Design and Implementation of Object Tracking System Based Mean-Shift with Locust Search Optimization on Raspberry Pi

Abstract

Having an optimal object tracking system is an advantage especially in everyday use. One of the example is use in victim localization. The optimal object tracking system as mentioned earlier means a system that capable of reaching the convergence point of region of interest (ROI) with small number of iterations. The smaller the number of iterations needed for reaching the convergence point, the faster the system follows the object’s movement. A system that is able to follow the movement of objects quickly, especially in accordance with case studies of victim localization, is certainly needed. In this paper, the object tracking system is built on Mean-Shift algorithm. Mean-Shift has a convergent search technique that requires a large number of iteration processes. In order to achieve an optimal object tracking system as the purpose of this paper, an optimized algorithm is carried out. Instead of shift the ROI point sequentially until reach the optimum point as done in Mean-Shift, the optimization algorithm will search the ROI optimum point randomly with a movement that pays attention to the previous optimum point. Optimization algorithm used in this paper is Locust Search algorithm. Object interest being tracked in this paper is human face. Object tracking system will be deployed in Raspberry Pi 3 Model B+ because of its characteristics that are suitable to be implemented in this paper’s case, in form of prototype.