Target Tracking Based On Particle Filtering

The problem of target tracking is used widely in civil and military applications, especially visual target tracking. It plays an important role in the video surveillance and traffic regulation. The main task of object tracking is to detect moving object accurate from each image frame, and position those object in the subsequent frames. Then we could get full trajectory of moving object. Because of the unique advantage and broad prospects in dealing with the state estimation and parameter estimation in the non-Gaussian and non-linear systems, Particle Filtering (PF) has become the hot topic in target tracking applications. So, studying PF intensively and applying it to visual target tracking have very important theoretical significance and practical value.The research content and innovation are as follows:1. This thesis summarizes some of existing stochastic filtering algorithms, the visual target tracking problem and their research status. It focuses on the optimal estimation of the linear Gauss system Kalman Filter (KF) and the suboptimal estimation of the nonlinear Gauss system Extended Kalman Filter (EKF). And it studies PF in depth, including the problems of Monte Carlo integral, importance sampling, resampling and the choice of importance function.2. It investigates PF’s weakness in details. According to the degeneracy phenomenon, it studies two improved PF algorithms including Auxiliary Variable Particle Filter (AVPF) and Regularized Particle Filter (RPF). Then, it analyses the performance of three algorithms contain EKF, PF and RPF using the simulation tools of MATLAB. Simulation results reveal that the estimation performance of PF and RPF is much better than EKF. The anti interference performance of RPF is more stable than PF, especially in low noise systems.3. The visual target tracking problems are dealt with, where the hydraulic synchronous value (HSV) color space model is utilized and a PF tracking algorithm is derived based on color features. This tracking algorithm considers target’s location, velocity and size, and adds angle information to state models and has high estimation accuracy when the color features of the tracking target are different from the environment around even the target is partially sheltered, and its anti interference performance is also good.