A Novel AI-Powered Technique for Ontario License Plate Recognition

The perception system is the key for autonomous vehicles (AVs) to sense and understand the surrounding environment. Monocular cameras, being cost-effective and mature, are employed to construct a detailed and accurate visual representation of the world surrounding the AV. The objective of this paper is to optimize a camera-based deep learning model with the transfer learning technique on real-time License Plate Recognition (LPR). This paper provides the following original contributions: (1) construct an Ontario license plate dataset, (2) train multiple license plate datasets worldwide using the Convolutional Recurrent Neural Network (CRNN) algorithm with a pre-trained model from a synthetic word dataset, (3) optimize the model by data augmentation methods and a post-processing classifier. The optimized model demonstrated good performance on Ontario LPR, achieving 97.53% accuracy with 72 Frames Per Second (FPS) inference speed.