Automatic license plate recognition with convolutional neural networks trained on synthetic data

Automatic license plate recognition (ALPR) systems have become an important tool to track stolen cars, access control, and monitor traffic. ALPR system consists of locating the license plate in an image, followed by character detection and recognition. Since the license plate can exist anywhere within an image, localisation is the most important part of ALPR and requires greater processing time. Most ALPR systems are computationally intensive and require a high‐performance computer. The present algorithm differs significantly from those utilised in previous ALPR technologies by offering a fast algorithm, composed of structural elements which more precisely conducts morphological operations within an image, and can be implemented in portable devices with low computation capabilities. The present algorithm is able to accurately detect and differentiate license plates in complex images. This method was first tested through MATLAB with an on‐line public database of Greek number plates and was 100% accurate in all clear images, and achieved 98.45% accuracy when using the entire database which included complex backgrounds and license plates obscured by shadow and dirt. Second, the efficiency of the algorithm was tested in devices with low computational processing power, by translating the code to Python, and was 300% faster than previous work.

Automatic license plate recognition (ALPR) systems critically depend on the robust and efficient detection of LPs under unconstrained environmental conditions, including significant viewpoint variations and complex backgrounds. To address these challenges, this paper introduces CAD‐Net, a novel corner‐aware LP detection architecture that combines a computationally efficient ResNet‐18 encoder with an efficient multi‐scale feature decoder for accurate LP corner localization. The decoder aggregates and refines features through group dilated convolutions, coordinate attention, and context gated attention, enabling enhanced focus on semantically salient regions while capturing intricate spatial dependencies. The detected LP corner points enable a polygonal region‐of‐interest alignment strategy for geometric rectification of LP features, which is integrated into an end‐to‐end LP recognition framework named CAR‐Net. Comprehensive experiments demonstrate the efficacy of our method. For LP detection, CAD‐Net attains LP detection rates of 99.9% on CCPD‐Base and 100.0% on AOLP‐RP, with a processing speed of 105 frames per second. For end‐to‐end LP recognition, CAR‐Net achieves state‐of‐the‐art performance on multiple benchmarks, CCPD (98.9%), AOLP‐RP (99.2%), PKUdata (98.5%), CLPD (82.3%), and OpenALPR‐BR (99.1%), while maintaining a real‐time inference speed of 72 frames per second. These results confirm practical viability for deployment in real‐world ALPR systems.

In this paper, we evaluate the influence of detection quality on OCR accuracy in automatic license plate recognition (ALPR) systems. Using a modular pipeline, we compare the performance of three OCR engines - Tesseract, EasyOCR, and PaddleOCR -when fed with outputs from two variants of YOLOv8n object detectors: one autotuned using Optuna and another manually fine-tuned. The fine-tuned YOLOv8n model, optimized for Indian license plates, achieved the best results, allowing PaddleOCR to reach an OCR accuracy of 99 %, marking a 4.16 % improvement over the baseline autotuned model. We use a range of evaluation metrics, including Character Error Rate (CER), Word Error Rate (WER), and Normalized Levenshtein Distance (NLD), to comprehensively quantify OCR performance under different detection scenarios. Our findings highlight the critical role of accurate detection in downstream OCR quality and offer valuable insights for building robust, end-to-end ALPR systems, particularly in real-world deployment scenarios.

With the advances in sensor and data transfer technologies, the usage areas of Automatic License Plate Recognition (ALPR) systems have been expanded in the public and private sectors. In public safety, ALPR systems are used to monitor and control traffic data at both individual and collective levels. To build an efficient sensor network, the locations of ALPR systems should be determined optimally. This study provides an approach to determine optimal locations of ALPR systems that maximize network coverage consisting of two measures: i) vehicle coverage and ii) road coverage. The former represents the daily average vehicle flow whereas the latter stands for the number of road-links covered. The relative importance of vehicle and road coverages are taken into consideration, and optimal solutions under various scenarios are presented. A close neighbor constraint is introduced to avoid inefficient distribution of ALPR systems on the network. A case study with numerical examples designed for two cities in Turkey is provided. The centralized and decentralized solutions are compared against the current state, and the results show that the network coverage increases substantially in the centralized case.

