Badhorse is a simple framework for performing automobile license plate recognition
We can count on a couple of items that will make our job easier (assumes fixed camera):
- The plate will always appear within a fairly defined zone (unless the driver is not in their lane!)
- Any angle or perspective will be the same every time, hence correctable
- The plate will always be the same size and shape
This means that we should always start with the following steps:
- Cropping (user defined area - tall but narrow for centered plates)
- User controls: crop area
- Perspective and rotation correction
- User controls: rotation and perspective
Next, we can actually perform the shape detection:
- Edge detection (Canny filter) on Hue
- Edge detection (Canny filter) on Value
- Combine above to produce composite
- Hough line filter, discarding non-vertical/horizontal lines
Before compiling test1, make sure you download the latest copy of Boost and set BOOST_ROOT in Makefile to the Boost folder You will also need OpenCV
Boost is found at http://www.boost.org