| Multi-channel video real-time mosaicing technology is a hot research topic in the field of computer vision, virtual reality, etc. Multi-channel video real-time mosaicing has widely used in such areas as traffic monitoring, video conference, home entertainment and so on. Currently we can obtain panoramic video streaming by wide-angle lens or fisheye lens, but these devices are expensive, and the existence of large distortion in the video images which obtained by these devices. Real-time video mosaicing technique is the essence of image mosaic, but it is different from image mosaic, there are more strict real-time requirements. In this paper, by studying the panoramic video realtime mosaicing technology, multi-channel real-time video acquisition and synchronization, image preprocessing, feature extraction and matching, image color correction, image mosaicking involved in image mosaic is introduced in detail and specific algorithm implementation is given. Finally a four-channel streaming video real-time mosaics system is designed and implemented in C++ language.The main content if this research are as follows:(1) During the video capture stage, through the operating system producers- consumer model and multi-thread mechanism, we have realized the collection and synchronization of real-time video streaming.(2) During the image registration stage, through classical SIFT feature extraction algorithm, combined with a certain video acquisition cases of specific situations, to find characteristics in specific areas, and then made use of the KD tree index with priority mechanism, improved the speed of feature extraction and matching.(3) During the image mosaic stage, first, we have used the ideal of optimization to correct the color difference and make the overall agreement between adjacent image in color brightness; Secondly, we have constructed a measuring error matrix, which is used for making the optical seam line as much as possible across the smooth area in the image; Finally we have combined Gaussian- Laplacian pyramid mosaic, to ensure the perfect stitching effect.(4) We have designed and implemented the multi-channel real-time video mosaicing system which is based on the image feature point. Through collecting four road network camera video streaming, in the process of video mosaicing, combined the respective advantages of CPU and GPU, used CUDA parallel computing architecture, builded collaborative programming model, to mosaic the video streams in real time, at the same time, we have made a group of test cases to statistic data and analyze the efficiency of the mosaic algorithm, The acceleration ratio of CUDA improved version arrives at 10:1. It shows that real-time video mosaics system is effective and feasible. |
PDF Full Text Request