openCV H detection

Question

For a drone contest, I need to do image processing with openCV to detect an “H” (for a helicopter landing pad). I have tried some classical algorithms, but the result is not satisfying.

• SIFT (and SURF): all the angles are the same (90 degrees) so even if it finds to “H”, it is mistaken about the orientation.
• matchTemplate: it is quite good, but it is not rotation and size invariant. So I need to make too many tests with different sizes and different orientations.
• Hough Line Transform: when the drone is too far from the target or too close to it, it doesn’t detect the same lines because of their thickness.
• Machine Learning for OCR: I ignore how to make it learn accurately because the template I am searching for is unique.

Can someone give me some advices please? :)

EDIT: Here is the "H" we need to detect:

Answer 1

The best approach for recognising a helipad is to train a Haar classifier, and then run it on:

1. original image
2. Images rotated by plus and minus 22, 45, 68 ,90 degrees

A Haar classifier is trained by adding small rotations, so the above angles should be good enough to cover all rotations of the helipad in an image. another approach is to train multiple classifiers for different rotations; this is more common because Haar classifiers give up with the earliest evidence, and it is fast to run multiple classifiers than rotate a high resolution image.

One can also try template matching with rotations, but that will need a much larger number of rotations.