A. Ahrary, L. Tian, S.-I. Kamata, and M. Ishikawa (Japan)
Stereo matching, Sewer inspection, Cooperative, Linear computation.
Stereo matching is an essential issue in computer vision. Recently, many stereo matching algorithms based on seg mentation, graph cuts and so on have been proposed. Be cause the disparities change continuously in sewer environ ment, these methods are not applicable to sewer systems and are computationally expensive. In this paper, we pro pose a new cooperative stereo matching algorithm that is suitable for this task. In our algorithm, a reference im age is divided into a feature pixel group (edges, cracks, etc.) and a non-feature pixel group (walls, etc.). Then, a proposed matching measure named Linear Computation is implemented in the feature pixel group. Also, the conven tional measure of the Sum of Squared Differences (SSD) is implemented in the non-feature pixel group. In order to improve the accuracy of the conventional SSD measure, we also impose constraints such as an evidence constraint and a neighboring similarity constraint on the SSD mea sure. The proposed cooperative algorithm achieves better results than other conventional algorithms in the sewer en vironment.
Important Links:
Go Back
