Research Of Object Tracking Based On Mean Shift Algorithm And Local Features

At present,the intelligent robot monitoring system can be seen in all walks of life.It has been widely used in factory automation,security and military security.Both of these technologies have scientific research value ands practical application with great prospect,and research on tracking technology is the basis of these,so the target tracking algorithm has became the focus of research of scholars at home and abroad.This paper will mainly introduce the scene for the presence of specific target in complex background.For the current main target detection and tracking algorithm,this paper firstly conduct the research to the classical algorithm.its advantages and disadvantages are analyzed,finding the deficiencies of the classical tracking algorithm and optimized.then,we analysis the theory with simulation experiment.Experiments demonstrate the improved algorithm performance in all aspects of moving target tracking.This paper uses local features to represent the tracking target tracking,meaning to achieve the classic shift algorithm.As the representative of the local feature,shape context can be very good to describe image features within the region tracking,Algorithm in the first frame to manually select target tracking,extract the target contour information and features,depending on the location and distance between sampling points established Shape Context histogram,the histogram of all points last Shape Context constitutes image Shape Context feature,finally tracking target position based on Mean Shift algorithm.Based on Mean Shift tracking algorithm Shape Context feature,based on the color can be a good solution to achieve Mean Shift tracking feature can not properly track is blocked,lighting and other effects can not be achieved tracking problem,which is because the use of Shape Context feature is a partial contour information of the sample point,it can solve the problem of partial shelter,and Shape Context feature is obtained by extracting the contour of sampling points,so for the illumination effect is negligible.Finally,after four simulation experiments prove tracking algorithm of this paper in real-time,color and shade,and the presence of morphologically similar interference in the case is still able to achieve a good track.Finally,the paper target based on Mean Shift Shape Context feature improved trackingalgorithm.This algorithm is introduced template update strategy.Since we need to detect the target,the target being tracked based mainly,background color information rarely,we pass the color histogram of the target area,the occurrence probability calculated for each pixel,the low probability of the point instead of using an average value of surrounding pixels,reducing background on the extraction of contour.Finally,after three sets of simulation experiments prove that the improved tracking system for the occurrence of a specific target for this target large deformation,complex background and moving backgrounds and so the effect is obvious.Based on the target detection and tracking system,the improved algorithm can be studied in the field of intelligent video surveillance provide theoretical support and reference.