Research On Video Sequence Stitching In Intelligent Monitoring

With the rapid development of computer application technology, smart surveillance gradually becomes one of the issues which people are very concerned about. In order to break through the limited vision of the existing monitor equipments, video mosaic has become a hotspot in this field. But it's still a difficult problem because of the complexity of scenes and content in the video. With the video captured by pan-tilt vidicon as the research object, the thesis achieves panorama mosaic of video sequence.The main content of the thesis is divided into three parts:global motion parameters estimation, moving object segmentation and image sequence mosaic. Firstly, phase correlation is used to obtain global motion parameters of pan-tilt vidicon, and then attain the key frames and the overlapped region between two adjacent images. Secondly, in order to remove the interference of moving objects which may exists in video sequence, this thesis uses a background subtraction algorithm based on Gauss Mixture Models in background modeling, and then locates the target region by background subtraction, finally accurate moving object region can be achieved through extracting the edge by Canny operator and other operations in target region. Thirdly, the feature promoter region can be obtained through the statistical approach to edge information to reduce the scope of corner extraction. On this basis, it realizes extracting corner by Harris operator in feature promoter region, matching them accurately, establishing the transformation matrix, and then considering the information of moving object region. Finally it achieve the seamless image through the eases in/out fusion algorithm.In the actual research course, the thesis proposes an updating method on images which has been mosaicked to reduce the total errors effectively in panorama mosaic and image blurring, and uses methods for updating moving objects and tracking moving objects to get panoramas of moving objects and tracks of moving objects.The thesis tests video sequence with pan-tilt vidicon, verifies the video mosaic algorithm can mosaic the image sequence with moving object rapidly and has a good real-time ability.