Revision history [back]

OpenCV can do that perfectly fine.

1. Train cascade classifier detectors for the most common speed signs, only seperating them in different shape models.
2. Use the detector to find shape regions.
3. Apply local thresholding (which works perfectly fine on speed signs) to make a blob of the speed symbol.
4. Apply shape based matching with a database shape.

How to do that is again your task, a good place to start is the search button on top of this forum...