Deep learning has become the main way of the object detection task for autonomous vehicles. Meanwhile, this method typically requires vast amounts of training data to reach their full potential. However, collecting the data from real world and labeling manually is an expensive, time-consuming and error-prone process. Synthetic image has the potential to replace real image for training neural networks, because image creation and labeling annotations are free in this way. For the network trained by synthetic images, the reality gap between real and synthetic images is the main obstacle to use it in the real world. And most previous works are only devoted to generate synthetic images with a good performance on model training, but lack of analysis of the domain gap that affects the performance. This work designs a method of generating the real and synthetic images to analyze and reduce the reality gap between synthetic and real images for car detection. Firstly, this work put one single car with no-background in a random background image to generate real single car and synthetic single car images. In order to further reduce the domain gap in content level, this method keeps the car distribution in synthetic images is similar with the distribution of car in real world. For the purpose of reducing the domain gap in appearance level, the parameters of camera model are same as the camera parameters of image collecting cars and the image is rendered by using the PBRT(Physically Based Ray Tracing) when we generated the synthetic images. Secondly, by training the neural network of instance segmentation with different datasets, the across validation result proves that the reality gap between synthetic and real images is no more than the domain gap between real images. Thirdly, the training results of datasets with different samples diversity show that the diversity of the samples yields better generalization between different datasets for car detection which can effectively reduce the domain gap.

In the past, people usually employ the facial feature extraction and shallow learners such as decision trees, SVM, Naive Bayes, etc. to classify faces of different races. Deep learning usually takes lots of time to train. But with the advances in hardware and new algorithm proposed, the training time problem is gradually alleviated. The deep convolutional neural networks have good effect on images classification. In this paper, we use the deep convolutional neural networks to try to solve the problem of classification faces of different racial origin. Because the convolutional neural networks usually require a huge amount of data for training for good performance, such training set of real racial faces is not available to us. As a result of small set of real racial faces, this study proposes to incorporate synthetic facial images in our training set to sufficiently increase the size of the training set. To the best of our knowledge, this study is the first to propose to incorporate synthetic racial faces to train a deep convolutional neural network to classify real racial faces. We compare the performance of only employ synthetic facial images and mixtures of synthetic and real facial images in the training set. Our experiments show that training with only the real facial images (2,500 images) can achieve 91.25% accuracy in classifying faces of three different race origins. However, the classification when training with a mixture of 2,500 real facial images and 15,000 synthetic facial images can be further improved to 98.5% in accuracy.

Automatic vehicle record-keeping systems have varied applications such as security for car parking facilities, tracking vehicle location and monitoring vehicular traffic. In this paper, we propose a system for vehicle authentication, book-keeping and tracking. The proposed system implements Automatic License Plate Recognition (ALPR) with Convolution Neural Network (CNN) followed by a Gated Recurrent Unit (GRU), which recognizes vehicles and automatically authenticates them with the records on a database to provide information about the vehicle. On the request of the vehicle owner, the system can send vehicle location data through SMS (Short Message Service) notification to the vehicle owner by using a GSM module. Tradition ALPR Systems employ Image Segmentation followed by individual character classification. In this work, we have used a sequence modeling technique that does not require image segmentation. It achieved a character recognition accuracy of 98% and a complete license plate character recognition accuracy of 88%, when trained on a modest data set consisting of 400 images of different fonts.

