MapBlend

Consistent Long-Term Point Cloud Mapping with Prior Maps

PRELIMINARY version

This is a PRELIMINARY version

Description

This repository is an extension to OpenLiDARMap, a zero-drift georeferenced LiDAR-only point cloud mapping algorithm.
We combine the mapping module of OpenLiDARMap with the object detection module of Autoware to filter movable objects before the mapping.
Additionally, we enable the update of point cloud maps through a raycasting-based voxel-occlusion update strategy (THIS IS NOT YET AVAILABLE IN THIS REPOSITORY).

Install

git clone --recurse-submodules https://github.com/ga58lar/MapBlend.git

./docker/build_docker.sh

Run

To run this repository, you will need a PC with a NVIDIA GPU. You also need to download the lidar_centerpoint pre-trained models from Autoware.

./docker/run_docker.sh <model_path> <config_path> <map_path> <scan_path> <output_path> <x> <y> <z> <qx> <qy> <qz> <qw>

For a detailed instruction the the core module OpenLiDARMap, please refer to the official GitHub repository.

Missing

• Object tracking module for robust movable object removal
• Life-long mapping module