Research Of Video Target Tracking Algorithm Based On Particle Filter

In recent years, with the continuous improvement of computer data processing ability, hardware configuration and advanced imaging technology, computer technology has been developing rapidly, video target tracking has become one of the key problems of computer vision. Video target tracking is widely used in the field of video surveillance, intelligent traffic, intelligent navigation, human-computer interaction, medical diagnosis and virtual reality, etc. Although a number of effective video target tracking algorithms have been proposed, the performance of target tracking is unsatisfactory in the real scene. It is often faced with various difficulties such as the indistinguishable color of target and background. target’s movement speed too fast, the diversity of target’s morphological changings and light mutation. Therefore, it is still a challenging task to design a robust video target-tracking algorithm.This paper researches the current of video target tracking and particle filter theory, then introduces the representative algorithm of moving target tracking based on prediction, including Meanshift, Camshift, Kalman filter and Particle filter algorithm; at last explains the advantages of particle filter algorithm being applied to target tracking. Aiming at the problem of static target detection, this paper proposes a detection algorithm based on particle filter framework. The multi-scenario experimental results reveal that the method has better detection precision and robustness to static target. Meanwhile, it reduces the false alarm rate. The main work of this paper is as follows:(1) This paper researches the common techniques used in video target tracking, including the target representation, target feature extraction, target tracking, classification and common target-tracking algorithms.(2) The basic theoretical knowledge of particle filter algorithm is stated, including Bias state estimation, importance sampling, particle scarcity and resampling, and the classical particle filter algorithm is introduced. Then, An improved algorithm is proposed, which to solve the drawbacks of the classical particle filter algorithm, And simulation experiment is compared. The experimental results reveal that the improved method compared with the basis particle filter method have more accurate result and the less error of estimation.(3) We study the target motion model and observation model in the video target tracking process. Due to the low robustness of particle filter target tracking using single feature, an adaptive multi-feature fusion method is proposed. A discussion is carried out with experimental data on this multi-feature method. The experimental results reveal that video target tracking algorithm based multi feature fusion compared with single feature have better robustness.(4) This paper introduces the shortcomings of particle filter method in the static target detection, and puts forward some improvement measures. The proposed method adds background model to the existing particle filter frame, using the center point and objective deviation, and then a relevancy measure is calculated between tracking object and the background template to determine the object’s final state. The experimental results reveal that the method improves the accuracy and robustness of static target detection.