Research&Application Of Intelligent Video Analysis Algorithm In Substation

In recent years, video surveillance system has been used more and more in substations. Along with the promotion of unmanned substation, the traditional video surveillance can not satisfy the need of unmanned operation and intelligence any more. Intelligent video surveillance has become the main content in the field of computer vision. The intelligent video surveillance, which is cored with moving target detection and tracking algorithm, can reduce human consumption and improve the real-time and accuracy of the video processing effectively. Therefore, the technology of intelligent video analysis is the necessary trend of the improvement of substations video surveillance and the study of the intelligent video analysis algorithm is of great significance.This dissertation studies the application in substations of two intelligent video analysis algorithms, such as target detection and tracking algorithm, and gives solutions to several classical problems. The following are the main research works:(1) Firstly, three main target detection algorithms (such as the optical flow method, the background difference method and the inter-fram difference) and two target tracking algorithms (such as the mean-shift and the cam-shift) are analyzed. Their principles and advantages are also introduced.(2) Secondly, the applied scopes of intelligent video analysis algorithms are studied. According to monitor the wearing of safe helmet of the workers in substations, this dissertation introduces a detection algorithm which is based on color space and centroid position. And the detective effect is validated by experiment.(3) According to the switch states and goods loss, which are not in high need of moving information detecting and tracking, the traditional background difference method is used to test and to analyze. For the situations which are in high need of moving information, such as person moving and foreign invasion, a detection algorithm combined with inter-fram difference and background difference method is presented in this dissertation because of the disadvantages of the normal target detection, such as the incomplete moving target space and the sensitive to environment. This algorithm is background modeled by median filtering method. After that, the effect of this method, adjacent frame differential method and three differential methods are compared by experiment.(4) On the research of object tracking, Kalman filter is used to forecast possible location of the target in the next frame, to determine the scope of search and to track target effectively combined with cam-shift algorithm. This dissertation aims to reduce the research calculation, and then to forecast and to track target’s trajectory quickly and reliably via optimizing algorithm.