A real time dual-camera surveillance system based on tracking-learning-detection algorithm

In this paper, a master-slave dual camera (MSDC) system is designed to monitor the scene and acquire detailed information of distant object at the same time. The selected target, in the wide field of view (FOV) of master camera, is detected and tracked by algorithms of codebook and Tracking-Learning-Detection (TLD), respectively. According to the method of Look Up Table (LUT), the slave camera (pan-tilt-zoom camera, PTZ camera) is directed to gaze at the target. In this way, the high resolution image of the object is available in the PTZ camera. Experiments of indoor and outdoor scenes have been conducted and the application of this dual-camera surveillance system in real time is illustrated. Furthermore, by comparing these two algorithms, it is found that TLD is more robust to illumination variation and the scene of multi motion objects. Besides, this set of system has the potential of being more intelligent through the employment of a video containing certain objects to initialize TLD, thus enabling the track of objects in real scene. Certain attempt in this point has been made in our experiment, which proves successful.