Pixel-Accurate Depth Evaluation in Realistic Driving Scenarios

Abstract

This work introduces an evaluation benchmark for depth estimation and completion using high-resolution depth measurements with angular resolution of up to 25" (arcsecond), akin to a 50 megapixel camera with per-pixel depth available. Existing datasets, such as the KITTI benchmark [13], provide only sparse reference measurements with an order of magnitude lower angular resolution -these sparse measurements are treated as ground truth by existing depth estimation methods. We propose an evaluation methodology in four characteristic automotive scenarios recorded in varying weather conditions (day, night, fog, rain). As a result, our benchmark allows us to evaluate the robustness of depth sensing methods in adverse weather and different driving conditions. Using the proposed evaluation data, we demonstrate that current stereo approaches provide significantly more stable depth estimates than monocular methods and lidar completion in adverse weather. Data and code are available at https://github.com/ gruberto/PixelAccurateDepthBenchmark.git.

CNN-Based Lidar Point Cloud De-Noising in Adverse Weather

“…Furthermore, a large data set with four very realistic road scenarios was recorded in CEREMA's climatic chamber to obtain constant and reproducible fog, rain and reference conditions [39], [40]. The data set will be published and will be available under the link in the abstract.…”

Neural network‐based semantic segmentation model for robot perception of driverless vision

“…At present, depth estimation is also an important aspect of the research of driverless driving, and object detection and distance prediction are carried out through depth estimation [13]. The depth‐sensing model is used to realise the depth estimation of the vehicle camera under different weather conditions [14]. The driverless vision understanding is realised based on 3D [15].…”

Rain rendering for evaluating and improving robustness to bad weather

“…The relatively new nuScenes dataset [9] have multiple labeled scenes containing rainy images, but variation in rain intensities are not indicated. Gruber et al [26] recently released a dataset with dense depth labels under a variety of real weather conditions produced by a controlled weather chamber, which inherently limits the variety of scenes (limited to four common scenarios) in the dataset. Note that [6] also announced-but at the time of writing, not yet fully available-a promising dataset including heavy snow and rain events.…”