obstacle_detection could detect holes about rtabmap_ros HOT 2 CLOSED

segmentFlatObstacles is used to segment planar objects like a table to be labeled as an obstacle instead of ground. All planes not coplanar to the biggest plane are considered as obstacles. To work, the camera must at least see always the ground and that the ground takes a big part of the camera image.

In the current code, holes are not segmented as obstacles. In your example, obstacles are detected on the other side of the hole, and the "hole" itself doesn't have any ground or obstacle points. Here is an idea:

We could project the points hitting under the ground plane (on the hole's wall) on the ground plane and label them as obstacles. This could be done by finding the intersection of the rays with the ground plane (doing it only with points under the ground plane). If you are in a 2D environment, ray tracing for all points under z = 0 could be done.

On the move_base part:

• laser : mark, don't clear
• kinect_obstacles (obstacles over ground plane including the projected points above): mark, clear
• kinect_ground: don't mark, clear

cheers

from rtabmap_ros.

Good stuff, sounds achievable in a usable way.

Lasers not clearing could throw up a few extra false-positives where move_base gets stuck trying to recover with a nearby non-existent collision, which then can't be cleared via the point_cloud/kinect/stereo input. Would have to rely on timeout clearing costmaps.

I'm by no means a mathematician but can brute-force my way through some Pythagoras when required. Got a lot on for the next month or 3, but will come back to this eventually if no one else does.

from rtabmap_ros.