Comparative Analysis of Mean-Shift Based Object Tracking Using Simulated Annealing and Locust Search Algorithm Approaches

摘要

Tracking process is a process to find the convergence value of 2 ROI boxes, i.e. object target ROI and candidate ROI. In other word, convergence is a situation where the value of those 2 ROI boxes has a high similarity value. Because the main process in tracking is update the object's position continuously, so it is important to pay attention to the processing time needed to reach the convergence point of the object target on each video frame while searching for ROI. Mean-Shift tracking algorithm has a deficiency in its technique during the ROI convergent search process. To overcome the deficiency of the Mean-Shift technique in searching for convergent ROI, the optimization algorithm is used. This research aims to produce a Mean-Shift based object tracking system with faster processing time performance to reach the converging point of the object target on each video frame. This research will provide analysis of optimization algorithm i.e. Simulated Annealing and Locust Search Algorithm in the process of finding optimum ROI point. Performance evaluation using one tail t-test testing technique with different variant assumptions. The performance result of both optimization algorithms will be shown in form of chart and t-test tables. The result summary obtained by using optimization algorithm due to searching for convergence point of ROI shows that Locust Search algorithm produce the better performance. The process time needed for Locust Search algorithm is 151.4 milliseconds, faster than Mean-Shift with 275.2 milliseconds and Simulated Annealing with 172.9 milliseconds.