Hi Eitan,

Thank you for your answer.

> A few questions that will help me understand your problem a little better:
>
> 1)  When you say that one PointCloud contains freespace information, do
> you mean that it is just the unfiltered information from your sensor?
> Including points on the ground? Is it generated from the same sensor as
> marking information?

> 2) What kind of sensor are you using on your robot and how is it
> configured? Is it a planar laser? An actuated sensor of some sort?
> Mounted pointed at the ground? Etc.

I use only one sensor: a (self made) stereo camera that is mounted on 
our robot and tilted such that I sees some ground and some obstacles 
(chairs, tables, legs, ...). With our ROS driver for the stereo camera 
and stereo_image_proc I get a point cloud.

 From this point cloud I extract a new point cloud "obstacle_pcl" with a 
PassThroughFilter (from pcl) that contains all points that are higher 
than 0.1 m and lower than 0.7 m and a second point cloud "free_pcl" that 
contains the remaining points. Thus free_pcl mainly contains points 
corresponding to the ground.

> 3) Have you tried setting the minimum and maximum obstacle heights
> differently for your clearing and marking sources? Typically, I'll be
> quite liberal in setting the threshold on an observation source that clears}

I am using the default settings. Because of noise the "ground points" of 
free_pcl are in a range from z = 0 m to z = 0.1 m. As they are above 0 m 
and free_pcl/min_obstacle_height is 0 as well, I would expect them to 
clear the cells. But I will play around with different settings in the 
next days and report about the results.

Sabrina