uses Broydent’s method to optimize the predicted human figure’s parameters to fit the person’s silhouette by using the cost function. We use the gait database SOTON from the University of Southampton, UK to test our system. The testing image sequences contain human walking in both the indoor and outdoor environment. The motion tracking results demonstrate that our system has an encouraging performance.

Date of Award	4 Oct 2004
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Kwok Leung CHAN (Supervisor)