Revision history [back]

Your approach is in principle correct. However note that you will always get some matches between your images. Therefore you need to filter them out, additionally to your min_dist-test you can

• use ratio-test
• use cross-check (for FLANN you need to code it yourself, BFMatcher has an option for that)
• try different RANSAC parameters
• warp the image according to your homography you found and see how good this warped image matches the other one (e.g. via cross-correlation)
• apply other geometrical constraints or verification steps

Maybe this similar question will also help you http://answers.opencv.org/question/983/object-detection-using-surf-flann/