Add tracking & more robust ANPR

[falkaabi]

Doing OCR on every frame of the video sometimes produces inaccurate license plate reading. To improve this, one may do the following improvements:

1. Vehicle Tracking and Image Selection:

• Tracking vehicles in a video and selecting the best frame for OCR.
• Choosing the best image for OCR could involve criteria like clear visibility of the license plate, good resolution, and minimal occlusions. (maybe use predefined line crossing to choose the best photo based on camera setup)
• Once the best image is selected, you can then apply OCR to extract the license plate number.

2. License Plate Template Matching:

• Creating a database of license plate templates is a useful approach for recognizing and categorizing plates with specific formats.
• You can use techniques like template matching, feature extraction, or deep learning-based object recognition to match the license plate against the templates.
• When a match is found, you can determine the region and plate number based on predefined rules associated with the template.

Regarding the recognition of license plates with region+plate numbers, this approach would be particularly beneficial in countries where license plates have complex formats. For example, in countries like the United States, Canada, or Germany, where license plates may have state/province-specific codes along with unique numbers, categorizing plates based on templates can greatly improve accuracy.

Looking forward to seeing some of these implemented on your next tutorials ;)