A Continuously Adaptive Template Matching Algorithm for Human Tracking
Shusheng He, Alei Liang, Ling Lin, Tao Song · 2017
A continuously adaptive template matching algorithm for depth camera is proposed for tracking human upper body. In contrast to traditional method for image patch matching, the background behind the target is separated, also the size of the tracking window is adjustable based on the distance of the target, which discard most of the disturbance in a relatively larger tracking window with smaller target. To better locate human head and shoulder, we record a mass center in the target window for the first confirm of Haar-feature detection, and continuously move our tracking window for each template match according to the shift of the mass center. The algorithm works in real-time and the target won't get lost during turning round.