Research On The Technologies Of Interesting Visual Object Tracking Based On Particle Filter

In recent years, in the video surveillance system to monitor the track of the moving target attracted the attention of scholars home and abroad. Video tracking integrates technologies of computer vision, pattern recognition, artificial intelligence and so on. It is important for military and civilian. In reality, the video target state has probability distribution of non-linear, non-Gaussian characteristics, and therefore the theory of technical studies related to the significance of target tracking has become an important field of research and development trends. Particle filter can be used for any state-space model representation of nonlinear systems, as well as traditional Kalman filter can not be expressed in nonlinear systems, which can be implemented by a Bayesian recursion process though a Monte Carlo simulation method.This thesis studies the theory of interest in video object tracking, which is based on particle filter. Based on the use of particle filter for target tracking theory to solve the technical problems faced by around how to improve the characteristics of target particle filter for target tracking presents a modified particle filter tracking algorithm; Description on how to improve the efficiency of target particles and algorithm for real-time communication proposes an effective solution method; Finally, a system of visual tracking based on particle filter has been designed. The main contributions of the thesis are as follows:1. First, we introduced to the theory of particle filter with systematic. And then we explained the advantages and disadvantages of particle filter, which compared to the field with the current video target tracking filter commonly used in the classic algorithm (Kalman filtering and extended Kalman filter). Pseudo-code of every variants of particle filter is given. Several variants of particle filter such as SIS, MCMC improvement strategies, UPF, APF, RBPF, RPF are compared. The advantages and disadvantages of them are discussed.2. A visual tracking based on particle filter algorithm has been proposed, which integrates both colour feature and the spatial information. By using spatial feature of object model and particle filter algorithm of colour histogram, the similarity of the Bhattacharyya coefficients and marginal spatial density ratio between two frames is compared. Computer simulation results demonstrate that the proposed algorithm is more robust as compare to the traditional visual tracking algorithm which just using colour feature. 3. In this paper, we studied the mean shift and Kalman filter algorithm for their integration into the framework of particle filter applications. As we know, the standard color particle filter spreads each particle with random walk method for target tracking. The traditional color histogram is used that can not reflect the characteristics of the target space. Thus, we propose a novel color-based particle filter target tracking algorithms. The second-order color histogram is applied to get the observation of particle probability density function. In addition, the Kalman filter is used to determine the spread of particle dynamic model of the uncertainty in the drift of the state. The distribution of particles is more accurately close to the probability distribution of the target, thus the use efficiency of particles is greatly improved. Computer simulation results demonstrate that the proposed algorithm is more robust as compare to the traditional color-based particle filter tracking algorithm.