Real-time Human Tracking in Stereoscopic Environments using Subtraction Stereo

Kenji Terabayashia, Yuma Hoshikawab, Yuki Hashimotoc and Kazunori Umedad

Chuo University/CREST, JST, 1-13-27 Kasuga, Bunkyo-ku, Tokyo 112-8551, Japan.


This paper reports real-time human tracking in stereoscopic environments using a calibrated stereo camera with subtraction stereo method. Subtraction stereo is a stereo matching method that focus on moving regions to restrict matching areas, and is fast and robust to correspondence problem, one of the most serious issues in computer vision. By focusing on moving regions, people can be detected easily based on standard labeling and information of object size that is directly obtained from subtraction stereo without information about the position and pose of installed camera. To Track multiple crossing people, the extended Kalman filter technique is employed to address occlusion problem usually seen in many people environment. The proposed method is verified by the experiments in unknown stereoscopic environments.

Keywords: Subtraction stereo, Human tracking, Stereoscopic environment.

     Back to TOC