Curb detection challenges and our proposed method. Three main challenges of curb detection are shown (a) height feature extraction (b) different density distribution of point clouds (c) Curb point cloud quantity proportion imbalance. Solution: First propose a 3D-Curb dataset. The MSCA module is designed for multi-scale spatial feature fusion and height feature extraction. The loss group is proposed to solve the imbalance problem. Finally, we use post-processing to further improve performance.
We have developed and proposed the 3D-Curb dataset based on the large-scale, open-source SemanticKITTI dataset, adding a new curb category with 3D label, while retaining the other original 28 semantic categories. This dataset was collected using a 64-line LiDAR, providing a comprehensive view of various street scenes as a universal autonomous driving dataset.
-
Our dataset can be visualized using SemanticKITTI API
-
The download link for the 3D-Curb dataset is HERE. (It's only half open now)
-
The annotation of the 3D-Curb dataset is based on the SemanticKITTI format. The point cloud file is stored in the .bin file and the label is in the corresponding .label file.
Because the Curb category is additionally added, we set the Curb labels category to 3. Therefore, when using API visualization, please add 3: "curb" to label: in config/semantic-kitti.yaml to increase the visualization of curb.
The data is organized in the following format:
/kitti/dataset/
└── sequences/
├── 00/
│ ├── labels/
│ │ ├ 000000.label
│ │ └ 000001.label
│ └── velodyne/
│ ├ 000000.bin
│ └ 000001.bin
├── 01/
├── 02/
.
└── 10/
If you find CurbNet or 3D-Curb Dataset useful in your research or applications, please consider giving us a star 🌟 and citing it.
@article{zhao2024curbnet,
title={CurbNet: Curb Detection Framework Based on LiDAR Point Cloud Segmentation},
author={Zhao, Guoyang and Ma, Fulong and Liu, Yuxuan and Qi, Weiqing and Liu, Ming},
journal={arXiv preprint arXiv:2403.16794},
year={2024}
}