Research On Moving Object Tracking Based On Particle Filter

At present, with the large number of applications in computer vision systems, more and more attention has been paid to the target tracking, such as stadiums monitoring, intelligent transportation, face recognition and detection, public safety and so on. As a key applications, target tracking is used widely in many areas, so it must be adapted to different environments. Traditional tracking methods with different characteristic could result in different effects. Especially, because the weather conditions, the target occlusion, the object deformation, they can not result in a good solution and the phenomenon of tracking failure or errors might be produced. To solve the above problems, in this dissertation an improved particle filter estimation algorithm is introduced to accurately track the moving target.How to track the moving target more stable and accurate is the core of this dissertation. Among the feature of color, shape and texture, the dissertation choose the feature of color as the observation model. Because some of the color characteristics of non-target information can be brought in, negative effect on tracking accuracy might be produced. The effect of the boundary useless information become weak by adding weight function to color model, while the effect of the object's information become strong. Real-time updating templates is very helpful to remove the interference colors.Particles degradation often takes place in the conventional particle filter algorithm. A weight optimal has been introduced in this dissertation. Which is increased partial particles participating of the weight in the algorithm, and samples the particles with bigger weight in the status's estimating. The problem of declining tracking accuracy due to the lack of diversity of the particles in the process of the particle degradation has been solved. The dissertation not only studied theory knowledge, but also carried out several simulation experiments divided into two kinds of static and motive background. As for static background, two simulation experiments have been finished to verify the algorithm in different environment. Compared with the tracking results, it shows that the method owns has higher accuracy.