Mapping result has noise points.

[esolk]

Greet job, I forked your project and add the function of saving result to PCD file. Then I tried this on my own dataset. The mapping runs great, even I dont have the precise extrinsic. Now I have 3 questions:

1. The PCD file has some noise points. Where are they come from?
   

2. Is it possible to add loop closure on this basis?

3. There is a online extrinsic optimizing during mapping. Where can I get the final extrinsic arguments ?

[HViktorTsoi]

Hello!

1. (1) Physically, there are several reasons that can cause noise, such as interference of multiple LiDARs, reflection from glass, and so on, please check your sensor setup.
   (2) Besides, while publishing the point cloud for visualization, we drop the points with very low intensities,

   PV-LIO/src/voxelMapping.cpp

   Line 427 in 8ec4ab0

   if(p->intensity < 5){

   
   therefore the noisy points may be the remaining low-intensity points (but still higher than those we droped) . You may try not to drop the low-intensity points.

2. Please see these repoes:
   https://github.com/gisbi-kim/FAST_LIO_SLAM
   https://github.com/hku-mars/STD
   https://github.com/hku-mars/HBA

3. The extrinsic parameter are stored in the state of kf. You may access the rotation and translation part of the extrinsic parameter in voxelMapping.cpp with

auto state = kf.get_x();
state.offset_R_L_I; // rotation of IMU-LiDAR extrinsic
state.offset_T_L_I; // translation of IMU-LiDAR extrinsic

We are also planning to add feature that publishes the extrinsic parameter as pose msg.