An Improved Algorithm for Dense Disparity Estimation Using Two--Level Dynamic Programming Approach

• Main Authors: Min-Hsien Hwang, 黃敏賢
• Other Authors: Kuo--Liang Chung
• Format: Others
• Language: zh-TW
• Published: 2001
• Online Access: http://ndltd.ncl.edu.tw/handle/78003619094174933823

碩士 === 國立臺灣科技大學 === 資訊工程研究所 === 89 === Suppose two cameras are arranged in a parallel-axis configuration and the maximal disparity allowable is assumed to d. Given two stereo images, say L and R, recently Tasi and Katsaggelos presented a new and efficient algorithm for solving the dense disparity estimation problem. For solving the same problem, this paper presents an improved two--level dynamic programming approach in fast and robust manners. In level 1, the main feature points are extracted from the rows of L and R, then we apply the dynamic programming technique associated with three elimination rules to find the matched feature pairs for any two corresponding rows in a robust manner. In level 2, for that two rows, based on the matched feature pairs obtained in the first level, each row is further divided into some small subintervals. Then a banded dynamic programming technique with respect to d is employed to speed up the determination of the matched point pairs between the corresponding two subintervals. The above two-level process is performed from the middle rows to the boundary rows until all the rows in L and R are processed. An experiment is carried out to demonstrate the computational and robust advantages of the proposed algorithm. The proposed improved algorithm is quite competitive with the current result by Tasi and Katsaggelos. Experimental results reveal that the execution time improvement ratio is about 53%.