Number Plate Detection and Reading from Vehicles

Inspiration

We have a lot of problems on roads, parking lots and other different places to find some specific problems. We also face problems like Car theft or vehicle tracking using CCTV cameras and finding any specific vehicle in a specific area. We cannot monitor every traffic camera. So we came up with idea to create a Number plate detection and reading model for cars in Pakistan

What it does

It takes video feed from CCTV cameras and extract vehicles from those images. Than we find number plate in that image and then we apply Deep Learning model trained using Tensorflow to read number from that number plate.

How I built it

We first collected dataset from roads and then labeled dataset. Then we trained three different model for Car detection, Number Plate detection and Number Plate Character Reading.

Challenges I ran into

We had to work a lot for collecting that dataset using different cameras. We then labeled dataset using different tools to train models. Also it took a lot of process to achieve good results.

Accomplishments that I'm proud of

We achieved 98.37% test accuracy on Number plate reading model, 96.71% on Number plate detection and 97.09% on Car detection from video feed. Also we got best results on non-standard number plates which are hard to identify and read from vehicles.

What I learned

I learned how to tain model for big data and different situations. I also learnt how to achieve good results by modifying my model layers and data.

What's next for Number Plate Detection and Reading from Cars

Next steps are to reduce computation time for image and achieve good results by applying different deep learning and common image processing techniques. We are also working to add this feature to all other vehicles like car, truck, bus and many other on road.

Built With

• docker
• python
• tensorflow