Overview / Usage

The current ANPR solutions are too expensive to run from a hardware perspective. They are also not tuned to specifically work in the Indian Context with its variety of number plate designs. This Project aims to tackle this issue, by using a custom-trained YoLov5 specifically exposed to Indian number plate images, we solve the issue of number plate detection. Further, we also transcribe the number plate by passing the detected number plate portion to the Tesseract API.

The biggest Advantage of the current implementation is that it can also be run on a mobile device.

Methodology / Approach

1.)Scraping images of number plates and cars from publically available places including OLX, Google Images etc.

2.)Using Roboflow to create synthetic data from existing ones

3.)Running training on yolov5 with custom dataset

4.)Deploying the model and connecting it to Tesseract for OCR.

Technologies Used

1.)RoboFlow

2.)Tesseract

3.)PyTorch

4.)Ultralytics

5.)YoLov5

Repository

https://github.com/ojasaklechayt/Brainihacks-VITISH/tree/main/Models/ANPR%20System