Cartesian Image

[tpanzarella]

I'm in the process of moving the xyzi_image from the ROS code into libo3d3xx and in the process I am converting the data types as discussed here with @graugans @semsitivity Now that I am in the code, I realize that we can't implement as discussed. The reason being that the X Y Z data are 16-bit signed ints while the amplitude is 16-bit unsigned. I'm going to propose that we drop the amplitude from that data structure since it is already available as a cv::Mat via the AmplitudeImage() accessor. The new structure will simply be a cv::Mat encoding the spatial components of the point cloud stored as 16-bit signed ints.

This will have the biggest effect on any ROS users that currently use the xyzi_image (i.e., me) but I think this is the correct change to make for the long-term. Any objections? Another option would be to keep the amplitude data bundled with this data structure but store all of the data as 32-bit ints. Is there a preference?

[semsitivity]

That was just a suggestion. Actually, expressing cartesian coordinates with an unsigned type is pretty impossible since the sign depends on the used axis convention. I realised the same thing after trying to rotate and translate those point to align them with my "detected floor". (my axis system choice being different than IFM and yours, x going to right, z for depth and y for up direction, setting then all my point positive according the origin position)
I think I have no more "global" suggestion concerning the xyzi_image, one will choose and convert to its own type depending the use case and constraints.