LAUSR

Accurate 3-D object detection with LiDAR is critical for autonomous driving. All existing research studies were based on the flat-world assumption. However, the actual road can be complex with steep sections, which breaks the premise. Current methods suffer from performance degradation in this case due to difficulty correctly detecting objects on sloped terrain. This work presents the first full-degree-of-freedom 3-D object detector, Det6D, without spatial and postural limitations to improve terrain robustness. We choose the point-based framework because of its flexible detection range. A ground-aware orientation branch leveraging the local ground constraints is designed to predict full-degree poses, that is, including pitch and roll. Given the difficulty of long-tail nonflat scene data collection and 6-D pose annotation, we present Slope-Aug, a data augmentation method for synthesizing nonflat terrain from existing datasets recorded in flat scenes. Experiments on various datasets demonstrate the effectiveness and robustness of our method on different terrains. The proposed modules are plug-and-play for existing point-based frameworks. The code will be available at https://github.com/HITSZ-NRSL/De6D.