Target Tracking Research And System Design Of Correlation Filtering And Machine Learning Fusion

With the rapid development of society,peoples life is more and more closely connected with computer vision.As a result,target tracking has attracted great attention in the fields of intelligent driving,robot and monitoring system,and the application of target tracking in various systems is becoming more and more extensive.At present,the target tracking algorithm still has many problems,such as the low success rate of the target tracking in the complex background,fast movement and other situations.In recent years,with the rapid development of machine learning-related technologies,the target tracking algorithm based on machine learning has also made rapid progress.However,it requires a large number of standard data of various scene pictures to train the convolution parameters,which involves a very large amount of computation.Although you can use the GPU to speed up the computation,but its required amount calculation task does not reduce,the weight of training in the use of the file to execute the tracking task,still need a convolution,makes the overall execution efficiency of the algorithm is not high,in the actual scene still cannot reach the effect of real-time processing,there is a barrier for practical application.The target tracking algorithm based on correlation filtering has a great advantage in processing speed,and the tracking effect is very good in a short time.However,as the attitude of the tracking target and the environment are constantly changing,the model of the correlation filter is very easy to be polluted and lead to tracking failure.This paper first analyzes the principle of correlation filter algorithm,this paper introduces the work process of feature extraction module,the gray features,shape features and edge detail contrast,according to their respective characteristics and combined with the feature of the expression of blind area,edge feature fusion of multiple features fusion algorithm is proposed,in terms of feature extraction to improve the accuracy of target tracking algorithm.Compared with the shape feature,the edge feature can describe the target in a more detailed way,and the edge feature,HOG feature and gray feature are fused together to improve the accuracy and coincidence degree of target tracking,and the tracking effect is significantly improved.Although the improved feature extraction module improves the stability of target tracking,it does not solve the problem that the model is easy to be polluted in the long time target tracking task,which may lead to tracking drift and model pollution.Therefore,introduces the machine learning module,the target detection based on machine learning algorithm design for relevant filter target tracking corrector,relying on the related filter in a short time,good tracking performance and machine learning good precision of target detection,make up for each other,according to the result of machine learning target detection at the same time,provide more reliable candidate box,The learning rate and tracking scale of the correlation filter module are modified,so as to realize the automatic adaptation of learning rate and scale information,and make the target features extracted by the correlation filter module more accurate,so as to ensure the accuracy of the target tracking algorithm.The performance test on the OTB open data set has effectively improved the target tracking accuracy.The improved target tracking algorithm based on the fusion of correlation filtering and machine learning is applied to a specific scene and a target tracking system based on the fusion of correlation filtering and machine learning is designed.According to this article to do related work,in the field of target tracking based on B/S software architecture was designed and implemented in this system,and through the user service layer,algorithm model and application support layer and hardware base layer four aspects to analyze system,made detailed instructions to its function hierarchy lists the system structure diagram,the process of the system and user interface display and description,It has certain creative significance and research value.