Automatic license plate recognition (ALPR) is a process of identification of the license plate from an image taken at any signal stop or parking station. The Identified information or the number plate can be used for keeping a check on signal stops or for monetary settlements at parking lots or toll stations. It can also be used around the world for law and administrative works like checking a vehicle if it is registered or not. It also finds its use in safety supervision system. The ALPR can be used upon various types of pictures like color images, black and white images and etc. ALPR may fail a number of times if the quality of the picture is not up to the mark. So, ALPR depend a lot in the image quality, and hence the challenges in its real-life application. It may be used both indoors and outdoors but it must be able to process the images both quickly and successfully in all sorts of environments, be it daytime or night time, sunny or rainy. Our proposed method mainly has three modules: 1) Detection of license plate 2) Segmentation of Characters 3) Text Box Generation. The number plates sometimes may be unclear for the cameras. It may sometimes have dirt and scratches on it and thus increasing the complexities of the ALPR system. Today nationwide ALPR has been adopted by several law and governmental agencies for its efficiency and usefulness. Public safety agencies are using the ALPR system to gather and store relevant information about people and to keep a check on the vehicular activities. It is becoming a significant tool for the Law Enforcement Agencies.

Vehicle license plate recognition has been a much studied research area in many countries. Due to the different types of license plates being used, the requirement of an automatic license plate recognition system is rather different for each country. In this paper, an automatic license plate recognition system is proposed for Malaysian vehicles with standard license plates based on image processing, feature extraction and neural networks. The image processing library is developed in-house which we referred to as vision system development platform (VSDP). The Kirsch edge feature extraction technique is used to extract features from the license plates characters which are then used as inputs to the neural network classifier. The neural network model is the standard multilayered perceptron trained using the back-propagation algorithm. The prototyped system has an accuracy of about 91%, however, suggestions to further improve the system are discussed in this paper based on the analysis of the error.

Deep learning architectures can automatically learn complex features and patterns associated with glaucomatous optic neuropathy (GON). However, developing robust algorithms requires a large number of data sets. We sought to train an adversarial model for generating high-quality optic disc images from a large, diverse data set and then assessed the performance of models on generated synthetic images for detecting GON. A total of 17,060 (6874 glaucomatous and 10,186 healthy) fundus images were used to train deep convolutional generative adversarial networks (DCGANs) for synthesizing disc images for both classes. We then trained two models to detect GON, one solely on these synthetic images and another on a mixed data set (synthetic and real clinical images). Both the models were externally validated on a data set not used for training. The multiple classification metrics were evaluated with 95% confidence intervals. Models' decision-making processes were assessed using gradient-weighted class activation mapping (Grad-CAM) techniques. Following receiver operating characteristic curve analysis, an optimal cup-to-disc ratio threshold for detecting GON from the training data was found to be 0.619. DCGANs generated high-quality synthetic disc images for healthy and glaucomatous eyes. When trained on a mixed data set, the model's area under the receiver operating characteristic curve attained 99.85% on internal validation and 86.45% on external validation. Grad-CAM saliency maps were primarily centered on the optic nerve head, indicating a more precise and clinically relevant attention area of the fundus image. Although our model performed well on synthetic data, training on a mixed data set demonstrated better performance and generalization. Integrating synthetic and real clinical images can optimize the performance of a deep learning model in glaucoma detection. Optimizing deep learning models for glaucoma detection through integrating DCGAN-generated synthetic and real-world clinical data can be improved and generalized in clinical practice.

Traffic control and vehicle owner identification become major problems in Bangladesh. Most of the time it is difficult to identify the driver or the owner of the vehicles who violate the traffic rules or do any accidental work on the road. Moreover, it is very time-consuming for a traffic police officer to physically check the license plate of every vehicle. So, an automatic license plate recognition system is a much-needed solution to solve these problems. The existing Bangla license plate recognition systems are mostly based on character segmentation and these methods are not implemented in real-time. In this study, two separate Deep Convolutional Neural Network (DCNN) models are used to identify the license plate and the characters on the license plate from the real-time video streaming. The first CNN model detects the license plate from the live video of a vehicle on the road. Than it crop the license plate area from the video frames. The cropped frame is then fed into the second CNN to detect the characters on that license plate. The characters are detected as individual objects. After detecting all the characters and numbers on the license plate, they are rearranged according to their position on the plate. To train the proposed model total of 292 images are collected used. Moreover, an open-sourced Bangla handwritten character dataset named BanglaLekha-Isolated is also used to train the model with synthetic character data. The trained model is tested using 18 live videos and 6 still image data. Finally, the proposed methodology gains a 100% precision on detecting the license plate, and 91.67% precision for detecting the characters on the license plate for the given test dataset.

