Research On Detection And Classification Of Camera Tampering Events

The intelligent video surveillance system has been widely used in various fields of national life, such as traffic monitoring, the robbery incident detection in bank, security monitoring of public places, etc. However, with the increasing demand for video surveillance system, it is also facing a growing number of disturbances events. These disturbances events make the surveillance images appear abnormal, which has affected subsequent flows of video analysis. When the camera is seriously disturbed, the images cannot be collected in the scene, which may result in the loss function of the video surveillance system. And the economy of public places is lost, and even the security of public places of is also threatened. So it is very important to detect the tampering events of the camera.This dissertation makes a detailed analysis and discussion on the problem of camera tampering detection and classification in intelligent video surveillance system, and the main research work is as follows.First, this dissertation proposes a novel background mapping model, which is used to describe the static content of video images in complex environment. The background mapping model is derived from the mapping background of each frame in one video sequence, and this dissertation uses the random updating strategy to update the background mapping model. In this dissertation, the background mapping model and other four background modeling methods including GMM, VIBE, SACON, and PBAS are compared in the scene. The results show that the background mapping model has a relatively high detection rate on camera tampering events, and the rationality and validity of the background mapping model are verified.Second, this dissertation proposes a detection algorithm of camera tampering based on background mapping model of this paper. This detection algorithm uses the mapping background image of background mapping model, and combines with the Canny edge detection to extract the edge features of the current gray image and mapping background image, then builds the edge feature function, and compares it with an adaptive decision threshold to detect whether a camera tampering event appears or not. At last, this dissertation compares with five kinds of camera tampering detection algorithms based on public video datasets and other video datasets which contains 122 video sequences in total of 17 hours. Experimental results demonstrate that this dissertation proposed method has lower time complexity and higher detection rate.Third, this dissertation proposes a classification algorithm of camera tampering based on multi-feature combination. The main function of this algorithm is to implement the classification of three kinds of camera tampering types, such as camera defocus, camera occlusion and camera displacement. Based on the detection algorithm of camera tampering in this dissertation, the camera tampering classification algorithm extracts 12 features such as the image color, edge, texture, corner and transform domain in each tampering frame at first. Next, the C4.5 classification algorithm is used to implement the classification of camera tampering. In six different scenarios, three classifiers including K Nearest Neighbor, Bayesian and Support Vector Machine are used to classify the camera tampering events. Finally, the experimental data shows that the proposed method can achieve 80.22% correct classification rate, and better than the classification results with other classifiers.At last, the dissertation summarizes the work of this paper, and the research and development of camera tampering detection and classification technology in intelligent video surveillance system are prospected.