Research On The Key Technologies Of Intelligent Video Surveillance System

Intelligent video surveillance system is one of new arising high-tech application fields.Compared to traditional video surveillance system,it has advantages of higher quality and less need of investment.It has cheerful prospect in the applications of surveillance for traffic,bank, hotel,shopping,etc.The dissertation aims to design intelligent video surveillance system with static cameras.Based on studying the current conclusion,the dissertation improved and realized the key technologies which relate to detection and recognition of moving object.The major contents of this dissertation can be summarized as followings:(1)In the dissertation,the method based on background model is adopted to detect moving object,and the algorithm is improved based on the traditional Gaussian mixture model.Firstly,the process of building and initializing background model is simplified,which make the application of the algorithm wider.Secondly,the models of S and V are built to increase the accuracy of object detection.Thirdly,using adjacent-frame difference for reference,the problems of background changes are solved by model rebuilding which is realized with the help of cycle counter and dynamic learning efficiency.(2)To solve the problem that the noise which could impact object feature extraction existing in the result of initial moving object detection, median filter and morphological filter are adopted to remove background noise.To remove the shadow area in foreground,an algorithm of shadow detection which is based on the RGB color model is proposed in the dissertation.(3)To deal with the situation that there could be many objects in a scene,the detection of 8-connected regions is adopted to mark and group the pixels of objects and remove noise area at the same time.To describe object area,an algorithm was proposed in the dissertation,which adopts circumscribed polygon instead of the traditional circumscribed rectangle and circumscribed circle.It can describe the moving object area more accurately.For this reason,the accuracy of object feature extraction is improved.(4)In the process of object recognition,the dissertation proposed that the result of object recognition is decided by the Euclidean distance between feature vectors of different objects,which are composed by invariant moments of object edge information.In this way,the object recognition system is not only advantaged by invariant moments but also made more efficient by a great decrease on the mount of calculation.The results of tests show that the algorithms in this dissertation are effective,which can be adopted in the condition of complex scenes to attain background model rapidly,detect and recognize moving object.