Real time occlusion handling using Kalman Filter and mean-shift

Abstract

Tracking objects using Mean Shift algorithm fails when there is a full/partial occlusion or when the background color and the desired object are close. In this paper we proposed a method using Kalman Filter and Mean Shift for handling these situations. Using similarity measure of Mean Shift algorithm we are able to detect an occlusion. Kalman Filter comes into the play for occlusion handling in a Buffer-Mode Process. We implemented this algorithm both on PC and DSP 64x+ Texas Instrument and the results are both tabulated. The results reveal the ability of our method to locate the object soon after occlusion disappearance.