Definition of residuals

[manuelgitgomes]

Good evening. When trying to calibrate two 2D LIDAR sensors with each other, there is a residuals between each point captured, correct? Because, if so, I have 541 points. To calibrate between each point, residuals are 541², equaling 292681 residuals, which might be too computationally intensive for my machine to handle. Am I formulating the problem correctly and simply need to wait for it to process all, or is there and error in my formulation?

[miguelriemoliveira]

Hi @manuelgitgomes ,

let's say you have Nx points in scan x, and No points in scan o.

scan x: x x x xxxxx x x x

scan o: o o o oooooo o o o

Now let's assume we the goal is to calibrate x in o, then you have to search for each x the closest o, and the residual corresponding to that x in the minimum distance between x and all os.

This way you have only Nx residuals and it works very fast.

[miguelriemoliveira]

Does it make sense?

[miguelriemoliveira]

for each x the closest o would mean

for each x, compute the distance to all os and take the minimum.

This is common in the ICP algorithm https://en.wikipedia.org/wiki/Iterative_closest_point

[manuelgitgomes]

It makes sense! I will try it ASAP. Thank you!

[manuelgitgomes]

It worked a lot better. Still some problem, will be better defined in #65. This issue is now closed.

[miguelriemoliveira]

OK, great. Bear in mind that you may have problems with the association, unless the initial value of the transformation is very close to the real solution.

This is a common thing with the ICP algorithm, which is know to often fall into local minima unless a good initialization is provided.

[manuelgitgomes]

Exactly, we are having that problem right now.