Superpixel tracking using Kalman filter

Abstract

In this paper, we propose an algorithm for tracking of moving objects in video sequences. Our method uses Kalman filter to predict the location of target and exploits superpixel based tracking algorithm to find the real position of target in a search region surrounding the predicted location. The motion dynamics and equations from mechanics physics are used to design a Kalman filter with assumption of constant acceleration motion. Using this Kalman filter makes our method able to handle long lasting occlusions. We have also devised a scheme that helps the tracker to find the target after long lasting occlusions.