FPGA Implementation for Character of License Plate Recognition System using CNN

Abstract

License plate recognition, which uses optical character recognition (OCR) techniques to identify vehicle registration numbers, has become an essential component of modern intelligent transportation systems. This research focuses on creating a Character of License Plate Recognition (CLPR) system using a Convolutional Neural Network (CNN) model to address applications in parking management, toll automation, traffic monitoring, and law enforcement. The objective is to create a CNN-based character recognition system and implement it on an FPGA for real-time embedded applications. A dataset of 35,500 grayscale, 28 x 28 pixel pictures from European license plates was used for testing and training. With 94% accuracy, the CNN model was created and trained in Google Colab using Python. Verilog HDL in Intel Quartus Prime 23.1 is used to extract and convert the trained weights and biases into decimal and hexadecimal text files for FPGA implementation. The model is validated through an Excel-based simulation, and floating-point parameters are changed to integers to guarantee compatibility. The Altera DE2-115 FPGA board houses the CNN module, which runs at a clock frequency of 50MHz and requires 40µs for character recognition and 8ms for LCD display. With a power consumption of just 137.24mW, the system is ideal for low-power, real-time embedded systems, according to power analysis.