Design of Mean Shift Tracking Algorithm Based on Target Position Prediction
Hui Wang, Xueying Wang, Lijun Yu, Fei Zhong · 2019
For the traditional Mean Shift algorithm in target tracking, the Mean Shift tracking algorithm can not adapt to the scaling, fast motion and occlusion of the target. This paper presents an improved Mean Shift tracking algorithm based on the predicted target position. The algorithm uses inter-frame clustering motion estimation and scale estimation to predict the initial search position of Mean Shift algorithm in each iteration. The algorithm can track fast moving targets while ensuring that the search window can adapt to changes in target scale. When the target is occluded, the target prediction value can be used to track and locate the occluded target. By analyzing the simulation results in complex environments, the improved Mean Shift tracking algorithm can track moving targets in real time quickly and accurately.