Research On Object Detection And Classification Technology In Intelligence Video Surveillance

In recent years, improving the intelligent level of traditional video monitoring system is the key task of technology upgrading in this field.Classification and behavior analysis technology based on rules of moving object that become one of the important research content of intelligent video monitoring.Because they are part of high-level image semantic analysis, there are still many difficulties that mainly come from influences of the low-level and middle-level image information.This thesis focuses on the classification technology based on geometric features and behavior analysis technology based on the trip-line detection rules of moving object.The main works are summarized as follows:(1) Aiming at the influence of moving object contour to the geometric features extraction of object classification and recognition, a more accuracy method to extract its external contour is designed by this thesis.Firstly,a rough moving object region is obtained by GMM foreground detection.Secondly,a rough external contour region is obtained by a marked watershed method.Finally,an accurate external contour region is obtained by HVS shadow removal method and object reconstruction method.So that the new method extracts the accurate external contour.The experimental results show that it could be get more accurate external contour.(2) In order to improve the average recognition rate of moving object based on geometric features,an improved method is designed.Firstly,about the predefined patterns of classes,according to the object contour of morphological difference,a predefined pattern class is subdivided into sub pattern classes;Secondly,the new method extracts high dimension features in the step of feature extraction.Then it uses pseudo division method to generate object sub feature vectors,they could describe the moving object in many ways;Finally,this step in the classification strategy,it uses the classification results of feature vector group to determine the final classification result.The experimental datum show that this method has excellent effects in four predefined patterns(people,car,bike,small crowd),the average recognition rate is 99.48%.Compared with the traditional methods,it is improved by 1.8% to 4.0%.(3) Aiming at the low intelligent level of the current trip-line methods,a kind of intelligent method in intelligent video surveillance is designed.Firstly,the method usestrajectory information and computational geometry algorithms to complete the single-direction and double-direction trip-line detection.The suspicious object is determined by expanding the definition of trajectory anomaly.Secondly, it uses moving object classification method to get classification result of the illegal object.Finally,it generates an available illegal event in intelligent video surveillance by using many basic information of illegal object and event generation mechanism.The experimental results show that this method has good detection results,the average correct rate of event information is 91.6%.