op Points2Resolution_RANSAC
Determine the px/mm resolution of the image from a regular pattern (e.g. smt tape with 4mm separation of the holes).
The input to this stage are points (e.g. circles detected by the op HoughCircles stage).
This stage is now implemented in FireSight. This page documents the idea, usage and results of the previous implementation in MATLAB.
The Hough circles detector is not perfect and often produces some false negatives and false positives. We know that the holes in the tape form a line, thus a RANSAC is employed to detect the line from all the circles. This gives a somewhat robust estimate of the line position. The line is fitted to the inliers (returned by RANSAC) by least-squares. Next, regularly spaced points on the line are found by second application of RANSAC with a different error measure. Finally, the median of distances of neighbouring points determines the resolution.
Sample json pipeline is given in the following code snippet. The output of HoughCircles stage is passed to points2resolution_RANSAC stage. The separation of holes in the smt tape is 4mm, which is set by the 'separation' variable.
[
{"op":"HoughCircles", "name":"circles", "diamMin":14.0, "diamMax":20.0, "show":1},
{"op":"points2resolution_RANSAC", "name":"p2r_RANSAC", "model":"circles", "separation":4.0}
]
In the images, all circles (red and green) are the output of the HoughCircles stage. The green ones are the inliers found by RANSAC.
The next step is to determine the distance (in pixels) of the holes' centres. For this a median of distances of neighbouring circles is used.
- true resolution: 11.77 px/mm
- estimated resolution: 11.74 px/mm
- true resolution: 12.05 px/mm
- estimated resolution: 12.24 px/mm
