Research On Target Tracking Method In Dynamic Scene

In the field of Computer Vision,target tracking is an important subject,It is widely used in the fields of Intelligent Surveillance,Military Navigation Guidance,Video Coding,etc.Target tracking is a process in which the positions and states of targets,appearing in the video frame for the first time,are determined automatically in the subsequent video sequence.According to the different video-shooting conditions,target tracking can be divided into two cases static scene and dynamic scene.The research object of this paper is that how to realize moving targets tracking in the dynamic scene.The essence of target tracking is to finish the target's identification and location in video sequence through a reasonable and efficient search mode after choosing a reliable feature description target.Firstly,feature description method based on random ferns is well used,considering special relationship and grey information between pixels,after analyzing of feature extraction algorithm.Hence robustness is well reflected relatively.Also,the feature is improved considering the issues to solve,and the feature extraction algorithm,fitting the issues in this dissertation,is obtained.After analysis of target scaling and posture changing's influence to feature extraction in the process of target tracking,and producing training set through initial target,initial feature distributing model of robust is obtained.Then concerning the complex moving condition like error accumulation and sheltering in target tracking,a guiding strategy based on combination of background compensation is put forward.Firstly,detect the moving border of current frame,then screen and structure the suspected moving regions.Then combine the random ferns feature and Naive Bayes Classifier;identify and classify the match range;then realize the target location of current frame.At last,reasonably update by feature model of tracking results leading target,and moving target tracking is finished.Experimental results show that the method put forward in this paper can adapt deformation of moving target.Also,it can realize exact orientation of target in the case of target occlusion and adhesion.Then moving target tracking is finished.