Research On Several Problems Of Fast Moving Target Tracking In Video Image

The field of computer vision reseaxch involves many disciplines,and the tracking of moving objects is one of the important research directions.In.recent years,with the development of computer image processing,pattern recognition,artificial intelligence and related disciplines,the target tracking technology has been greatly developed.At present,the research in this field mainly concentrated in two directions;one is tracking a based on priori knowledge of the target,based on target motion information to obtain the initial on the track;the second is not dependent on prior knowledge of the target tracking technology research,according to the recognition in the video and detect in-teresting targets tracking and related work.In recent years,the research in these two areas has made great progress,but there are still a lot of problems need to be solved in the process of target tracking.These problems not only include target representa-tion,the external interference,the robustness of the algorithm,but also related to the application level of the general problem.This article is interested in the prior knowl-edge of target tracking problems based on the fast moving target tracking research,put forward four methods of fast target tracking,Moving object model representation,target search mechanism,target matching method and the robustness of the algorithm research,through the comparative analysis of experimental results,study shows they are robust to tracking of moving targets.Research efforts may focus on the following aspects:(1)The key of template based target tracking is the template generation,template matching and updating,a single template can not be very good to express the infor?mation of the target,in the tracking process is likely to lead to the failure of tracking.Multi template can better express the feature information of the target,but there are some difficulties in the problem of target matching and template updating.The pro-posed method generates multiple target template by affine transformation,this paper proposes an improved similarity method,considering the effect of different templates in the past frame in the template update,calculated to achieve multi template update different weights,to realize the target tracking under Kalman filter framework.(2)Feature extraction and target tracking is a key feature of the target model is established based on target tracking,tracking methods commonly used to extract the target color,texture,gray and other information to realize the goal of fusion of multiple features multi feature fusion method of target tracking by extracting,establish target tracking model,existing problems is to model more complex,with the increase in the number of model features,resulting in computational complexity significantly increased.The proposed method of target model was established by extraction of color,edge,texture feature,motion feature information adding moving target,improved the matching method of target model,target tracking in mean under shift framework.(3)Local Binary Pattern is the classical operator texture recognition,texture recog?nition has been widely used in target,texture recognition operator extended through the rotation invariance of time and space,to enhance the operator recognition accu-racy,dynamic texture texture recognition standard for movement in the time domain and space domain makes innovative research.The proposed method combined with the characteristics of rotation invariance and dynamic texture,considering the needs of fast moving target tracking,the rotation invariant operator in space and time is extended,the texture feature model of target discrimination,using D-S evidence theory to realize target tracking.(4)Visual salience originate from human physiological characteristics,the objective of visual interest feature extraction to establish a model of visual saliency.The proposed method is used to extract the saliency map of the visual static color saliency feature and the target motion saliency feature,and to complete the target tracking by using the particle filter.