SGM-Based Disparity Estimation Under Radiometric Variations

Abstract

The semi-global matching (SGM) performances excellent in stereo correspondence field, it reaches a good trade-off between correspondence accuracy and computational complexity. However, the performance of SGM is limited under radiometric variations, such as varying lighting and exposure conditions. In this paper, an improved SGM method is presented to remedy this problem. To eliminate the discrepancy of illumination between the stereo images, both histogram equalization and binary singleton expansion are adopted in pre-processing stage. The weighted median filter is adopted after conventional LRC in the disparity refinement stage to remove the outlier errors and preserve edges. The stereo images from the Middlebury benchmark are used in the experiment. The experimental result show that the average RMSE of the improved SGM method is 12.64% lower than the raw SGM. The proposed method can effectively improve the accuracy of disparity map compared to the SGM algorithm.