Multi-level based stereo line matching with structural information using dynamic programming

In this paper, dynamic programming is used to solve the correspondence problem in stereo vision. A multi-level matching technique is used so as to improve the accuracy of the matching process between the left and right images. The method first matches those that have a similarity larger than a threshold T₁. In the second match, a lower threshold T₂ is used and all previous matched pairs are used to provide structural information in measuring the similarity. The matched results are then updated and the process is repeated until a predefined level n is reached. The proposed method uses the multi-level matching technique so as to reduce the errors of missed matches due to imperfect feature extraction such as missing lines and broken lines. The algorithm has been tested on real scenes to confirm the usefulness of the proposed method