Robust semi-global matching under different radiometric conditions

This paper presents an improved semi-global matching algorithm which is robust to the different radiometric conditions. Firstly, Census transform instead of mutual information is adopted to calculate the matching cost because of its robustness to radiometric changes. Then, an additional self-adaptive matching penalty is added into the cost aggregation along every accumulation path of semi-global matching to preserve depth continuities. Finally, the winner-take-all strategy is applied to calculate the optimal disparity value and a few post-processing methods are performed to refine the disparity image. Experiments on Middlebury datasets reveal that the proposed method can always perform better than the SGM, GF and ELAS algorithms under different radiometric conditions, and it can also be used in the real-world scene successfully.