Depth Estimation with Online Calibration for Autonomous Driving
Monocular and stereo depth estimation, with camera calibration and absolute scale recovered on the fly.
Monocular and stereo depth estimation for autonomous driving, with the cameras calibrated online while driving.
- Depth estimation: a RAFT-based approach, using attention to compute correlations and stacked GRUs to refine the disparity.
- Online calibration: recovering the camera extrinsics and absolute scale from feature tracking, known objects and the camera’s height.