Abstract
Recently, human tracking method using cameras and image processing has been developed as the monitoring system. These methods are possible to detect human position and behavior with much circumstance. However, pattern recognition with computer vision is always accompanied with the difficulty process and condition such as occlusion problem caused by complexity indoor environments and various change of human feature caused by perspective characteristic. Therefore, we propose a robust tracking method with stereo omni-directional cameras in an indoor scene. Our approach uses 3D cylindrical model based on the feature of human profile in the 2D image. Moreover, we employ useful constraint by common foot position in the stereo images to eliminate the tracking error. These ideas give us the robust recognition for estimation of human position, even if the captured human feature has some deficits by the obstruction.
