CamShift-Based Tracking in Joint Color-Spatial Spaces

摘要

This paper presents a visual tracking algorithm that is based on CamShift. Both the face and upper body are utilized simultaneously to perform tracking. They are first tracked independently by applying two separate CamShifts which continue tracking from the locations determined in the last time step and use only color probability images. Next, the candidate locations are subjected to CamShift which operates on distributions reflecting additionally geometrical relations between the face and the body. The aim of the CamShift-based searching in the joint color-spatial space is to find the mode. Experimental tracking results on meeting video recordings are presented. They demonstrate that this algorithm is superior over traditional CamShift. Furthermore, it is very simple and computationally fast.