So I’ve set up my light and camera and am trying to figure out how to make the image more useable. Do I go with the Roberts edge detection algorithm, or the Sobel? Perhaps Prewitt would be better…. Or does this need the Laplacian?
Suddenly realization strikes - if the choice of tool is that important, then I’m doing it wrong! Seriously. If my image processing approach is sensitive to the type of edge detection routine I use, then it’s probably not robust enough to work day-in, day-out, in an industrial environment. In fact, what my dilemma really shows is that I need to reengineer the way I light and view the part I’m looking at.
It’s been my experience that 90% of industrial inspection-type tasks can be done using either blob analysis, histograms or simple edge detection tools. No, they’re not as sophisticated as some of these fancy convolutions, but they can be made to work all day, every day. And isn’t that what we’re really trying to achieve?
Subscribe to:
Post Comments (Atom)
Posts
No comments:
Post a Comment