Abstract:

As a widely used traditional tracking technique in visual surveillance, Mean Shift algorithm has a deficiency in handling moving targets with high speed or large scale change. In order to sove this problem, a bandwidthadaptive tracking algorithm based on Mean Shift and Kalman prediction was proposed. The algorithm uses Kalman filter to predict the positions of fast moving objects in the successive frame, which are as the initial positions for Mean Shift iteration. Bandwidth trials is utilized to adjust the bandwidth automatically for targets' scale change. The experimental results of pedestrians and vehicle tracking show that our algorithm is effective and robust.

Key words: Mean Shift;object tracking;Kalman prediction;bandwidth trials