Automatic license plate recognition (ALPR) plays a vital role in various applications including traffic surveillance, toll collection, parking systems, and theft prevention. The key goal of ALPR systems is to detect license plates in images and video streams and recognize the characters without extensive human involvement, thereby saving time and resources. While traditional ALPR techniques relied on image processing and machine learning, recent advances in deep learning have led to highly accurate ALPR systems based on convolutional neural networks (CNNs).
However, ALPR remains an active research problem due to factors such as lighting conditions, angles, distances, backgrounds, and license plate layouts differing across countries. Most existing ALPR research has focused on Latin script plates from Europe, United States, Brazil, and China. Limited work has been done for Arabic script plates prevalent in the Middle East.
Arabic car plates, like any other Arabic content-based area, did not receive as much attention in the literature as non-Arabic car plates. Few works are connected to ours, and we strive to summarize and focus on the most current and relevant to ours.
This article aims to provide a comprehensive overview of the Egyptian license plate system, its unique challenges, and the technological advancements being applied to overcome them.
Cairo, the capital of Egypt, is notorious for heavy traffic. Traffic monitoring isn’t an easy task for authorities. They have to read and recognize the license plates of tens of thousands of vehicles. Also, they have to do it as quickly as possible if traffic flows at a rapid pace. In addition to the extreme number of vehicles to be monitored, license plate recognition is further challenged by the wide variety of plates. Egyptian license plates feature small Arabic letters and different background colors to clearly distinguish vehicle types from each other. Moreover, placing plates on vehicles isn’t regulated. This means that they can appear in different positions which makes reading and recognizing them quite challenging.
Read also: Egyptian Adventure
Thanks to one system, the client resolved multiple problems at once. For one, they got a centralized, fully automated system that can handle information coming in from 15 different IP cameras scattered across the city of Cairo. Since the new system heavily utilizes these cameras, there was no need to install special devices either. The software client arranging the acquired data in a straightforward manner solves another issue: data storage. The solution significantly reduced the size of data without compromising quality. The connected servers store data for a year before deletion. Data can be easily searched for a year thanks to the intuitive data storage solution. Speaking of exporting data, the servers are connected to the databases of Egyptian law enforcement authorities. As such, the user can easily perform cross-checks for stolen or wanted vehicles.
Due to the complexity of reading Arabic license plates, Adaptive Recognition’s flagship automatic number/license plate recognition (ANPR) product, Carmen® FreeFlow ANPR engine was chosen. By being on the OCR market for many decades, Carmen® had the time to develop into a true powerhouse that can deal with any license plate regardless of location. However, reading and recognizing Egyptian license plates wasn’t the only reason why Carmen® was chosen for this project. This number plate recognition software is also fully confident with vehicles coming from countries neighboring Egypt. It can also categorize plates based on the country of origin. Plate color isn’t an issue for Carmen® either; the software can easily retrieve license plate characters, regardless of light and weather conditions.
More impressively, Carmen® also solves the issues regarding the position of license plates on the vehicle. In addition to that, this project perfectly demonstrates two of the key features of Carmen® FreeFlow. Firstly, Carmen® has no problems with reading and recognizing tens of thousands of license plates daily. Feedback from the end-user has been overwhelmingly positive. The system perfectly satisfies customer needs, especially regarding the accuracy of license plate recognition, which has been the main concern before the realization of the project. It also solved additional issues such as the daily management of 40,000 vehicles, centralizing data collection, and direct connection to road traffic authorities.
Examples of Arabic/Egyptian license plates confidently recognized by Carmen® ANPR Image and displayed via the interface of 6SS.
Thanks to Carmen®’s versatility regarding IP camera brands, you can integrate a new set of cameras into any system similar to what we just presented, whether or not you have already installed those devices. Governmental facilities are well aware of the importance of monitoring the access and parking of those arriving by car.
Read also: Learn Arabic in Egypt: Guide
A credible License Plate Recognition System has to have the latest algorithmic technology and cloud-based connectivity. The LPR technology can decode license plate automatically in split seconds without the need to have a security guard taking the details manually. By automating this process, all license plates will be recorded accurately despite vehicles types and velocity, and there will be no cases of human error or unsanctioned vehicle entries by guards.
Having LPR Technology that is connected to cloud-based technology like TimeTec LPR Technology, access to information is around the clock in real time manner, making monitoring effective. Deployment of LPR technology can elevate and improve access security of residential and office buildings because the system will only authorize occupants of the building to enter the secured area, and prevent unauthorized vehicles from accessing.
Current Egyptian License Plate System
Egyptian vehicle registration number plates are used for official identification purposes for motor vehicles in Egypt. The current vehicle registration plates, which have been used since 2008, are rectangular in shape and made of aluminum.
Egyptian vehicle registration number plates
Numbers go from 1 to 9 and are chosen randomly. Note : These plate codes do not apply for army, police and diplomatic vehicles. Use of Latin letters and Western Arabic numerals below Arabic letters and Hindu-Arabic numerals was abandoned early due to the unease of reading the Arabic letters and numbers because they were too small. To reduce the risk of confusion on account of the visual similarity between Arabic letters, only a limited number of letters are used.
