A Parallel Image Mosaic Method Based On Feature Matching

Currently, Google's street view is the most representative way of the city street visual technology. This method uses spherical panoramic images to provide fixed-point 360°view of the scene here and produce many pieces of panoramic images down the street. Using the jump of panoramic images achieve the purpose of roaming. The method subsists a larger defect, namely the transition between panoramic images difficult to obtain smooth results, user's immersion is not enough, and there is "ghosting". This paper presents a measure of image stitching technology which achieves city street visualization: Parallel Image Mosaic. This method uses the camera shot both sides of blocks along the street scene, and then apply the image mosaic technology end to end to collect of images together into the shape of the wide field image. The approach supports the user in the horizontal direction of any move, to enhance people's sense of immersion, and solve the "ghosting" problem effectively. This paper presents a method about block image capture, view selection, feature matching and image stitching, specifically the following:(1) Design a convenient and practical image acquisition system which uses to collect the city street.(2) Select key images from a great deal of images which use to mosaic city street. According to the particularity of the city street, and in order to meet the needs stitching, after elimination of perspective distortion and to facilitate the work of three-dimensional rendering, the paper develops a view of the principle of selection, and uses depth estimation and plane detection algorithm classified as critical view of options available basis, and then through our selection strategy for selected the mosaic view.(3) The paper extract the image's feature by using the method of limited the region from the adjacent images ,then process feature matching. Improve the operation speed of feature extraction and reduce the false feature matching, so obtain accurate transformation matrix between the images.(4) Next the image's fusion is necessary. This paper combines Poisson image fusion method and optimal seam technology, which solve the crack when the images fuse effectively, "ghosting" and the phenomenon of exposure.