Feature/semantic layers

[mktk1117]

Large update of adding functionality of having multi-modal layers such as color, semantic or feature layers.

[georgflick]

@mktk1117 Great! I saw that there is some kind of semantic sensor package in the code. Does that mean, the software can then extract the colors (rbg) from the camera image and match / fuse it with the corresponding cell from the elevation map? or how does it work? Thanks.

[mktk1117]

Yes, we now added the functionality of raycasting the single image to the elevation map cells.
We'll have a paper explaining how it works. We'll add the link to the paper once it's ready!

[Becktor]

This is cool stuff!

Some assumptions break the point cloud generation for us (we use realsense cameras that use mm instead of m for their depth images) in pointcloud_node.py currently we are just scaling the depth image but this should perhaps be handled by a parameter?

Also on line 251 I'm assuming the 8 should be replaced by a parameter instead of being hardcoded?

Also due to ros_numpy not receiving an update for quite a while, it breaks with newer numpy versions as the np.float attribute no longer exists. A hack is to redefine it after importing numpy np.float=np.float32

I can't add these changes to the pull request but here are some initial bugs/findings

[mktk1117]

Hi Becktor, thanks a lot for your reports!