Research About Key Technology Of Surveillance Video Mining Based On Motion Feature

Video surveillance has been widely used in many fields, by using video miningtechniques,characteristics can be extracted from huge video data, effectively analysis, foundhigh-level semantic knowledge and mode, and realize the video for monitoring automationand intelligent application. At present, the domestic and foreign research to monitor videomining just underway, some key issues, such as surveillance video data feature extraction,incident detection have not effectively solved.Video mining is found effective, novel, implied, valuable, understandable mode fromvideo concentration, namely knowledge, draw event relevance and trend, to provide userswith problem solving levels of auxiliary capability. The specific technology of video miningcan be divided into two ideas: video structure mining and. Video results, according tostructural characteristics of video content structure and according to certain algorithm rules,video is divided into several levels of structural unit; Then retrieve useful feature of everyhierarchical structure and characteristic of structural unit itself between characteristics; Finally,according to the similarity units of every level or other rules to get video structure of thetectonic pattern get the semantic information which the tectonic model can reflect. videosports mining, Segmentation, tracking motion object from the video, in this process to extractsubstitutive characteristics of the moving object and these features association rule orspace-time relationship between these features to get the feature meaning of moving object,or its behavior tendencies and event model, thus the high-level semantic expression dig videoinformation.This paper based on the scene without content structure video, applied video sportsmining technical ideas of the video monitoring data mining to extract video object from videothinking, tracking its movement, combined with the time characteristics forming a time seriesdata only for which march the abstract, clustering and anomalous detection, realize themonitor video data mining purposes.The main innovative achievements are described as follows.1) Video feature extraction method research. Using the background updates based onframe differential algorithm and based on the movement of the time series of extractionalgorithm express video feature in data sequences ensure video abstract extraction, clusteringand anomalous detection. 2) The video abstract of extraction based on wavelet transform, the proposed algorithmcan form multi-scale video abstract, solves the problems which in different conditions, videoabstract needs the length of different.3) In order to reduce the abnormal situation searched, propose anomalous detectionbased on k-means algorithm. By theoretical analysis and experiment, verify the validity andsuperiority of the proposed algorithm.