Research On The Visual Object Tracking Based On Particle Filter And Mean Shift

Video target tracking technology is computer vision, machine learning, digital image processing, is a hot and difficult problem in the field of research, the technology has been more and more widely used in intelligent surveillance,intelligent transportation, human-computer interaction and video compression, and other fields. In recent years, many scholars on the video target tracking technology is studied in-depth, and put forward many improved tracking algorithm. In video target tracking techniques commonly used in tracking algorithm with particle filter tracking algorithm and Mean Shift tracking algorithm. In this paper, the particle filter algorithm and Mean Shift algorithm in video target tracking applications, mainly in the complex environment to improve the tracking algorithm robustness, accuracy and real-time tracking process, for the following research:To solve the problem of a single feature lead to tracking failure easily and computationally in a complex environment, this paper presents a multiple characteristics of particle filter tracking algorithm. The algorithm based on the color, structure, such as edge information fusion target observation model established strong, the observation model is established through fusion of color, texture and structure feature information, and various characteristics in the process of tracking fusion coefficient can be as the change of environment in the process of tracking adaptive changes, in the actual tracking several feature information can complement each other, when the failure of a feature, but also through other feature information of target tracking, so as to ensure the robustness of tracking.To solve the problem of the particle filtering algorithm using multi-feature fusion in tracking to ensure the robustness at the same time, the complexity of the algorithm increases, and proposes an adaptive strategy to adjust the number of particles are required in the process of tracking. When the target in a simple environment, due to the interference by the surrounding environment is small, a high degree of similarity with the target template candidate target template can be appropriate to reduce the number of particles required to achieve real-time tracking.If the target is in a complex environment, due to the greater impact on the surrounding environment, low similarity target template and the candidate target template, this time can be achieved by increasing the number of particles of the target tracking stability. The policy adjustment process to track the number of particles ensures the timeliness of the tracking process through adaptation.To solve the problem of the large amount of calculation of particle filter algorithm, Mean Shift algorithm is easy to fall into extreme, on the basis of integration of color, structure and texture features proposed that particle filtering tracking algorithm based on multi-feature fusion and Mean Shift. In particle filter algorithm using the Mean Shift clustering algorithm for tracking the process of particle clustering operation, which increases the particle weights, more close to the true state of the target, the algorithm ensures the robustness of tracking using the less number of particles.