First week in the plan has now passed, and we are basicly one week behind :)
Well, I will admit that the plan was very optimistic, but sometimes we are just to dumb. We still have problems finding the intersection of horizontal and vertical lines, although this should be a very simple problem, we still have problems regarding, I guess, the aspect ratio... As we are using linear parametric hough transform, we cannot describe vertical lines.
We suspect that the trouble arise from using hough transform two times on the same image, one time for horizontal, and one time for vertical, simply by flipping the x and y coordinates.
The coordinates are normalized to [-1:1[ on both axises, everything should really be fine.
The problem is that the slope of the vertical lines are... drifting... I don't know how to better explain it, with a single word, but with multiple; when the lines are exactly vertical, the lines are also, then as the angle of the line changes, the distance to the real lines increases.
Any ideas ? solutions, don't hesitate to leave a pointer of what can go wrong here.
Outr sollution is probably to change to a generic hough transform, where we use the angle of the line and distance, wikipedia description here. And ultimately, if we cant get our own to work, use the one in openCV.
I will, right now, work on making some openGL application where we can visualize the calibration object, the laser plane and the 3Dpoints when we at some point in the future have a working linedetection etc.
16 October 2007
Ok, so it didn't work that well, but it seems like it is working now, also with real points.
The subproblem of disallowing the same peak in the hough-space to create multiple lines was solved by maskin out the area such that we are sure there is some minimum distance betwen the lines ( that is distancebetween the a and b parameters )
The two images shows, first the laser line on a dark image, and second the lines detected on a thresholded image.
14 October 2007
It is done, the detection works. At least whit simulated points, but why wouldn't it work with real problems :)
There are still some sub-problems whith the line detection which must be solved, shuch as, we wan't 2 lines ( one laser line hitting a 90 degree corner) , and we must be sure that we select the correct