Research On Adaboost Based Multicolor Space Learning For Target Tracking

Computer vis ion is one of the research hotspots in the current field artificialintelligence, and a key role of the human visual is to track the moving object.Because of A large amount of information contained in the movement in real life,tracking of moving object has become an important research area of the fie ld ofcomputer vision. Research of tracking of moving object involves many fields,such as pattern recognition, image processing, machine learning, signalprocessing, statistical and optimization methods and so on. Research of objecttracking has important theoretical value. And because of wide application of thevideo tracking system, accurate tracking of moving object has impo rtant practicalsignificance in the field of human-computer interaction, navigation traffic,movement-based object recognition, intelligent control, video compression,weapons guidance, etc.So-called moving object tracking means to estimate the trajectory of thetarget motion in the video. A lot of work has been done by many researchers inthis regard. But due to the complexity of the real scene, such as clutteredbackground, a wide range of occlusion as well as the goal itself apparent changeswhich involves object pose and scale changes, these works are able to solvespecific problems in the specific scenario, but difficult to achieve the desiredresults in practical applications. Therefore, in order to get a robust trackingalgorithm, there is a lot of work to do.In this paper, we mainly focus on how to obtain the discriminativeinformation between the target and background to conduct apparent modeling forthe target and how to use the information effectively in the process of tracking. Inthis article, we adopted the method of multi-color space to obtain thediscriminative information between target and background, and then learn theinformation by AdaBoost learning method. Finally, the information was used inMean Shift method and Particle Filter. The main work of this paper is as follows:Firstly, we propose a new method to calculate the weight of the point in theMean Shift framework. In our algorithm, we use the strong classifier that islearned by AdaBoost method to classify each point in the target area. And each point is assigned a weight, then applied to the Mean Shift method to track themoving target.Secondly, we proposed an algorithm to calculate the particle weight underthe framework of Partice Filter. In our algorithm, the particle weight calculationis divided into two parts: one part based on the overall information of the image,the other based on local point information. The we will calculate the weightedsum of particle weights of these two parts to get the final particle weight, whichwill be used in the particle filter method to track the object.