PY-CrackDB: A pavement crack dataset from paraguayan roads for context-aware computer vision models

Area of Science:

• Civil Engineering
• Computer Science
• Data Science

Background:

• Road maintenance relies on accurate defect detection.
• Early identification of pavement distress, especially fine cracks, is critical for cost-effective repairs.
• Existing datasets may not adequately address the specific needs of early-stage crack detection in asphalt pavements.

Purpose of the Study:

• To introduce PY-CrackDB, a novel dataset of asphalt pavement images.
• To provide a resource for developing context-aware artificial intelligence systems for road maintenance.
• To specifically support the detection of fine cracks (< 3 mm wide) for early-stage intervention.

Main Methods:

• Collected 569 asphalt pavement images from national routes in Paraguay.
• Annotated images for the presence or absence of cracks, with a focus on fine fissures.
• Ensured data quality through standardized collection and verification by civil engineering professionals.

Main Results:

• PY-CrackDB contains 369 images with cracks and 200 without.
• The dataset is suitable for training AI models for semantic segmentation and early defect detection.
• Annotations were rigorously cross-reviewed for accuracy.

Conclusions:

• PY-CrackDB is a valuable resource for advancing AI in road maintenance.
• The dataset facilitates the development of preventative maintenance strategies through improved crack detection.
• Focusing on fine cracks enables earlier and more efficient road repairs.