metaheuristics for robust object tracking in video sequences multi object real time tracking.

Abstract

In these recent years, object tracking has attracted the interest of many researchers, because it offers many challenges as a scientific problem, as well as its potential industrial and commerce applications in human life, robotics, and surveillance. Moreover, it is a part of many prominent levels in computer vision problems, such as motion analysis, and activity recognition. Visual object tracking is a challenging task, as many approaches have been proposed, it is still unresolved problem because of the high number of the challenging factors. By considering the research trends in this line, three lines of researches were developed namely: the object presentation, the search mechanism, and the updating model. In this thesis, we are specifically focused in the study and the developing of metaheuristic based searching techniques used for tracking. The role of metaheuristic search algorithm is to find the most similar candidate to a previous defined template. Many similar related works have been proposed in this line, their main disadvantage is the convergence at local minima which make them unable to find the exact position; to overcome this drawback, we proposed four different tracking frameworks, whose three, are single object tracking based methods, and one is multi-object tracking based method. The first part of this thesis is dedicated to implement Stochastic Fractal Search (SFS) algorithm, study and analyze the effect of using different population sizes; also the implementation results of the Harris Hawks Optimizer (HHO), in the same context. When the second part of this thesis is dedicated to implement, study and analyze the effect of using both population sizes, and iterations number on the tracking accuracy with WOA algorithm, combined with a very discriminative appearance model