We can use a combination of Deep Learning and Shape Context to detect and identify license plates. DL to detect the vehicle and corners of the license plate and Shape Context to recognize the alpha numerals in the license plate. See the OpenVINO model is able to detect the vehicle and license plate as shown below.
- After finding the corners of the license plate, compute the perspective transform matrix and warp the license plate to get a frontal projection.
- Preprocess the frontal image to get ROI and run contour detection. Filter unnecessary contours based on the shape and size of the contour.
- Apply shape context matching on each contour to identify the alphabets and numerals in the license plate.
See the output of shape context detection on different Indian license plates: