Parameter Tuning

[ashnarayan13]

Hi,

Could I get some information on the parameters being used?

I didn't understand the search radius, deviation and buffer.

Thanks!

[jacobhuesman]

I wouldn't recommend using this package. I developed this Freshman year of college while learning C++. While it kind of works, I never finished the implementation and it's very computationally intensive.

If you're looking for something that'll take in a point cloud and give you an occupancy grid, check out http://wiki.ros.org/costmap_2d.

[jacobhuesman]

To answer your question,

• search radius: You can give the point cloud library (PCL) a point cloud, a three dimensional index, and a search radius and it will return a vector that represents its closest approximation of the normal vector to the surface contained at the index provided within the search radius.
• deviation: I took the difference between the normal vector returned from the PCL function mentioned above and compared it to a reference vector. If the difference between the two was greater than deviation, I marked that cell in the occupancy grid as occupied.
• buffer: Poorly named, this is the minimum number of points over an occupancy grid cell that are calculated to be occupied, before the occupancy grid is marked as occupied. The normal vector calculation generates some false positives, so I had to have a threshold for the number of values that get marked positive before the space is considered occupied.