Title:Uncertainty depth estimation with gated images for 3D reconstruction

Abstract:Gated imaging is an emerging sensor technology for self-driving cars that provides high-contrast images even under adverse weather influence. It has been shown that this technology can even generate high-fidelity dense depth maps with accuracy comparable to scanning LiDAR systems. In this work, we extend the recent Gated2Depth framework with aleatoric uncertainty providing an additional confidence measure for the depth estimates. This confidence can help to filter out uncertain estimations in regions without any illumination. Moreover, we show that training on dense depth maps generated by LiDAR depth completion algorithms can further improve the performance.