Title | ||
---|---|---|
Warping-based spectral translation network for unsupervised cross-spectral stereo matching |
Abstract | ||
---|---|---|
Recently, a pair of RGB and near-infrared (NIR) cameras is applied to stereo vision systems for all-day vision applications. The images captured by the RGB-NIR stereo vision system have spectral ranges that differ significantly. Hence, the NIR image displays richer image information during nighttime. By contrast, during daytime, the RGB image generally provides abundant information. Therefore, these images can complement each other’s disadvantages in all-day environments. However, from the perspective of image matching, it is difficult to search for correspondences between two images because of their different spectral ranges. Although various methods for translating RGB images into NIR images have been proposed to solve this problem, hight-quality conversion results have not been achieved. Incomplete conversion results cause the inaccurate estimation of disparity during stereo matching. Therefore, we propose a warping-based spectral translation network (WASTNet) to enhance the training performance of a disparity estimation network by improving the performance of image translation. |
Year | DOI | Venue |
---|---|---|
2022 | 10.1016/j.ins.2021.12.075 | Information Sciences |
Keywords | DocType | Volume |
Cross-spectral stereo matching,Spectral translation,Image warping,Depth information | Journal | 588 |
ISSN | Citations | PageRank |
0020-0255 | 0 | 0.34 |
References | Authors | |
0 | 3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Yong-Jun Chang | 1 | 0 | 0.34 |
Byung-Geun Lee | 2 | 0 | 0.34 |
Moongu Jeon | 3 | 456 | 72.81 |