Sequential Monte Carlo Particle Filter And It's Applications In Visual Tracking

Sequential Monte Carlo, namely Particle filter is a real time inference algorithm. Because of its unique characteristics such as efficient to process nonlinear problems, extensive applicability, particle filter has been paid more and more attention in recent years. Although there are some theoretical and practical achievements, the use of particle filter is still in its infancy, there exist many basic issues need to be investigated.This thesis focuses on the key techniques of improvement strategies to increase particle filter's performance and its application to visual tracking. To improve the particle filter, weight degeneracy and choice of importance density are studied. In the application of visual tracking, this thesis proposes an incremental learning based visual tracking algorithm using adaptive particle filter. This method has three characteristics, firstly, the eigenbasis vectors in subspace are trained online. Secondly, by using RSR resample method the number of particles can adaptly be changed acording to the size of the sum weight of these particles. Thirdly, according to the size of the sum weight, the updating method of the eigenbasis vectors in subspace can be adjusted adaptly in oder to track the target. Experiments show that the proposed method is more precise and robust under conditions such as large appearance variation, pose variation and lighting variation.