gisbi-kim
commented
3 years ago Hi @SiyuanHuang95 thank you for using the project.
Q1) However, I thought in your code, the default angle is still 360 degrees. In such a case, most of the columns will be set to zero? Am I right? -> Yes, you are right. I get them remained as zeros for the ease of implementation convenience. and it works as you might see.
Q2) By the way, I found the reverting LC is quite difficult for Horizon, any idea for that? -> scan context works well when the overlap between two scans is 100%. Therefore, scan context works well in the forward direction even with lidar that only acquires 80deg. However, in the case of reverse revisit, it is difficult for the overlap to be the same in the case of lidar of limited hfov. My one suggestion is: First, livox lidar scans are stacked over poses. Then, the submap point cloud is reprojected to create a virtual scan with full hfov like KITTI, and use it to create the scan context descriptor of the place.
SiyuanHuang95
huangsiyuan0717
hr2894235132
CLboyrobot
Hi Kim,
thanks for your sharing and wonderful work! It did help a lot.
However, I had some unclear questions currently wrt the Scan Context used in FAST-LIO where Livox Horizon is used. The FOV of Horizon is about 80 degrees, far less than the Hesai or Ouster series. However, I thought in your code, the default angle is still 360 degrees. In such a case, most of the columns will be set to zero? Am I right?
By the way, I found the reverting LC is quite difficult for Horizon, any idea for that?
Bests,