DSST Target Detection And Tracking And System Implementation Of High Confidence Strategy

With the multimedia and informationization of power systems,a large number of substations and transmission lines have adopted remote monitoring systems and collected a large number of video images.At present,the detection and tracking thereof is basically done manually,which may lead to false detection and missed detection,and miss the best processing time of the event,so in the scene,firstly,it is necessary to automatically track the target in video without manual intervention,extract key useful information from the video source;secondly,it can be able to track the target quickly;finally,the substation site environment is complex,it is likely to lose the target in the process of tracking,and need to be as accurate as possible.In order to solve the above problems,the correlation filtering-based tracking method has unparalleled advantages.Therefore,this paper firstly selects suitable target filtering method for this background,then improves the existing problems of the tracking method,tests the algorithm in different scenarios,analyzes the tracking performance of the algorithm,and finally realizes an intelligent video monitoring system to facilitate staff operating.The main research contents of this paper are as follows:1)This paper proposes a DSST tracking algorithm with high confidence.In order to solve the problem that the DSST tracking algorithm needs to manually mark the initial frame position,real-time speed and update strategy,this algorithm combines the three-frame difference target detection algorithm,the non-linear scaling algorithm based on Sobel operator and uniform interpolation on the basis of the original algorithm,and use the PCA method to reduce the dimensionality of the Hog features and color histogram features into a joint feature vector.At the same time,in the model updating strategy,the values F_maxax of the highest point in the response graph and APCE（average peak-to correlation energy）indicator are introduced.In order to verify the effectiveness of this algorithm,50 groups of video images in the substation data set were used to test the algorithm.Through the analysis of the experimental results,the algorithm further improved the tracking accuracy and tracking speed,and was suitable for target tracking in substation scenarios.2)The general verification of this algorithm.In order to further prove the robustness of this algorithm’s other scenarios,this paper tests on open data sets OTB and VOT.Firstly,the experimental results on the OTB data set show that the algorithm is slightly better than other advanced algorithms in overall performance evaluation.It is slightly inferior to the Staple algorithm in the two attributes of low pixel and deformation,but it is the best in the other nine attributes;Secondly,the experimental results on the VOT dataset show that the tracking performance of this algorithm is slightly consistent with that of the Staple algorithm,except that the occlusion feature is slightly better than the Staple algorithm,and the remaining five features are slightly lower than the Staple algorithm.Through a series of experimental comparisons and analysis,the algorithm of this paper shows strong robustness and tracking efficiency in many scenarios.3)an electric intelligent video motion detection and tracking system.In order to facilitate the video detection and tracking operation of substation workers,an electric intelligent video motion detection and tracking system was completed based on the algorithm of this paper.The system consists of three modules,which are algorithm testing,batch detection and real-time monitoring.The algorithm testing module compares the advantages and disadvantages of the algorithm and the original algorithm in a graphical way.The batch detection module selects multiple substation monitoring video detections at a time,and then displays the tracking information by strip,and stores it.The real-time monitoring module is a real time preview tracking the moving target under the camera,and displays the abnormal situation through the log file,in addition,the feature is integrated to display some extra functions.