A multi-level dynamic programming method for line segment matching in axial motion stereo

This paper propose a multi-level dynamic programming method to solve the line segment-based correspondence problem in axial motion stereo. In this method, a local similarity measure (LSM) is calculated For each line segment pair between the front and back images. A threshold T and certain constraints are used as selecting criteria for choosing potential matching pairs in each level. In Level I, threshold T is set to a relatively high value to ensure the probability of correct match in the first level is very high. In this level, the matching probability between line segments is represented by their local similarity measure. Dynamic programming is then used to search for the best match for those selected potential matching pairs. Matched pairs are used to assist the matching process of the next level. By considering the geometric properties between the matched and the remaining line segments, a global similarity measure (GSM) is calculated for each remaining line segment pair. An overall similarity measure (matching probability) for each remaining line segment pair is then obtained by the LSM and the GSM. The algorithm then proceeds with the second match, but with a slightly lower threshold T-2. New matched results are then used to modify the GSM and the overall similarity measure of the remaining line segment pairs. These processes are repeated until a predefined level n(stop) (or a predefined condition) is reached. By using the GSM and the multi-level searching technique, the proposed technique increases the matching accuracy and the number of matches while reducing the number of unmatched line segment due to misordering when dynamic programming is used for axial motion stereo matching. (C) 1998 Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved.