Hello, I'm working on a high speed camera project to recognize golf ball and club. So according to my experience, it's sure that you can use a camera to build an automatic identification system. It's easy to recognize some circles and lines in a picture by Hough transformation. But object like fastener is hard to find single feature as a line or a circle to represent itself. So my advice is that you can mark some circles on the fasteners. According to the location sequence of circles, you can recognize the fasteners. This is how GC2(a product of foresight sports) recognizes golf club face.

Presumably you can stage the camera view of the fasteners, under roughly the same lighting conditions, without being blocked by clutter or other fasteners, ideally without one fastener touching another. That helps photo/video recognition a lot. Otherwise be prepared for many errors. 

You might consider model matching: Give a computer a 3D model (CAD design, or laser scanner image) of every fastener you want to recognize. Then the model rotates the models, constructing is own synthetic image of the model fastener in each rotation for comparison with the actual image, picking the fastener whose model creates the best-fit synthetic image of the actual image. It's pretty much what humans do in their head for fastener recognition presumably.

The synthetic-to-actual image best fit is defined under some distance measure or objective function based on image feature sets that you will have to determine and experiment with.

Model matching will very computationally intensive, especially if you have thousands of different kinds of fasteners. So you will have to pay close attention to computation speed and its optimization, presumably with some progressive refinement toward ever smaller subsets of possible-fit fasteners under increasingly intensive objective functions.

You might consider combining camera and (3D) laser scanner images of fasteners for better performance -- basing the initial possible-fit fastener subset on the camera, and the final fastener selection on the laser.

Ronald

salam

have a good day. can you take a look to this code program https://github.com/rishisij/workhouse

Thank you for all the response. 

Mr Yu luo, never thought about marking the fastener. But that's a good idea. However, I dont think I have circles on the image of fastener. The image is just a fastener (eg a screw) lying on a white surface.

Mr Kessel, I have no problem with clutter because the image contain just one object on a white surface.  I think it is a great idea to use 3d image because we get to see the type of head of the fastener( philip or slot). So far, I just have the screw image without the having the head type image.