Semantic-Vector HD Map

An open-source HD vector map (HDVM) generation pipeline for autonomous vehicles, integrating GNSS, INS, LiDAR, and camera data.

HDMap provides a complete pipeline for constructing high-definition semantic and vector maps, designed for autonomous driving in complex urban environments. Unlike traditional methods that rely on planar assumptions, our approach fuses multi-sensor data to produce accurate 3D HD maps.

Pipeline Overview:

1. Semantic extraction — Extracts semantic information from raw images using Vision Transformer (ViT) and Swin Transformer architectures
2. 3D reconstruction — Obtains absolute 3D coordinates of semantic objects from LiDAR depth data
3. Precise localization — Uses GNSS-RTK and INS for high-precision pose estimation
4. Vector map generation — Extracts vector features (e.g., lane markings) to form the HD vector map
5. Error analysis — Provides an error propagation scheme analyzing segmentation and LiDAR-camera extrinsic calibration errors

A Docker version of the pipeline is available for easy deployment.

Citation:

@article{hu2024hdmap,
  author={Hu, Runzhi and Bai, Shiyu and Wen, Weisong and Xia, Xin and Hsu, Li-Ta},
  title={Towards high-definition vector map construction based on multi-sensor integration for intelligent vehicles: Systems and error quantification},
  journal={IET Intelligent Transport Systems},
  doi={https://doi.org/10.1049/itr2.12524}
}

GitHub: https://github.com/ebhrz/HDMap