4
$\begingroup$

I'm trying to use computer vision to identify objects an image. I am currently using the SURF algorithm, but Harris and SURF both find virtually no points on our objects (a paper ball and a deodorant can) due to their simple features. If it does find points it finds one or two that aren't the same points it find in the new frame. SURF also finds a large amount of points on our floor due to the contrast of the black dots on the carpet. Here's a sample image:

image that SURF has problems with

Is there a feature detector that doesn't use corner detection/curvature?

$\endgroup$
2
$\begingroup$

If you have prior knowledge of the objects expected size you should ignore interest points of too small scale. I think these objects (thanks to round shape and contrast with the background) it should be an easy task for blob detection algorithms, you'd get one detected blob / item. Note that for SIFT or SURF the major orientation isn't well defined for blobs with rotational symmetry but it might be sufficient to assume fixed orientation (like on U-SURF).

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.