Stitching algorithm of large parallax images in underground roadways

In order to solve problems of high cost of fixed -point rotating cameras for video surveillance in underground roadways as well as limited visual field and image stitching effects, a camera arrangement is designed for large parallax shooting. For difficult image stitching caused by large parallax, a stitching algorithm of large parallax images in underground roadways is proposed which is based on multi -plane and multi -perception seam. Firstly, scale -invariant feature transform algorithm is used to detect and match feature points of the input images to obtain feature matching points. Secondly, feature matching points are grouped based on multi -plane, and corresponding alignment candidate homography matrixes are generated. Each homography matrix aligns to one plane in the image, so as to solve plane inconsistency problem in large parallax scene of roadway. Finally, multi -perception seams based on color, edge and saliency are calculated in each group of local aligned images, and the aligned image with the minimum stitch energy is selected to synthesize the stitched image, so as to reduce local area dislocation. The experimental results show that compared with classic image stitching algorithms including APAP, ANAP, SPHP, NISwGSP and RobustELA and color -based stitching algorithm, the proposed algorithm effectively eliminates local area dislocation and ghosting, and achieves more natural and seamless stitching image.