Unified system boosts license plate reading speed on small devices

A Unified Framework for License Plate Identification and Character-Recognition

Machine LearningArtificial IntelligenceComputer Vision and Pattern Recognition

Summary

Reading license plates from pictures can be tricky, especially on small, low-power devices like those in cars or traffic cameras. The authors designed a single system that finds license plates and reads their letters directly from image features without extra steps. They trained it with many real license plate images from all U.S. states, including tough conditions like bad weather and blur. Their system runs very fast and uses little memory and power compared to older methods, making it better for devices with limited hardware.

What this means in practice

  • For embedded system developers: Build low-power camera devices that quickly identify license plates on edge hardware using a unified detection and recognition model
  • For smart city infrastructure teams: Deploy efficient license plate reading solutions with high accuracy for traffic monitoring and law enforcement on existing small computing platforms

Authors

Saurabh Kumar, Dewant Katare, Prakriti Jain

Abstract

This paper presents a unified framework for license plate detection and character recognition on low-power edge hardware. The proposed network uses a shared CSP-Darknet and PANet feature encoder with an anchor-free localization head and a RoIAlign-based sequence-recognition branch. Character decoding is performed directly on aligned feature maps, avoiding image-space plate cropping and a second recognition backbone. The model is trained on an AI-assisted dataset of approximately 37,000 manually verified images covering license plate layouts from all 50 United States. The dataset includes variations in illumination, viewpoint, weather, blur, occlusion, and plate design. The framework is converted to an INT8 TensorRT engine and evaluated on an NVIDIA Jetson Orin Nano with 8 GB memory. The INT8 model shows an mAP50 of 97.8% and a full-sequence accuracy of 90.7%. It records 147 FPS at 6.8 ms end-to-end latency, with 315 MB inference memory and 9.4 W average power consumption. Relative to the YOLOv8s+CRNN baseline, the proposed deployment provides a 7.1x throughput increase and a 77.8% reduction in allocated inference memory