octomap in dynamic environment
Hi,All,
I am using a 3D lidar together with octomap package to build a 3D map for an indoor environment. Since my lidar is stationary, I actually don't need any transform between lidar frame and map frame. Everything works fine except that when there are moving objects, the 3D map produced seems to be messy. Those occupied cells are not cleared even when the moving objects moved away. I know this has sth to do with the update rule of the octomap and I have been trying to play around with the octomap parameters but no success. What am I missing here? Thanks a lot!
My platform: Ubuntu 12.04, ROS Fuerte, Octomap is the one checked out from ROS wiki page.
They should be cleared. Are they not changing at all or at least changing a bit, but slowly? Were they at max likelihood before? Are you getting lidar measurements through the places where objects were?
Another strange observation: if I pass the PCL2 generated by the lidar directly to the octomap server, the dynamic objects are actually cleared!. However, if I use the PCL filter (such as outlier removal) to filter the raw PCL2 and pass the filtered PCL2 to the octomap server, the dynamic objects are not cleared. A walking person will left a trail (3D cells) behind him. And, sometimes I got "Not enough inliners found to support a model (0)! returning the same coefficients" error. Back to your questions: Where can I check max likelihood? My lidar is in the centre of an indoor space where people walked by occasionally.
That seems to me like the filter, filters out something. Might be time to look at both point clouds and see what's happening.
The filtered PCL looks normal, at least in rviz.
Is it maybe downsampled?
In the case of voxel filter, the PCL are downsampled. The dynamic object won't clear completely. It left a sparse trail behind. In the case of passthroughfilter (Z axis), the PCL are limted to certain height, the dynamic object will left a trail on the floor.
I guess that might be the problem. If it's downsampled, you have less end points to go through. Thus you get some traces, but not enough, so that some cells "in between" the rays are never visited and thus not cleared. If that's the case it should look like the objects are "thinned out in density".