This paper presents a deep learning- based approach for automatic number plate detection using Convolutional Neural Networks (CNN). The system focuses on accurately identifying and localizing vehicle license platesin real-time from images captured undervarious conditions. A custom CNN model is trained on a diverse dataset to extract spatial features and detect number plates effectively. The proposed method demonstrates high accuracy, robustness, and practical applicability in traffic monitoring and intelligent transportation systems.
Introduction
The demand for intelligent transportation systems has increased, relying heavily on Automatic Number Plate Recognition (ANPR) for applications like traffic monitoring, law enforcement, toll collection, and vehicle tracking. Traditional ANPR methods struggle with lighting, angles, and background noise. To address these issues, deep learning—particularly Convolutional Neural Networks (CNNs)—has become popular due to their ability to automatically learn features and excel at image recognition.
CNN Architecture:
CNNs consist of multiple layers—input, convolutional (for feature extraction), ReLU activation, pooling (to reduce data size), flattening, fully connected layers, and output layers—that work together to identify key features like edges and textures. This architecture enables accurate detection and localization of number plates in images.
Proposed System:
The system uses a CNN-based approach to detect and recognize license plates. It improves image quality through preprocessing techniques (median filtering, masking, thresholding) and applies morphological operations for better feature extraction. The license plate region is detected and then passed to an Optical Character Recognition (OCR) module to convert images into readable text.
Methodology:
Dataset of 864 images across 36 classes was used for training and validation.
The system supports image upload or real-time camera capture.
Number plates are detected by CNN, cropped, and processed by OCR for character recognition.
Recognized plates can be displayed or stored in a database for applications like vehicle tracking or toll enforcement.
System Design:
Database schema includes tables for detections, cameras, and vehicles to log and manage recognition events efficiently.
Uses OpenCV for image preprocessing and Haar cascades for initial plate detection.
Character segmentation involves thresholding, erosion, dilation, and contour detection.
Characters resized and fed into a CNN classifier trained with data augmentation, optimized with Adam and categorical cross-entropy loss.
Performance:
The system is evaluated using validation datasets, with performance metrics such as accuracy and confusion matrices to assess effectiveness.
Conclusion
The project focused on developing an automated Number Plate Detection System using Convolutional Neural Networks (CNNs) to accurately identify and recognize vehicle number plates from images taken under varying real-world conditions. Traditional image processingtechniques often struggle with inconsistencies in lighting,angles,andnoise,makingdeeplearning a more robust choice. In this project, a comprehensive datasetofvehicle imageswas used, covering diverse lighting, angles, and plate formats. The images were preprocessed through resizing, normalization, and grayscale conversiontopreparethemfortraining. A customCNN model was built and trained for number plate detection,with YOLOv4exploredasan alternativeforimprovedlocalization.For character recognition, the CNN was integratedwith an OCR module to extract alphanumeric characters from the detected plates. The trained models achieved a high detection accuracy of 97.69% and 99.54. It also showed efficient processing times, averaging under 0.5 seconds per image on standard hardware. The model demonstrated robustness in challenging scenarios such as low light, angled views, and cluttered backgrounds, outperforming traditional methods. However, challenges such as motion blur, stylized fonts, and low text-background contrast occasionally affected performance. The resultswere visualized with clear outputs showing successful detection and recognition pipelines, while training graphs confirmed strongconvergence and generalization. Overall, thesystemprovedeffectiveforreal-world applications like traffic surveillance, parking automation, and toll systems, with high accuracy, fast processing, and reliable performance across diverse conditions.
The current study on number plate detection using CNNs can be significantly enhanced in futurework through various directions. A key recommendation is the real-time implementationof the system using lightweight models like MobileNet or Tiny-YOLO for deployment onedge devices. Expanding the model to support multinational license plates with varying formats, fonts, and languages—alongside multilingualOCR capabilities—would increase itsapplicability. Future systems should aim forimproved character segmentation using advanced techniques like UNet or Mask R-CNN and enhanced robustness to environmental challenges such as low lighting, motion blur, and occlusions. Integration with larger systems like vehicle tracking, smart parking, and traffic enforcement would extend its practical utility. Furthermore, deployingthe model via cloudplatforms or asedge solutions can ensure scalability and accessibility. Implementing continuous learning mechanisms to adapt to new data, along with ensuring data privacy and compliance with legal regulations, is also crucial for real-world adoption.
References
[1] M. Montazzolli and C. Jung, \"Real-Time Brazilian License Plate Detection and Recognition Using Deep Convolutional Neural Networks,\"IEEE Transactions on Intelligent Transportation Systems, vol. 20, no. 3, pp. 1097–1107, Mar. 2019, doi: 10.1109/TITS.2018.2836308.
[2] H. Li, P.Wang and C. Shen, \"Towards End-to- End Car License Plate Detection and Recognition With Deep Neural Networks,\" IEEE Transactions on Intelligent Transportation Systems, vol. 20, no. 3, pp. 1126–1136, Mar. 2019, doi:10.1109/TITS.2018.2847291.
[3] D. Chanson and T. Roberts, “License plate recognition system.” www.manukau.ac.nz./EE/research /2002/dc.pdf, 2002.
[4] H. Yang, L. Xu, and L. Shi, “Design and implementation of license plate recognition system,” in First IEEE International Symposium on Information Technologies andApplications in Education,ISITAE ’07., pp. 602 – 605, 2007.
[5] J. A. G. Nijhuis, M. H. T. Brugge, K. A. Helmholtz, J. P. W. Plum, L. Sonnenburg, R. S. Verma, and M. A. Westenberg, Car license plate recognition with neural networks and fuzzy logic, in IEEE Int. Conference on Neural Networks, vol. 5, pp. 2232–2236, 1995.
[6] H. A. Patel, K. G. Patel and A. V. Shah, \"Automatic Number Plate Recognition Using CNN Based Self Synthesized Feature Learning,\" in Proc. 2018 2nd Int. Conf. Inventive Systemsand Control (ICISC), Coimbatore, India, 2018, pp. 500–504, doi: 10.1109/ICISC.2018.8398991.
[7] R. Laroca, E. Severo, L. A. Zanlorensi, L. S. Oliveira, G. R. Gonçalves, W. R. Schwartz, and D. Menotti, \"A Robust Real-Time Automatic License Plate Recognition Based on the YOLO Detector,\" arXiv preprint arXiv:1802.09567, Feb. 2018. [Online].Available: https://arxiv.org/abs/1802.09567.
[8] https://www.kaggle.com/datasets/sarthakvajpayee/ai-indian-license-plate-recognition-data
[9] T. Joseph and R. D. Nadiad, \"Plate Recognition Using Backpropagation Neural Network and GeneticAlgorithm,\" in Elsevier Procedia Computer Science, 2017.
[10] H. Jorgensen, \"Automatic License Plate Recognition, \"Norwegian University of Science and Technology,2017.
[11] S. Larsson and F. Mellqvist, \"Automatic Number Plate Recognition for Android,\" The author(s) and Karlstad University, 2019.
[12] H. L. Nay, L. A. Yan and K. K. Win, \"Automatic Vehicle License Plate Recognition System for Smart Transportation,\" in IEEE, 2018.
[13] Madhusree, Mondal; Parmita, Mondal, \"Automatic Number Plate Recognition Using CNN Based Self Synthesized Feature Learning,\" in IEEE, 2017.
[14] Surajit, Das; Joydeep, Mukherjee, \"Automatic License Plate Recognition Technique using Convolutional Neural Network,\" International Journal of Computer Applications, vol. 169, 2017