TY - GEN
T1 - Pseudo Depth Maps for RGB-D SLAM
AU - Zhang, Yue
AU - Wu, Taoyu
AU - Zhao, Haocheng
AU - Du, Shuang
AU - Yu, Limin
AU - Wang, Xinheng
N1 - Funding Information:
The authors would like to thank Suzhou Inteleizhen Intelligent Technology Co. Ltd and Xi’an Jiaotong-Liverpool University for their financial support to conduct this research by Key Program Special Fund in XJTLU under project KSF-E-64, XJTLU Research Development Fund under projects RDF-19-01-14 and RDF-20-01-15, and the National Natural Science Foundation of China (NSFC) under grant 52175030.
Publisher Copyright:
© 2022 IEEE.
PY - 2022/12/10
Y1 - 2022/12/10
N2 - In visual SLAM, RGB-D cameras can actively obtain pixel depth information but are expensive and unsuitable for outdoor use. In contrast, monocular depth estimation is relatively inexpensive and more readily available. There have been some systems that fuse monocular depth estimation with SLAM algorithms, but no method that can directly generate depth maps has been tested with experiments. We propose that using monocular depth estimation results can generate 16-bit pseudo depth maps, which can be combined with monocular images as pseudo RGB-D. The relevant camera parameters can be configured according to the general RGB-D camera SLAM requirements. Experiments with Monodepth2 and ORB-SLAM3 on the KITTI dataset demonstrate that pseudo RGB-D can achieve as satisfactory results in SLAM as in stereo computation. The research paves the way for researchers can quickly examine monocular depth estimation results in more SLAM frameworks, reducing the cost of testing new depth monocular estimation frameworks.
AB - In visual SLAM, RGB-D cameras can actively obtain pixel depth information but are expensive and unsuitable for outdoor use. In contrast, monocular depth estimation is relatively inexpensive and more readily available. There have been some systems that fuse monocular depth estimation with SLAM algorithms, but no method that can directly generate depth maps has been tested with experiments. We propose that using monocular depth estimation results can generate 16-bit pseudo depth maps, which can be combined with monocular images as pseudo RGB-D. The relevant camera parameters can be configured according to the general RGB-D camera SLAM requirements. Experiments with Monodepth2 and ORB-SLAM3 on the KITTI dataset demonstrate that pseudo RGB-D can achieve as satisfactory results in SLAM as in stereo computation. The research paves the way for researchers can quickly examine monocular depth estimation results in more SLAM frameworks, reducing the cost of testing new depth monocular estimation frameworks.
KW - SLAM
KW - depth estimation
KW - pseudo RGB-D
KW - pseudo depth maps
UR - http://www.scopus.com/inward/record.url?scp=85146430850&partnerID=8YFLogxK
U2 - 10.1109/HDIS56859.2022.9991394
DO - 10.1109/HDIS56859.2022.9991394
M3 - Conference Proceeding
AN - SCOPUS:85146430850
T3 - 2022 International Conference on High Performance Big Data and Intelligent Systems, HDIS 2022
SP - 168
EP - 172
BT - 2022 International Conference on High Performance Big Data and Intelligent Systems, HDIS 2022
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 4th International Conference on High Performance Big Data and Intelligent Systems, HDIS 2022
Y2 - 10 December 2022 through 11 December 2022
ER -