Research Of Specilfic Human Abnormal Activity Recognition In Power System

There are several safety hazards in power system, the majority of the safety incidents occurr in the electricity production, some events have sudden and unpredictable nature, but personal injuries caused by man-made causes are avoidable, traditionally, video surveillance can be used to reduce the economic losses and injuries; the traditional video surveillance system is limited by intrinsic factors, the monitoring staff are easily fatigued, the real-time performance is poor, reducing the safety and practicality of the whole system, intelligent video monitoring system enables computer to analyze and understand the content of the video, to make up for the shortcomings of traditional monitoring system, the degree of injury can be minimized. The aim of this study is the detection and identification of human behavior based on the intelligent video monitoring system.In terms of human behavior detection, firstly, the paper studied the human body moving target detection method and analyzed the advantages and disadvantages of the basic method, the conventional Gaussian background model is done by weighting average of the current frame and the current background frame of the video sequence to update the background, however, it is ineffective due to the light mutations and the impact of other external environment, the paper finally proposed the target detection method based on the combination of adaptive multi-dimensional mixed Gaussian background model and frame difference, using the improved OTSU method to select the threshold, experimental results showed that the method had a good effect.In terms of human behavior identification, the paper firstly introduced the basic method of the motion feature extraction and behavior recognition, it proposed a method of human behavior identification based on improved (?) transform and the Discriminative Random Fields; improved (?) transform makes up the (?) transform’s shortcomings which the two-dimensional planar function need to be scale standardization when extracting the features, enhancing the features’robustness. The Discriminative Random Fields combine the characteristics of the traditional Conditional Random Fields model and Hidden Conditional Random Fields model, improving the Hidden Conditional Random Fields’performance, which can only deal with sequence data, this model gets the behavior’s internal dynamic characteristics and the external dynamic contacts. By comparing the experimental results, as can be seen that this method had better classification results, and could also better classify human behavior.