Research On Methods Of Generating Stereo Panorama Of Geographical Scene Based On Video Sequence

With the convenience of obtaining video sequence, there are a lot of researches on panorama and stereo panorama based on video sequence. A panorama is a single wide-angle image of the environment around the camera. Panoramic images are valuable tools to display the real 3-D world. The maps are more receptive in the form of panoramic landscape.Panoramic images are quite useful in evaluating location and characteristics of the neighborhood surrounding them. Panoramas have been widely used in mobile navigation,digital city, panoramic map etc. With the development of Volunteered Geographic Information,geographic information data sharing service mode for "everyone is sensor" and "all for one, one for all" has been rapidly developed. Constructing a panoramic representation and stereo panoramic visual model by stitching image sequences captured by ordinary cameras has become a hot issue. Therefore, this paper studies panoramic mosaic and stereo panoramic construction methods based on video sequence taken by ordinary users. The main contributions of this paper are as follows:(1) Image feature detection is analyzed, classified and summarized in this paper. Aiming at the problems of FAST (Features from Accelerated Segment Test) algorithm such as threshold setting and order selection of continuous points on circle, adaptive threshold compass detection (ATCD) based on video sequence is proposed. On the basis of ATCD algorithm,feature point filtering algorithm based on local entropy is proposed. Feature points were filtered by local entropy and some with larger entropy were kept as the matching points.The parallel algorithm based feature filtering was designed which could greatly reduce the number of feature points and speed up the matching efficiency. Finally, the performances of algorithms were analyzed through contrast tests.(2) As for panoramic images stitching based on video sequence, image matching is the basis of image registration. Matching results directly determine the precision of image registration and influence on panoramic image stitching result. In order to solve the problems of high computational complexity and uncertainty of point-pairs selection in present image matching algorithm, a variable circular domain compasses matching algorithm was proposed. Finally, the experimental results show that the proposed algorithm is faster and has high accuracy and well-stability.(3) Global search strategy is used for image matching, which makes feature points in non-overlapping images still need to be searched. Therefore, local search strategy based on grid was proposed, which can avoid the problems of global search strategy. Local matching search can greatly speed up the search efficiency between images and be applyed to ordinary images with small overlap area. Finally, matching accuracy and efficiency under the local matching search was verified by contrast experiments.(4) The numerous characteristics of video sequence and high overlap between images cause the inefficiency of panorama stitched. The reasons of low efficiency of stitching video panorama were analyzed,including a large number of feature points, global search strategy, et al. Therefore, Parallel video panoramic stitching method based image blocking was proposed. Parallel algorithms were built from two aspects: feature detection parallel algorithm and image matching parallel algorithm. Finally,the global rotation matrix was constructed to avoid panorama twisted or tilted. Experimental results show that parallel panoramic stitching algorithm has a good effect on panorama and is suitable for generating panorama with video sequence captured by ordinary cameras.(5) Stereo panorama is constructed by video sequence. Aim at the deficiency of existing method of generating stereo panorama, circular projections was used to extract left and right strips based on video sequence. Pasting together left strips taken from each image approximated the left panoramic image cylinder. Pasting together right strips in the same way. In order to get a stereo panoramic model, left panorama and right panorama were synthetized to generate red-cyan 3D stereo panorama.