Car License Plate Detection Dataset - 3 809 704 Images for Plate Detection and OCR

Overview
The Car License Plate Detection Dataset is a large-scale commercial image dataset produced by Unidata, designed for training and evaluating license plate recognition systems — including LPR, ALPR, and ANPR solutions. The dataset contains 3,809,704 annotated images collected across 93 countries, making it one of the most geographically comprehensive resources available for plate detection and OCR tasks.
Data was collected by parsing videos from real-world road environments and traffic cameras, ensuring diverse coverage of traffic conditions, road types, lighting scenarios, and urban areas across multiple continents.

Dataset Composition
- Total images: 3,809,704
- Countries covered: 93
- Image format: PNG
- Annotation format: CSV
- Tasks supported: Detection, Classification, OCR
- Labeling fields: model, details, plate text, tag, country
The OCR-annotated subset covers 1,200,000+ images with full character recognition markup, enabling training of recognition algorithms for automatic number plate identification across different regions and environments.

Geographic Coverage
The dataset spans all major continents, with the heaviest concentration in Europe:
- Europe — 2,434,844 images
- Asia — 128,627 images
- North America — 84,629 images
- Oceania — 5,265 images
- South America — 4,687 images
- Africa — 2,596 images

Top 10 countries by image count:
1. Russia — 1,299,630
2. France — 222,912
3. Poland — 220,474
4. Germany — 200,062
5. Ukraine — 189,947
6. United Kingdom — 135,718
7. Spain — 125,919
8. Hungary — 107,272
9. USA — 105,136
10. Netherlands — 101,582
Additional well-represented countries include Italy, Belgium, China, Serbia, Belarus, Lithuania, Turkey, Kazakhstan, Austria, and Romania, among others. The dataset also includes plates from Vietnam, Latvia, Czech Republic, Croatia, Norway, Canada, Indonesia, Greece, Georgia, Finland, Estonia, Thailand, Uzbekistan, Israel, UAE, Sweden, Moldova, Azerbaijan, Kyrgyzstan, Armenia, Singapore, Argentina, Brazil, Morocco, Mexico, Japan, South Korea, Saudi Arabia, Egypt, Mongolia, and more.

Vehicle Tags & Plate Categories
Each image is tagged by vehicle type and plate category, enabling fine-grained filtering for specialized training scenarios. Top tags by frequency:
- Truck — 266,654
- Transferred / re-issued plate — 226,645
- Tractor unit — 148,792
- Trailer — 106,685
- Bus — 83,659
- Cabriolet — 73,272
- Oldtimer — 71,065
- New letter combination — 56,709
- Motorcycle — 52,111
-Taxicab — 43,570
-Motorhome — 35,760
- Non-standard plate — 26,572
- Plate for brand or model of vehicle — 26,268
- Abandoned vehicle — 22,309
- Dump truck — 15,975
- Vinyl wrapping — 15,728
- Electric vehicle — 15,150
- Police — 14,185
- Damaged — 11,235
- Tuning — 9,556
This breadth of vehicle and plate types ensures that models trained on this dataset can handle real-world traffic flows with high accuracy — including edge cases such as damaged plates, non-standard formats, and specialty vehicles.

Technical Specifications
Images are provided in PNG format with CSV annotation files. Each annotation record contains: plate text (OCR), country of origin, vehicle model and tag classification.
Collection methodology: frames extracted from road camera footage and video streams captured across diverse road conditions — highways, urban streets, parking lots, and intersections — under varying lighting and weather conditions.

Use Cases
- Intelligent Transportation Systems. Training LPR and ALPR models for real-time vehicle identification in traffic management and monitoring systems. The dataset's coverage of 93 countries supports multi-region deployment without additional data collection.
- Parking & Access Control. Building automated parking management solutions that recognize number plates across different vehicle types, including trucks, buses, motorcycles, and electric vehicles.
- Law Enforcement & Security. Powering ANPR camera systems for vehicle registration tracking, enforcement workflows, and security monitoring in smart city infrastructure.
Autonomous Driving Research. Supplying training data for autonomous vehicles and ADAS systems that require reliable plate detection across diverse road environments and traffic scenarios.
- OCR Model Development. The 1,200,000+ OCR-annotated images with character-level labels and segmentation masks make the dataset directly applicable to training and benchmarking plate recognition algorithms for multi-language and multi-script number plates.

Summary
The Car License Plate Detection Dataset is a production-grade resource for teams building LPR, ALPR, and ANPR systems at scale. With 3,809,704 images spanning 93 countries, 20+ vehicle and plate tag categories, it covers the full spectrum of real-world plate detection challenges. The combination of geographic diversity, traffic condition variety makes it suitable for training high-accuracy recognition models deployable across different regions, road networks, and transportation systems.