DORIE: Dataset of Road Infrastructure Elements

DORIE is a novel, high-resolution object detection dataset specifically curated for real-time patrol vehicle monitoring and automated road infrastructure inspection. This dataset addresses a significant gap in existing driving benchmarks by focusing explicitly on safety-critical static infrastructure elements that are often underrepresented. The dataset is designed to facilitate the training and evaluation of object detection models for classes crucial to road safety and maintenance.

Key Data Characteristics

• Total Images: 938 manually annotated images.
• Total Object Instances: Over 6,800 object instances.
• Annotation Type: Manually annotated with YOLO bounding boxes.
• Resolution: High resolution at 5,568 x 4,872.
• Collection Site: Collected along the A2 motorway in Spain (from Madrid to the boundary between the provinces of Guadalajara and Soria). The road section is maintained by Acciona Construction S.A.
• Collection Setup: Acquired using a GoPro HERO13 Black mounted on the hood of a patrol vehicle, simulating the viewpoint of a typical dash camera.
• Inherent Challenges: The dataset includes real-world challenges such as small object size, class imbalance, varying lighting and weather (including high sun glare), and occlusion by other vehicles or environmental elements.

Annotated Classes

DORIE features ten safety-critical categories, including both static infrastructure and dynamic traffic participants:

• Bollard: Roadside safety posts or poles. 
• Delineator: Reflective roadside markers used to guide traffic.
• Prohibitory sign: Standard traffic prohibition signs (circular signs with a white background and red border). 
• Danger sign: Warning signs indicating hazards ahead (triangular signs with a white background and red border). 
• Mandatory sign: Signs prescribing specific driving actions (circular signs with a blue background). 
• Other sign: Additional traffic signs not falling into the previous groups. 
• Guardrail: Roadside safety barriers.
• Road: The drivable road surface. 
• Car: Passenger vehicles visible on the road. 
• Truck: Larger vehicles such as lorries and freight trucks.

Dataset Split

The dataset is split into training, validation, and test subsets to facilitate standardized model development and evaluation:

• Total Instances: 6,852
• Training Set: 4,408 instances
• Validation Set: 1,081 instances
• Test Set: 1,363 instances

Funding

This work was supported by the European Union’s Horizon 2020 Research and Innovation Programme under grant agreement No 955356 (Improved Robotic Platform to perform Maintenance and Upgrading Roadworks: The HERON Approach)