Skip to content

Latest commit

 

History

History
75 lines (52 loc) · 3.48 KB

README.md

File metadata and controls

75 lines (52 loc) · 3.48 KB
Raw

Det6D: A Ground-Aware Full-Pose 3D Object Detector for Improving Terrain Robustness

Junyuan Ouyang, Haoyao Chen

[project page] [paper] [video]

This repo is the official code release of Det6D, the first full-degree-of-freedom 3D detector.

Citation

If you find this project useful in your research, please consider citing:

@ARTICLE{ouyang2022det6d,
  title={Det6D: A Ground-Aware Full-Pose 3-D Object Detector for Improving Terrain Robustness}, 
  author={Ouyang, Junyuan and Chen, Haoyao},
  journal={IEEE Transactions on Instrumentation and Measurement}, 
  year={2022},
  volume={71},
  number={},
  pages={1-9},
  doi={10.1109/TIM.2022.3219469},
  publisher={IEEE}
}

News:

  • [2022/11/17]: release source codes.
  • [2022/10/26]: 💥 Det6D accepted by IEEE Transactions on Instrumentation and Measurement.
  • [2022/08/01]: Det6D is now on arXiv, and code will be released after the paper is accepted.
  • [2022/06/28]: Add project page.

Introduction

Accurate 3D object detection with LiDAR is critical for autonomous driving. Existing research is all 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 3D 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, i.e., including pitch and roll. Given the difficulty of long-tail non-flat scene data collection and 6D pose annotation, we present SlopeAug, a data augmentation method for synthesizing non-flat 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.

Preview

KITTI

fontview sideview

GAZEBO

results scene

GTA-V

results scene

Acknowledgement

We thank the authors of OpenPCDet for their open source release. Parts of our code base on the recent work SASA.