Biased/constrained object pose in 2D

asked 2018-10-29 08:12:58 -0500

mrath gravatar image


I'm trying out some code to locate the 2D pose of a PCB, in a fixed distance from the camera, so only x,y,angle. I want to detect this from a render of a Gerber (CAD-like) file of the PCB (so an "ideal image").

So far it works pretty well with both template matching and SIFT feature matching. However, template matching is assuming a bit too much, while feature matching is slow, and sometimes give skewed results.

I want to know if there's any existing best practice for finding objects on a 2D plane, where I'm not interested in other transformations than translation, and orientation around the camera z axis? While still being invariant to position and z axis rotation of the object, of course. Maybe some sort of iterative template matching (e.g. search different sizes and orientations)? Or constrained feature matching (e.g. no y and x axis rotation permitted)?

Also, with the constraint that I know the distance from camera to PCB, what is the simplest way to obtain the real-world coordinate of the PCB, once it is located in pixel coordinates?


edit retag flag offensive close merge delete