search

HKUST-Aerial-Robotics / G3Reg

A fast and robust global registration library for outdoor LiDAR point clouds.
MIT License
193 stars 11 forks source link

About the effect of real-time frame and map matching #9

Open tust13018211 opened 1 month ago

tust13018211 commented 1 month ago 
  Thanks for your amazing work, Have you tested the effect of matching single frame and map, Will the point cloud of the map be denser and the extracted features more abundant, leading to matching problems？
qiaozhijian commented 1 month ago

For submaps, the point cloud is denser, and the extracted GEMs will be more numerous and complete. I believe this will produce better results than using a single frame.

However, you should note that our registration relies on the results of point cloud segmentation, and the current segmentation algorithms, such as TRAVEL, are designed for single frames. Its parameter file contains some parameters specific to scans, such as the number of beams, angular resolution, upper and lower angle limits, and so on. You need to adjust these parameters, for instance, by setting a higher number of beams (since the map is not as sparse as a scan), and setting the upper and lower limits based on the actual situation.

I have tried registering submaps and found that adjusting these parameters can achieve fairly good segmentation results, and the registration effect is also good. However, I haven't conducted extensive experiments, but I suspect that using submaps would be a more practical choice.

tust13018211 commented 1 month ago

ok , thanks your reply ,I will test use the submap effect, I will update the result.