Closed leewo0 closed 5 years ago
Do you notice any accuracy performance improvement? I see that the original code uses the same params.
http://docs.ros.org/indigo/api/loam_velodyne/html/laserOdometry_8cpp_source.html
I see that too. I haven't tested any cases with ground truth. It's almost the same when i tested in a scenario with rich features by doing close loop. In the original code, we can still find corresponding features in two rows below. I think it could improve if there are less features.
Apparently
cornerPointsSharpNum
is much smaller thanlaserCloudCornerLastNum
andsurfPointsFlatNum
is smaller thanlaserCloudSurfLastNum
. Thus, no fatal error. But this causes loss of corresponding features.