Design And Implementation Of Real-time Wide View Video Mosaics Based On SIFT

With increasing of modern social awareness on terrorism and security awareness and the need in the military and industry, video surveillance is widely used in residential regions, shopping mall and urban traffic etc. Traditional video surveillance uses many cameras to cover the region being monitored. If the single scene is very large, using one wide view video is more intuitive than multi-cameras. In this paper, we study the video mosaics methods and produce a real-time wide view video mosaics system based on SIFT features.The whole system includes calibration of cameras, acquisition and correction of video sequences, matching and fusion of key frames, rapid integration and display of the remaining frames etc. First, we adjust the two cameras to a suitable location(there is only a relationships of translation or rotation by PTZ). We do the calibration and correction of the cameras by the board method.Next, we select a key frame who is simultaneously from the corrected-videos by the two cameras to preprocess respectively,then we match the two key frames by using registration algorithm. In this paper, we select the SIFT features that have higher accuracy for feature detection and matching. Finally, we synthesize the overlap region of the two video frames by using fusion algorithm to obtain the wide view video frames of the current key frames. In this paper,real-time wide scene synthesis is based on the fixed camera position, the positions of video frames are basically unchanged by the two cameras obtained. Therefore, the positions of each video frame’s overlap region are substantially unchanged, we only need to record the position of the overlapping area during the first matching, then all the following frames only need to fusion together, so the efficiency of is improved greatly, and the real-time requirement of the system is achieved.The experiments of this paper achieve the real-time wide view video mosaics of the two analog cameras.The system is provided with better real-time, meet the needs of monitoring. This algorithm can be extended to the synthesis of three and more video cameras.