Intelligent Video Surveillance And Anomaly Detection Based On Wireless Network

This thesis proposes a novel solution for the Intelligent Wireless Visual MonitoringSystem. Wireless Mesh network will be the carrier to translate video data. In order tohave better noise suppression and integral goals after anomalies goals appear, I willmodel background with the improved adaptive mixture Gaussian background modelalgorithm. Therefore, I can extract all moving targets. To avoid the target adhesionsproblems among motion detections, Mean-shift tracking algorithm will work ontracking the moving targets. Besides, multiple sub-blocks of gray matching method maysolve or mitigate the problem of goal losing and messy which happens during tracking.At the same time, reliability of following research and applications would be ensured.The local image which contains anomalies frame will be transported via TCP protocol.Therefore，manager of monitoring client could find out the timestamp when abnormityappeared in history video rapidly. And then, recent video could be stored into SD card.Finally, we could transport history video to monitoring server via FTP.The key step in target recognition is to isolate suitable moving targets for videosurveillance systems. While maintaining the detection accuracy, corresponding adaptivecapacity is necessary for the continuous variation of the light during target separation.We address a novel implementation of the improved adaptive mixture Gaussianbackground model algorithm. Based on this implementation, we separate abnormaltargets easily with analysis/determination and identification follows.We optimized the adaptive mixture Gaussian background model in the system so asto adapt to the background construction and prospects control. Algorithms ofbackground construction and prospects control are described as follows:Firstly, build a static background image and initialize the background model withvideo sequence which contains moving objects in the scene and a static backgroundimage. Secondly, The adjustment on prospects ablating time brings in prospects ablationtime control mechanism and independent learning efficiency model. And then, byjudging results of foreground and background, we can identify whether there areabnormal moving targets in video frames. Video frames which contain abnormal moving targets will be saved to local folder as images. Experiments show satisfyingresults with good robust and accuracy of this algorithm.We tracked algorithm to track the moving targets which have been separated fromtheir original videos. Whenever there is real-time tracked goal matches anomalycharacteristics in anomaly behavior library, we will try to find it and give an analysis onit. Meanwhile, the monitor server bell rings. Finally, we implemented the whole subjectof software system and the subject passed inspection.