Algorithm advice: defect detection

asked 2016-01-21 22:21:49 -0500

Staring Frog gravatar image

updated 2016-01-23 13:02:16 -0500


I'm looking for some advice with developing an algorithm for defect detection.

Here are the kinds of defects I'm interested in identifying:

  • rocks in sand
  • rocks and branches in grass
  • cracks and potholes in roads
  • cracks and knots in wood

Hopefully you get the idea. I need to be able to capture the texture of the good parts of the object (offline, probably) and detect defects or significant deviations from that texture. I would train a different classifier for each of the specific situations described above, but I think the overall approach should be basically the same.

I'm writing the code in python.

Looking forward to seeing what you come up with!


Here's an example of what I'd like to be able to do: image description

In this image, I'd like to be able to get the cracks and potholes in the road. You can go a decent ways with using edge and color information, but I'd really like to be able to capture the texture of the road somehow. This should capture the fact that there are small rocks embedded in the road that create edges, but that they are part of the normal road texture.

edit retag flag offensive close merge delete


Without case specific images, impossible to give a decent answer :D

StevenPuttemans gravatar imageStevenPuttemans ( 2016-01-22 04:30:03 -0500 )edit

How about cutting down patches of road structure, patches of non road structure, calculate some features from it like color based, edge based, ... then feed that info into an SVM and do a sliding window based classification?

StevenPuttemans gravatar imageStevenPuttemans ( 2016-01-24 12:24:53 -0500 )edit