Read also: Clothing in Ancient Egypt
Vehicles with unpaid customs: Yellow; foreign vehicles that cross into Egypt are required to carry this type of license plate during their stay in Egypt and pay the fees associated with it, as well as purchasing an Egyptian vehicle insurance and receiving a temporary Egyptian ownership title. In most common cases, such as tourists crossing from neighboring countries, the plate is valid for two weeks.
Police vehicles: Dark blue; additionally, the Arabic and English for "Police" replace "Egypt" in the upper bar. Before the introduction of the new alphanumeric plate system in August 2008, Egyptian vehicle registration numbers were purely numeric, and registered by governorate and registration type.
Automatic arabic number plate recognition (ANPR) for Egyptian cars licence plates.
Proposed Methodology for ALPR
In this work, we propose a two-stage automatic license plate recognition (ALPR) system for detecting and recognizing Egyptian car license plates. The proposed ALPR framework relies on an image captured from camera to generate a segmented image. The first phase is the detection of the car plate. This is achieved using YOLOv8 deep learning model, which is currently witnesses a widespread use of the YOLO series models due to their efficiency for different object detection. We have implemented version 8 of this algorithm to ensure a more reliable and precise detection of car plates which will affect the second phase of the system. The second phase involves the recognition of the text and number within the detected car plate. For this task, we employed two different approaches: Easy-OCR tool and a custom CNN deep learning model, which we trained with letters and images from plates dataset with Arabic characters to match the problem at hand.
The proposed ALPR framework
In the training of the models, we perform image preprocessing and data augmentation. Accurate localization of license plates within complex scenes is a key first step. We leverage YOLOv8, the latest version of the popular YOLO model family, as our detector due to its improved speed, accuracy, generalization ability, and robustness to occlusion. YOLOv8 utilizes a transformer-based backbone architecture instead of the prior Darknet models.
In the realm of computer vision, YOLOv8 stands out as a notable iteration, further refining the delicate equilibrium between accuracy and speed. With its core principles rooted in a unique detection pipeline, YOLOv8 excels in directly predicting bounding boxes and class probabilities in a single pass-through input image. This enhances the contextual reasoning capability leading to better detection performance, especially for small objects like license plates.
In the context of car plate recognition, YOLOv8's capabilities become particularly significant. Leveraging a grid-based approach, the algorithm divides the image into cells, each responsible for predicting bounding boxes and class probabilities for objects within its domain. The incorporation of anchor boxes, predefined shapes with varying scales and aspect ratios, enhances YOLOv8's versatility in handling car plates of diverse sizes and proportions.
The strength of YOLOv8 lies in its adept utilization of a deep convolutional neural network (CNN), adopting a variant of the DarkNet architecture. This backbone network seamlessly integrates multiple convolutional and fully connected layers, facilitating effective object classification. The algorithm's prowess in real-time processing is especially valuable in applications like car plate recognition for video surveillance and autonomous driving, where swift and accurate responses are imperative.
Despite its remarkable attributes, YOLOv8 encounters trade-offs between accuracy and speed inherent to its single-shot nature. Challenges may arise in detecting smaller or low-contrast car plates compared to multi-stage methods.
Related Work
Several studies have explored various techniques for car plate detection and recognition. Here's a summary of some recent efforts:
- 2020: An improved Faster R-CNN model with an adaptive attention network and a lightweight Inception V3 model was used for real-time automatic license plate detection of multi-style Egyptian license plates.
- 2021: A deep machine learning-based system compared classical machine learning techniques to newer deep learning approaches for Egyptian vehicle license plate recognition. The study recommended an optimal system utilizing Faster R-CNN for plate detection, connected component labeling for segmentation, and CNNs for character and whole plate recognition.
- 2022: An automatic license plate recognition system for cars in Saudi Arabia used Canny edge detection and OCR to read the English and Arabic text on the plates.
- 2022: A real-time ALPR system for Egyptian license plates using Tiny-YOLOV3 achieved high mean average precision values for both license plate detection and character recognition.
- 2023: A deep learning approach for Iranian license plate recognition used YOLOv4-tiny for detection and CRNN with CTC for character recognition, eliminating the need for character segmentation.
- 2023: The MVSR normalization algorithm was used to improve car license plate recognition, especially for low resolution or rotated images.
- 2023: An efficient automated vehicle license plate recognition system used image processing techniques to achieve high recognition accuracy.
Table 1 summarizes these works mentioning the dataset used, the algorithms implemented, and the highest accuracy reached by each of them.
| Year | Method | Dataset | Accuracy |
|---|---|---|---|
| 2020 | Improved Faster R-CNN | New Arabic license plate dataset | High detection accuracy, 23 FPS |
| 2021 | Faster R-CNN, CNN | Egyptian vehicle license plates | Varies by stage |
| 2022 | Canny edge detection, OCR | Locally collected Saudi dataset | 96% (English), 92% (Arabic) |
| 2022 | Tiny-YOLOV3 | EALPR dataset | 97.89% (detection), 92.46% (recognition) |
| 2023 | YOLOv4-tiny, CRNN, CTC | Created datasets of Iranian license plates | Fast processing times, high accuracy |
| 2023 | MVSR normalization algorithm | N/A | High accuracy, real-time detection |
| 2023 | Image processing techniques | N/A | 94.17% |
Popular articles:
tags: #Egypt
