| In recent years, visual tracking has become a hot issue. There is a demand and broad prospect in the field of video surveillance, video retrieval, virtual reality, medical diagnosis and sports video detection etc. But the complexity of the real tracking conditions challenges the state-of-the-art trackers, including the effect of illumination, occlusion, object deformation and motion uncertainty etc. Although many researchers have put forward many effective methods, but there remain many suspended problems.This paper made main research on the visual model constructing problem in the framework of particle filter, and proposed a optimization based model. In detail, through the study on the existed visual model extraction methods, we defined a optimization model, and employed the genetic algorithm to optimize the parameters of multi-cue integration model. The main contents of this paper include:1.The Particle Filtering FrameworkFirstly, study on the basic particle filter framework, and summarize the key problems of the particle filter, to achieve the basic framework of particle filter algorithm.2.Visual ModelSecondly, through the analysis on the existed visual model, summarize the problems of existed multi-cue integration model, and explore the possible solutions.3.Genetic Algorithm based multi-cue integrationFinally, this paper puts forward genetic algorithm to optimize the multi-cue integration model. We transform the visual model problem into an optimization problem by constructing optimization model. Specifically, the genetic algorithm is employed to achieve the goal of robust modeling, by introducing the genetic algorithm into the multi-cue integration model.In this paper, the above model was tested in the video database, and the experimental results was analyzed and proved robust to the problems of complex scene, occlusion, appearance changes and abrupt motions. The proposed method is validated to be robust, stabile and efficient. |
PDF Full Text Request