Pattern Detection

I got a situation: Design a card/pattern, and select an detection algorithm, make the results as best as I can.

So, I designed a card with many black-white cross, and I trained them using adaboost,
but the results were not as good as I thought, especially when the camera was tilt ?

so… anybody has some fantastic way, I will be grateful.

best regards.

blackball

Hi google for SURF (Speeded Up Robust Features) and/or SIFT (Scale invariant feature transform) … I think at least SURF is implemented in OpenCV since version 2.0.

These are algorithms to detect and describe local features in a image. It’s possible to use both width different camera angles and they are in general robust to changes in image scale, noise, illumination and geometric image distortion.

Maybe it’s possible to use a cascade of them to detect different objects. On the other side I think Chris O’Shea (or somebody different?!) posted some time ago some experiments with the haarcascade and recognition of poker cards.

[quote author=“m9dfukc”]Hi google for SURF (Speeded Up Robust Features) and/or SIFT (Scale invariant feature transform) … I think at least SURF is implemented in OpenCV since version 2.0.

These are algorithms to detect and describe local features in a image. It’s possible to use both width different camera angles and they are in general robust to changes in image scale, noise, illumination and geometric image distortion.

Maybe it’s possible to use a cascade of them to detect different objects. On the other side I think Chris O’Shea (or somebody different?!) posted some time ago some experiments with the haarcascade and recognition of poker cards.[/quote]

SURF & SIFT are based on feature points detection, I tried them, but I found they were not robust in a dynamic environment .
the cascade method…I think,you meant using a boosting algorithm, as I said before, adaboost did not agree with me. Maybe the pattern I designed was not a good one.

anyway, thx for your reply.