Camera Tampering Detection Based On Corner

Nowadays, video surveillance has got into all fields of our everyday life. The camera are installed everywhere such as the living quarters, the bank automatic teller machine (ATM) and the traffic area. Video surveillance can be effective in preventing all kinds of criminal behaviors. It can also provide information of the traffic flow and the road situation. However, along with the popularity of video surveillance, some suspicious people, who want to avoid being monitored, make some camera tampering, such as a person holding his/her hand in front of the camera, spray painting the lens, or turning the camera so that it points in a different direction. Therefore, the computer automatically and timely recognizes these tampering: Real-Time Detection of Camera Tampering is very important.First, the exact definition of what constitutes camera tampering must be established. For this work, camera tampering is defined as any sustained event which dramatically alters the image seen by the camera. Such an event must be sustained for several seconds in order to be detected. For this purpose we should address three problems: first, the method must work online in real time; second, camera tampering detection should be sensitive to any significant camera motion, and is insensitive to changes in illumination and other non-tampering events. The number of false detection events should be minimized; Third, insensitive to transient changesCamera tampering detection is based on the video scene change detection, that is, to detect the change of the image frame when tampering occurs. We referenced many algorithms of the scene change detection which are based on the pixel domain and the compression domain, and compare algorithms'advantages and disadvantages. As the algorithms based on the pixel domain is better, and algorithms based on compression domain are not real time, so we search algorithm based on the pixel domain. We build a function to describe image, and then get the differentials.As corner minimize the amount of data, and is immune to rotation, insensitive to illumination changes, so select the corner as the description of the image. So we can get the shape and location information of the image, at the same time can be insensitive to illumination changes.A function describing the image characteristic is built based on the SUSAN corner, which is used to describe the image. When the scene changes, the function will be significant changes, then we can detect the change. SUSAN corner don't need detect the edge algorithm first, is insensitive to noise, can detect all types of corner. At the same time, we use fast SUSAN corner detection algorithm in order to improve the speed of calculation.In order to be effective with the sudden change and slow change we should not just compare two frames. We build two buffers, the first buffer stores new frames, the second buffer stores old frames. This method enlarges the tampering interference, achieves very good results and can be real time.