Abstract: In view of the geometric local constraints in line matching of close range images, there are some problems, such as missing selection and weak matching of descriptors under certain scene transformation. In this paper, a line hierarchical matching scheme for close range image is proposed. In this scheme, the kernel equation and perspective transformation matrix are obtained with reliable homonymous points, the target line is screened twice, and whether there is a reliable camera projection matrix in close range image is selected as the check method. If there is a camera projection matrix, the corresponding re projection error is calculated, Otherwise, the cosine similarity between the optimized MSLD descriptor vectors is calculated, and their reasonable thresholds are set as the final threshold to accept the line with the same name. The algorithm is analyzed and verified by using the network public data set. The experimental results prove the feasibility of this scheme, and can obtain accurate and stable line matching results of the same name. At the same time, it has good adaptability to a variety of transformed images.

Key words: close range image, hierarchical matching, global constraint, reprojection error, descriptor constraint