Depth Estimation from LiDAR and Stereo
DESCRIPTION
In this work, we designed a convolutional neural network architecture that performs disparity estimation and semantic segmentation tasks in combination. We proposed a model in which segment embedding learned from semantic segmentation is fused into the process for disparity estimation, which is helpful for estimating disparity in ill-posed regions. We demonstrated on KITTI and Cityscapes datasets that our unsupervised method achieves comparable results to supervised methods on KITTI and even outperforms some of them in background regions. We also developed a self-supervised method to generate high-quality dense depth maps from low-resolution LiDAR data.
RELATED Publications
J. Zhang, M. Srinivasan Ramanagopal, R. Vasudevan, and M. Johnson-Roberson, "LiStereo: Generate Dense Depth Maps from LIDAR and Stereo Imagery," IEEE International Conference on Robotics and Automation, 2020, Accepted. [arXiv]
J. Zhang, K. A. Skinner, R. Vasudevan, and M. Johnson-Roberson, "DispSegNet: Leveraging Semantics for End-to-End Learning of Disparity Estimation From Stereo Imagery," in IEEE Robotics and Automation Letters, vol. 4, no. 2, pp. 1162-1169, 2019. [arXiv] [IEEE Xplore] [code] [video]
@article{zhang2019listereo,
title={LiStereo: Generate Dense Depth Maps from LIDAR and Stereo Imagery},
author={Zhang, Junming and Ramanagopal, Manikandasriram Srinivasan and Vasudevan, Ram and Johnson-Roberson, Matthew},
journal={arXiv preprint arXiv:1905.02744},
year={2019}
}
@article{zhang2019dispsegnet,
title={Dispsegnet: Leveraging semantics for end-to-end learning of disparity estimation from stereo imagery},
author={Zhang, Junming and Skinner, Katherine A and Vasudevan, Ram and Johnson-Roberson, Matthew},
journal={IEEE Robotics and Automation Letters},
volume={4},
number={2},
pages={1162--1169},
year={2019},
publisher={IEEE}
}