Design Of A Cooperative Dual-cameras System For Object Detection And Eagle Eye Surveillance

The main aim of modern video surveillance is to capture clear imagery of ROI(region of interesting) in vast range scene. However, it is hard to achieve the goal by zooming when the ROI is far from the surveillance system. A station-motion cameras system is designed to acquire both the detail and overall imagery of objects at distance. Firstly, improve moving objects detection using Codebook algorithm. Secondly, analyze the physic model of dual-camera calibration to prove feasibility of look-up-table method. Thirdly, propose a dual-camera coordination control algorithm based on the motion state of pedestrian. Finally, devise the system and realized the function. This paper focuses on the following:1) Improve the codebook algorithm.In order to accurately detect the moving object, two methods are added. One is finding the errors of the Codebook result then fixes it. Another is introducing the neighborhood pixel and updates the Codebook model. Then Kalman filtering is used for tracking and marking the moving object. 2) Analyze the physics model of dual-camera calibration to prove feasibility of (LUT) look-up-table method. Firstly, build the physic model of dual-camera coordination, and deduce the process of the dual-camera calibration. Secondly, summarize the advantages of the method: flexible and accurate; disadvantages: complex and impracticable. Finally, prove the feasibility of LUT based on the model of dual-camera.3) Propose a dual-camera coordination control algorithm base on the motion state of pedestrian.Present a dual-camera coordinating method based on the real world object motion states. Control the PTZ(pan-tilt-zoom) camera to track when the present object is moving. Once the object stops, steer the PZT camera focus on and zoom in to acquire clearly imagery.While the moving object is pedestrian, execute a face detection procedure using the images received from PTZ camera. After the face is detected, PTZ camera puts it in the center of the FOV(field of view) and zooms to acquire high-definition imagery. When the pedestrian continues moving, PTZ camera receives the location of pedestrian from the station camera and tracks it continually. In addition, a procedure is used for finding errors in the face detection and dealing with these faults to improve the system robustness.4) Devise the dual-camera system. Select hardware and software development kits according to the demands of the system. Develop dual-camera system based on VS2008 and OpenCV, and speed up the algorithm to detect the moving objects. Experiments on real indoor and outdoor environments show that the proposed wide FOV camera objects detection algorithm is robust to illumination variations, the PTZ camera can track and zoom in the face of pedestrians, and the real-time performance of the method is good.