hku-mars/HBA

PointClouds are not visualized properly

pauldeee opened this issue · 6 comments

pauldeee commented

When using poses from the GTSAM graph with LIO-SAM the point clouds are not aligned properly.

Any ideas how to fix this?

Thanks

samsdolphin commented

Hi @pauldeee, I think this is the LIO-SAM issue. There is some extrinsic between the LiDAR and the imu in the LIO-SAM configuration. Please make sure you have the point cloud and the corresponding poses in the same frame, i.e., both in the LiDAR body frame or the imu frame.

pauldeee commented

Hi @samsdolphin,

I figured it out. The documentation states the pose file should have the format tx ty tz qw qx qy qz.

image

When I switch my file to have the poses in the format tx ty tz qx qy qz qw it works.

Thanks!

pauldeee commented

Sorry, I may have spoken too soon.

It seems while the change I made allows the visualize.launch to work correctly, hba.launch does not produce the expected results...

visualize.launch
image

After running hba.launch with the default values then running visualize.launch again:
image

samsdolphin commented

Hi @pauldeee, maybe you could share your point cloud and poses so I can figure out which parameters are suitable for your case? It looks very weird the overall point cloud quality decreases so badly after just one iteration :(

pauldeee commented

Hi @samsdolphin,

Here is a link to some test data: https://drive.google.com/drive/folders/1tn7ZFGnVPS39j8jd3yWux9ykgKQkKUXV?usp=sharing

This works fine with the visualize.launch. Could it be a TF issue that's causing it not to work with the hba.launch even though it works fine with the visualize.launch?

Thanks!

samsdolphin commented

Hi @pauldeee, I checked your data and I found only 97 poses (pcds). If this is your total number of point clouds, I would suggest you use the BALM directly as HBA is designed to handle large-scale point cloud mapping, e.g, 1K or 10K or more poses.