search

zidik / soccervision

Soccer-playing robot computer vision processing
2 stars 0 forks source link

Corner detection #7

Closed zidik closed 8 years ago

zidik commented 8 years ago

Corners should be detected from vision. This can be achieved by finding field-lines (white to black transitions) from vision. Found transitions are translated to points in real world. Points are analyzed (how?) to determine if it represents a corner

First implementation can just return the point with the largest distance as a corner. (should be close enough) The next implementation could use some linear regression.

jaantti commented 8 years ago

Basic corner detection is working. That means I check each black-white color transition point. If the distance to the point is increasing and then starts decreasing, that indicates a corner.

Next step would be eliminate noise and implement some checks, to determine the probability of the corner.

jaantti commented 8 years ago

Now the code will calculate slopes for points first->corner and corner->last. If -1.2 <= slope1*slope2 <= -0.8 then it is a corner. This should remove the false positives. I have to do more testing to see, if there are some more small improvement I can make.