LHMap-loc: Cross-Modal Monocular Localization Using LiDAR Point Cloud Heat Map (ICRA 2024)

Xinrui Wu, Jianbo Xu, Puyuan Hu, Guangming Wang, Hesheng Wang

Our paper is available at:

• Arxiv
• IEEE

Requirements

The dependencies installation and data prepration please refer to CMRNet.

LHMap construction

python main_single_map.py with batch_size=8 data_folder=./KITTI_ODOMETRY/sequences/ epochs=120 max_r=10 max_t=2 BASE_LEARNING_RATE=0.0001 savemodel=./checkpoints/ test_sequence=00 

LHMap saving

Change folder path to save LiDAR point cloud in "./models/CMRNet/CMRNet_single_save.py"

python main_single_save.py with batch_size=1 data_folder=./KITTI_ODOMETRY/sequences/  weights='./checkpoints/xxx.tar' test_sequence=00
python main_single_save.py with batch_size=1 data_folder=./KITTI_ODOMETRY/sequences/  weights='./checkpoints/xxx.tar' test_sequence=03
python main_single_save.py with batch_size=1 data_folder=./KITTI_ODOMETRY/sequences/  weights='./checkpoints/xxx.tar' test_sequence=05
python main_single_save.py with batch_size=1 data_folder=./KITTI_ODOMETRY/sequences/  weights='./checkpoints/xxx.tar' test_sequence=06
python main_single_save.py with batch_size=1 data_folder=./KITTI_ODOMETRY/sequences/  weights='./checkpoints/xxx.tar' test_sequence=07
python main_single_save.py with batch_size=1 data_folder=./KITTI_ODOMETRY/sequences/  weights='./checkpoints/xxx.tar' test_sequence=08
python main_single_save.py with batch_size=1 data_folder=./KITTI_ODOMETRY/sequences/  weights='./checkpoints/xxx.tar' test_sequence=09

LHMap localization

Change folder path of LiDAR point cloud in "Dataset_kitti_localization.py"

python main_single_loc.py with batch_size=12 data_folder=./KITTI_ODOMETRY/sequences/ epochs=150 max_r=10 max_t=2 BASE_LEARNING_RATE=0.0001 savemodel=./checkpoints/ test_sequence=00 

Evaluation

python evaluate.py with test_sequence=00 maps_folder=local_maps data_folder=./KITTI_ODOMETRY/sequences/  weight="['./checkpoints/iter1.tar','./checkpoints/iter2.tar','./checkpoints/iter3.tar']"

Results

You can acquire the weights for map generation here, and the weights for localization here1, here2, and here3.

The iteration results in KITTI:

	Median Transl. error	Median Rotation. error
Iteration 1	0.21 m	0.94°
Iteration 2	0.06 m	0.35°
Iteration 3	0.03 m	0.33°

Citation

If you use LHMap-loc, please cite:

@INPROCEEDINGS{10610718,
  author={Wu, Xinrui and Xu, Jianbo and Hu, Puyuan and Wang, Guangming and Wang, Hesheng},
  booktitle={2024 IEEE International Conference on Robotics and Automation (ICRA)}, 
  title={LHMap-loc: Cross-Modal Monocular Localization Using LiDAR Point Cloud Heat Map}, 
  year={2024},
  volume={},
  number={},
  pages={8500-8506},
  doi={10.1109/ICRA57147.2024.10610718}}