Automatic Number Plate Detection Using Python
Number plate detection: I’ve been experimenting with deep learning, namely convolutional neural networks, over the past few weeks. Google’s Multi-digit Number Recognition from Street View is a recent notable paper. Using a single end-to-end neural network, this study offers a system for retrieving home numbers from street view data. The authors go on to show how the same network may be used to crack Google’s CAPTCHA system with human-level precision.
I decided to develop a system to solve a similar challenge to gain some hands-on experience with neural networks: Automated number plate recognition. There are three reasons why I’m doing this:
- I should be able to use the Google paper’s network architecture: Because the Google infrastructure has been proved to solve CAPTCHAs as well as read number plates, it’s logical to think that it’ll do so as well. As I learn the ropes of CNNs, having a known-good network design would drastically simplify things.
- I’m able to quickly generate training data. The demand for a large amount of labelled training data is one of the primary challenges with training neural networks. To successfully train a network, hundreds of thousands of annotated training images are frequently necessary. Fortunately, the homogeneity of number plates in the United Kingdom allows me to synthesize training data.
- For plate localization, normalization, segmentation, character identification, and other functions, traditional ANPR systems have relied on hand-written algorithms. As a result, these systems typically have tens of thousands of lines. It’ll be interesting to see how good a system I can build with only a little domain knowledge and a tiny amount of code.
Tools For Number Plate Detection
#NumberPlateDetection #RegistrationPlateDetection #LicensePlateDetection