Summary: | 碩士 === 國立中央大學 === 資訊工程研究所 === 100 === In the application of security surveillance and robot vision, object detection and tracking always play an important role. How to establish a stable intelligence surveillance platform is a final purpose we all want to accomplish. But we usually face the problems of the narrow angle of view and dead space in single camera. So we may use some wide-angle lens, PTZ (pan-tilt-zoom) camera, or even multi-camera system to improve this kind of issue.
In multi-camera surveillance system, we proposed a robust and efficient mechanism to solve the foregoing problem. At first we use the progress background modeling and calibrate the demarcation (overlapped area) between the cameras, and the connected-component to extract the interest object of the foreground, and then put a PSO tracker on that. When this tracking object is about to leave the current camera, we design a cooperative protocol to take over the tracking token and go on tracking the object in another camera. The cooperation between cameras can achieve reliable and robust in continuous tracking. Finally we proposed a NGT measure to evaluate the tracking performance.
Compares to the other multi-camera approaches, most of them prefer to put the cameras toward the overlapped area, in order to get the depth information as a criterion. We tend to get more surveillance view instead of the overlapped area. And with the uncomplicated setup of the camera calibration, we can realize the intelligent surveillance for multi-camera cooperative object tracking.
|