Research On Correlation Filtering Target Tracking Algorithm

Video target tracking has attracted much attention in the field of machine vision.Researchers have proposed many advanced target tracking techniques through years of research,but in the actual tracking process,target scale changes,motion blur,fast motion and occlusion are still studied.Difficulties and hot spots.Therefore,it is still very challenging to study the target tracking algorithm with high robustness in various application scenarios.This paper mainly studies the problems of poor tracking accuracy due to changes in target scale and how to use context information more effectively to improve tracking accuracy.The main research contents are as follows:In view of the problem of poor tracking accuracy due to target scale changes during the tracking process,this paper adds a scale estimation strategy based on the color feature tracking algorithm,respectively training the two filters of position and scale,and improves the update mechanism to achieve Adapt to color feature scale tracking.Firstly,the algorithm obtains the position-dependent filter through the least squares classifier,calculates the maximum value of the response as the target center position of the next frame,and then forms a plurality of rectangular regions of different sizes around the target center position according to the set scale factor.Calculate and learn the color features of each region to obtain the scale correlation filter;then determine the size of the tracking target according to the maximum value of the response;finally update the position and size of the target.By testing the video sequence of the data set OTB100,the results show that the proposed algorithm has better performance than the improved color feature tracking algorithm in the video with obvious target scale change.For the context-aware algorithm to treat the context information with the same proportional coefficient,the maximum value of the context information can not be exerted,and the weighting processing strategy is added,so that the context information can be utilized more effectively,the tracking precision is improved,and single channel and multi-channel are given.Solutions in the original domain and solutions in the dual domain.The algorithm mainlycalculates the weight matrix according to the similarity between the different regions of the context and the tracking target,and assigns different weights to the context region.The background information around the target is combined with the target features to jointly train the classifier.After a large number of experimental tests on the data set OTB100,the results show that the proposed algorithm is superior to the context-aware algorithm before the improvement in the target scale change,motion blur,fast motion and occlusion.