License Plate Recognition System Using Computer Vision and Optical Character Recognition

Abstract

The increasing volume of vehicles in India has made manual traffic monitoring and vehicle record-keeping slow, error-prone, and unsuitable for large-scale deployment. License Plate Recognition (LPR) systems are now commonly used in tolling, parking, and surveillance applications to automatically extract vehicle registration numbers from images or video streams. But most high-accuracy LPR solutions use deep-learning models and GPU-based infrastructure, which makes them more expensive and harder to use in places like colleges, gated communities, and city deployments where money is tight. This paper describes a cheap LPR framework that uses Python, OpenCV, and Tesseract to do classical computer vision and Optical Character Recognition (OCR). The system is specifically designed to work with Indian High-Security Registration Plates (HSRP) and tests performance in the day, night, shadow, and blur. Testing on a dataset of 500 images achieved an average plate-detection accuracy of 94% and OCR accuracy of 91% on CPU-only hardware, with a mean inference time of 1.5 s per image. The results show that traditional image-processing pipelines can work well in India without the extra work of deep learning. This makes the system good for smart parking, automated entry gates, and low-cost traffic management.

Introduction

Automated vehicle identification plays a vital role in modern transportation for tolling, access control, law enforcement, and traffic management. With India adding over 20 million vehicles annually, manual number recording has become inefficient and error-prone. License Plate Recognition (LPR), also known as Automatic Number Plate Recognition (ANPR), uses optical and computational methods to extract alphanumeric text from vehicle plates. However, India’s adoption of ANPR has been limited by inconsistent plate designs, low-quality cameras, multilingual text, and high system costs. The introduction of High-Security Registration Plates (HSRP) under the Motor Vehicles Act (2019) has standardized plate design, creating an opportunity for scalable, India-specific LPR solutions.

This paper proposes a lightweight, open-source, CPU-based LPR system that uses classical computer vision (CV) and Tesseract OCR, avoiding reliance on deep learning and expensive GPU hardware. Designed for affordability and edge deployment (e.g., Raspberry Pi or PCs), the system offers:

1. An end-to-end LPR pipeline optimized for Indian HSRP plates.

2. Evaluation under four real-world conditions—day, night, shadow, and motion blur.

3. Comparative analysis with deep learning models in terms of cost, accuracy, and hardware use.

4. A real-time web interface built with Flask.

The literature review shows LPR’s evolution from heuristic edge-based detection to deep learning methods (YOLO, CRNN, Transformers). While deep models achieve >97% accuracy on Western datasets, they require GPUs and fail on Indian multilingual or low-resolution plates. India-specific research remains limited to small datasets and daylight-only conditions.

The proposed system fills these gaps with a modular, explainable pipeline comprising image acquisition, preprocessing, plate localization, character segmentation, OCR-based text extraction, and rule-based validation. Implemented in Python using OpenCV and Tesseract, it achieves around 94% detection accuracy and 91% OCR accuracy on Indian plates using only CPU hardware—demonstrating a cost-effective, deployable alternative to GPU-dependent LPR systems.

Conclusion

This paper presented a CPU-based, open-source License Plate Recognition system optimized for Indian High-Security Registration Plates using classical computer-vision techniques and Tesseract OCR. The system was evaluated on a dataset of 500 real-world vehicle images under four environmental conditions and achieved an average plate-detection accuracy of 94 % and OCR accuracy of 91 %, with an average processing time of 1.5 s per image. The results demonstrate that a classical CV + OCR pipeline, when properly engineered, can deliver competitive accuracy without the computational expense of deep-learning models, making it suitable for cost-sensitive deployments in India such as college campuses, gated societies, and medium-scale traffic monitoring applications. The system meets the primary design goals of affordability, deployability, and reproducibility, and provides a practical middle ground between traditional image-processing methods and GPU-dependent deep-learning solutions. Several extensions are planned to improve the system’s robustness and real-world applicability: 1) Deep-Learning-Based Detection Module – Replacing contour-based detection with YOLOv8 to enable real-time multi-plate detection in video streams. 2) Custom OCR Model for Indian Fonts – Training a lightweight CNN-based OCR model for state-wise font variations and bilin-gual plates. 3) Night-Time Enhancement – Integrating IR-assisted cameras and histogram equalization to improve low-light performance. 4) Video-Based Tracking and Logging – Adding multi-frame vehicle tracking for traffic enforcement and toll collection. 5) Edge Deployment and Cloud Integration – Porting the system to Raspberry Pi, NVIDIA Jetson, and linking to dashboard-based analytics for smart-city applications.

References

[1] G. Bradski and A. Kaehler, Learning OpenCV 4, Sebastopol, CA, USA: O’Reilly, 2018. [2] R. Gonzalez and R. Woods, Digital Image Processing, 4th ed., Pearson, 2018. [3] M. Rahman, S. Saha, and P. Dey, “Vehicle License Plate Recognition Using Image Processing Techniques,” International Journal of Engineering and Advanced Technology, vol. 9, no. 3, pp. 532–538, 2020. [4] D. Soumya and S. Patel, “Hybrid Approach for License Plate Extraction Using SVM and Edge Operators,” International Journal of Recent Technology and Engineering, vol. 7, no. 6, pp. 101–106, 2018. [5] J. Redmon and A. Farhadi, “YOLOv4: Optimal Speed and Accuracy of Object Detection,” arXiv preprint arXiv:2004.10934, 2020. [6] X. Li, Y. Zhang, and L. Hu, “Deep Learning-Based Automatic Number Plate Recognition,” IEEE Access, vol. 9, pp. 11203–11215, 2021. [7] P. Kiran, A. Singh, and R. Shetty, “License Plate Detection Using Morphological Operations and Contour Analysis,” IJERT, vol. 8, no. 5, pp. 120–124, 2019. [8] R. Smith, “An Overview of the Tesseract OCR Engine,” in Proc. ICDAR, 2007, pp. 629–633. [9] S. Jain et al., “Indian Number Plate Recognition Using YOLO and CRNN,” in Proc. IEEE ICCCNT, 2022. [10] Flask Documentation, https://flask.palletsprojects.com, accessed Jan. 18, 2024. [11] OpenCV Documentation, https://docs.opencv.org, accessed Jan. 18, 2024. [12] Ministry of Road Transport and Highways, “High-Security Registration Plates Order,” Government of India, 2019. [13] S. Patel and M. Sharma, “Performance Evaluation of OCR Models for Indian License Plates,” IEEE SPICES, 2023. [14] A. Das and K. Rao, “A Review of ANPR Systems for Smart City Applications,” IEEE Xplore, 2022.

Copyright

Copyright © 2025 Renu Chaudhary, Akshat Kumar Srivastava, Tanishq Mann, Kanan Saini, Harshita Rawat. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.