| Title |
Iterative Improvement of Stereo Disparity Image through Minimum Cost Voting of SGM Aggregation Matching Cost Volume |
| Authors |
정원제(Won-Je Jeong) ; 이민재(Minjae Lee) ; 박순용(Soon-Yong Park) |
| DOI |
https://doi.org/10.5573/ieie.2021.58.7.63 |
| Keywords |
Stereo matching; Semi-global matching; Disparity |
| Abstract |
In the stereo matching, cost aggregation improves disparity images using surrounding cost value. The SGM(Semi global Matching)method[2], one of the traditional cost aggregation steps, improves disparity by calculating a minimum value according to an aggregation direction of surrounding costs from a cost volume generated by initial cost computation. In this paper, we propose a method to improve the disparity image by calculating the probability of the aggregation matching cost of SGM. Our method generates a probability volume by using the minimum value selection in each aggregated volume generated by the aggregation direction of the SGM. Next, a new penalty is created as a probability volume, added to the cost volume, and aggregated again to calculate the probability volume. The proposed method improves disparity by iterative calculating the calculated penalty and correcting the cost volume. Next, the probability volume is recalculated by adding a new penalty calculated as the probability volume to the SGM. Then, the cost volume is updated by iteratively calculating the probability volume. As a post-processing step, we improve the disparity by performing a left-right consistency check, hole filling, and weight median filtering[6]. As a result of evaluating the Middlebury data set method, it confirmed that the proposed method is superior to the SGM. |