Technology Research On Intelligent Video Surveillance In Unmanned Substation

With the improvement of power grid dispatching automation, the operation of unmanned substation becomes the developing trend in current power system. Remote video monitoring is used on the basis of the traditional "four-remotion"(remote measurement, remote communication, remote control, remote adjustment). Remote video monitoring can monitor substation site and equipment at the real-time and can make the power grid operation more safe and reliable. So the remote video monitoring system is very necessary supplementary for unmanned substation.Although the remote video monitoring system has monitoring and alarm function in substation, it mainly focus on sensor alarm analysis. Video detection is mostly assistance method. With the increasing of monitoring points, the lack of traditional video surveillance revealed:extremely rely on artificial mode to find abnormality, and lack of active analysis for abnormal video data. Intelligent video surveillance technology can overcome these shortcomings. It adds automatic video analysis technology in the current video surveillance system. It can automatically analyze image sequence recorded by camera to locate, recognize and track the objects in the dynamic scene, and further analyze and judge object behavior to give real-time alarm on suspicious behavior, which almost needs no human intervention.In order to meet the need of the fire alarm and theft protection in substation video surveillance, in this paper intelligent video surveillance technology is adopted for moving object detection, recognition and tracking. Main work consists of three parts. The first part is moving object detection:Gaussian mixture background model is chose to detect moving objects and objects involve people, animals, ordinary flames (red and yellow flames), white flames and interferences (incandescent lamps). The second part is feature extraction:the selected features include Hu moments, flames color characteristics, flames color brightness criterion, circularity and other features for multiple objects recognition of people, animals and flames, and so on. The third part is moving object recognition:hierarchical classifier structure was generated from the confusion matrix. Support vector machine (SVM) was used as the basic binary classifier. AdaBoost algorithm applies weighted voting on SVM whose classification accuracy was not ideal in order to increase the classification accuracy. On the basis of object recognition, Mean-shift algorithm is used to track object that is recognized as people and describe its running track.Simulation experiment using multi-video data were implemented. Experimental results show that the design method can get a better recognition for people, animals and flame and eliminate the impact of interferences such as incandescent lamps. When other object-"car"(take car for example in this paper) is tested and background illumination change, some situation emerges and the corresponding solution is determined. The implementation issue of intelligent video surveillance system is discussed. It can provide the necessary conditions for the realization of unmanned substation.