A visual tracking framework based on differential evolution algorithm

Visual tracking has become a hot research topic in the field of computer vision at present. The tracking process can be converted into a dynamic optimization problem. In this article, an image fitness metric which named the Structural Similarity Index (SSIM) is used to estimate the object location in the experiments. In addition, DE (differential evolution), a population based stochastic meta-heuristic algorithm is adopted for this optimization problem. According to the characteristics of the visual tracking, this article initialized the particles of the DE algorithm by Gaussian distribution which make the search faster and more accurate. By improving the iterative formula, selecting the optimal distribution of CR and F could balance the global and local search automatically. By introducing a new updating mechanism into the process of object tracking, the errors which generated in the tracking process can be compensated effectively. Then we combine the improved DE and SSIM to form a novel visual tracking framework which based on gray level information. Through the contrast test with other tracking algorithm including PF (particle filter), SPSO (sequential particle swarm optimization) and QPSO (quantum-behaved particle swarm optimization), experimental results show that the new tracking framework based on differential evolution algorithm achieved better tracking performance than others.