Nosille small bug fix #1412
Nosille small bug fix #1412
Conversation
corelib/src/LocalGridMaker.cpp
Outdated
| bool normalSegmentationTmp = normalsSegmentation_; | ||
| float minGroundHeightTmp = minGroundHeight_; | ||
| float maxGroundHeightTmp = maxGroundHeight_; | ||
| if(scan.is2d()) | ||
| { | ||
| // if 2D, assume the whole scan is obstacle | ||
| normalsSegmentation_ = false; | ||
| minGroundHeight_ = std::numeric_limits<int>::min(); | ||
| maxGroundHeight_ = std::numeric_limits<int>::min()+100; | ||
| } | ||
|
|
||
| createLocalMap(scan, node.getPose(), groundCells, obstacleCells, emptyCells, viewPoint); | ||
|
|
||
| if(scan.is2d()) | ||
| { | ||
| // restore | ||
| normalsSegmentation_ = normalSegmentationTmp; | ||
| minGroundHeight_ = minGroundHeightTmp; | ||
| maxGroundHeight_ = maxGroundHeightTmp; | ||
| } |
There was a problem hiding this comment.
This is done in a special case when we combine 2D lidar data and 3D depth data into the local grid. Is it really never called even in that case?
There was a problem hiding this comment.
It grabbed my attention because of multiple checks for is2d() and occupancy == 0. However, after a second look, you are correct that occupancy == 2 could still get there.
My bad.
| @@ -3530,7 +3530,7 @@ LaserScan adjustNormalsToViewPoint( | |||
|
|
|||
| float result = v.dot(n); | |||
| if(result < 0 | |||
| || (groundNormalsUp>0.0f && ptr[nz] < -groundNormalsUp && ptr[2] < viewpoint[3])) // some far velodyne rays on road can have normals toward ground | |||
There was a problem hiding this comment.
viewpoint is an Eigen::Vector3f object, with indexes 0=x, 1=y and 2=z. 3 would be out of bounds.
There was a problem hiding this comment.
After, sending this I dug a little deeper and noted that there are several other places in the code where viewpoint[3] is used. The difference is that in those cases the vector is an Eigen::Vector4f, so it doesn't crash like this did. However, it seems to me that viewpoint[2] is likely the intended array value in those cases as well. Do you agree? Or am a missing something? It seems like a simple comparison between point.z and viewpoint.z, but I guess you could be doing something I am not following with the 4th value in the homogeneous coordinates.
There was a problem hiding this comment.
Thanks for the comment, the other places using viewpoint[3] are wrong. Here the ones I've found:
https://github.com/search?q=repo%3Aintrolab%2Frtabmap%20viewpoint%5B3%5D&type=code
I created an issue about it #1413
Nosille
force-pushed
the
nosille_small_bug_fix
branch
from
5e6d9b9 to
3e0d4c2
Compare
I stumbled across a small indexing bug in adjustNormalsToViewPoint. I also noted a piece of unreachable code in LocalGridMaker.