Multi-Object Tracking Based on Improved Mean-Shift Algorithm

A multi-object tracking algorithm is proposed for road & bridge traffic scene. Firstly, background reconstruction was conducted based on a statistical model, and the background was updated using Kalman filter at regular intervals. Secondly, the background differencing was conducted to obtain potential objects. An improved mean-shift tracking algorithm was put forwarded for image sequences without obvious color information, and the histogram difference was computed between the histograms of background and moving object. Then, the back projection image was obtained according to the histogram difference. The feature extraction is effective to distinguish the background and foreground. Afterward, a multi-object tracking link list and tracking state list were proposed for tracking. The proposed algorithm has been compared with basic mean-shift algorithm. Result shows it provides better accuracy and achieved good real-time performance.