Visual Target Tracking Method Study In Complex Scene

Visual target tracking is one of the most important and popular field of computer vision. It combines image processing, pattern recognition, artificial intelligence, image content understanding and other related fields of knowledge. It is a kind of technology that detect target from image sequences, obtain target state, automatic target tracking. It can be used for target detection, target recognition, image content understanding, scene classification, behavior understanding, human-computer interaction, robot navigation and other fields. Complex scenarios single target tracking and multi-target tracking is the most difficult visual target tracking problem has not been solved.Target tracking under complex scene due to the complexity of tracking scene result it has a higher degree of difficulty. In the complex scenario because changes in light intensity, background interference and the same scenario objects have strong similarity, so target characterization is very difficult. As the target motion state in complex scenes with non-linear, non-Gaussian, multi-mode, high noise, and so tracking algorithm also has more stringent requirements. Tracking method based on particle filter is the main tracking method in complex scenes. However, the number of particles in particle filter is also the more higher the complexity of the algorithm, since resampling to a large sample weights are selected many times, resulting in decreased diversity of sampling results, making particle filter poor problem occurs. In the multi-target tracking is a difficult target association probability estimates.Address these issues my work includes: Color and gradient direction dual information were used to describe in complex scenarios. SIFT feature was used to describe object in multi-target tracking. The particle swarm optimization (PSO) was introduced to optimize the particle filter. The particles move toward the region of larger posterior probability density to reduce the required samples, while increasing the diversity of samples. This paper presents a new image tracking algorithm PSO-PF.A new association probability estimation method was proposed based on conditional random field in the multi-target tracking.