Automatic License plate detection and recognition system is special form of optical character recognition and has been an active research domain in image processing field. However, the accuracy is varied due to different styles of number plates endorsed. Besides, the characters on Malaysia license plate are in one or two lines. Thus, the proposed license plate recognition (LPR) technique of this research is able to achieve the best recognition performance based on Malaysia license plate vehicle registration number characters. This paper presents a study of applying the template matching method for character image recognition. The database of characters and license plate image has been created by collecting images from various type of car. The initial pre-processing involves image enhancement, binarization, filtering and segmenting of license plate. There are 100 license plates that contain 693 characters have been tested, and the result shown that 92.78 % of all characters is correctly recognized. Thus, template matching can be classified as one of the promising algorithm for recognizing Malaysia license plate.

In urban environments, the issue of unauthorized parking in designated no-parking zones persists, leading to traffic congestion and safety hazards. inaccurate license plate recognition, License plate (LP) detection is a crucial task for Automatic License Plate Recognition (ALPR) systems. Most existing LP detection networks can detect License plates, but their accuracy suffers when license plates (LPs) are tilted or deformed due to perspective distortion. This leading to difficulties in identifying vehicle owners. To address this challenge, this project present TraceMe, a predictive system utilizing advanced machine learning algorithms. The system employs YOLOv8 for efficient object detection, focusing on identifying vehicles in no-parking zones, and Tesseract OCR for accurate license plate recognition. The extracted license plate information is then processed by a machine learning model trained to predict the owner of the vehicle. The proposed system involves collecting and annotating a diverse dataset, training YOLOv8 and LPRNet model for vehicle number plate detection, utilizing Tesseract OCR for license plate extraction, and implementing a machine learning model for owner identification. Real-time processing and integration with surveillance systems allow for immediate identification of unauthorized parking incidents. The system generates alerts or notifications, aiding law enforcement in enforcing parking regulations.TraceMe not only provides a technological solution to mitigate unauthorized parking but also contributes to improved traffic management and public safety.

License plate recognition (LPR) system is an important system in our life. LPR is an image processing and a character recognition system that used to recognize any car from the others. An automatic license plate recognition system for the three different Iraqi car license plates was proposed in this paper. Differentiating between the three styles were done depending on the plate size. An optical character recognition (OCR) is used with correlation approach and templates matching for plate recognition by segmenting each number, character and word into sub images. The software used is MATLAB R2014a. The algorithm is successfully constructed with sample of images correctly identified.

Automatic license plate recognition is a technology that relies on optical character recognition methods used on images for reading license plate text. These techniques are, in general, used with image processing techniques, whether the system is composed of them entirely or in addition to deep learning techniques and neural network models. Such a system can be used by CCTV cameras, smart cameras, and other such devices, for tasks in areas like law enforcement, toll collection systems, or parking lot ticket systems. These systems can be used to capture images, locate license plates and recognize the text from the license plate. Depending on the region, post-processing can be conducted to further refine the recognized text based on a set of predefined rules, such as the position of letters and numbers in the license plate string. This work aims to create an automatic license plate recognition system with the use of image processing techniques, mainly from the OpenCV python library. The system will attempt to recognize license plate text from vehicles parked in parking lots. This can be useful for people, who do not remember the place where they parked their vehicle, or in other areas such as law enforcement. For large parking lots, there are generally camera systems already in place, in this case, the system should be able to work with multiple angles and views of the parking lot to get results with the highest accuracy ratings.