An annotated street view image dataset for automated road damage detection.

Area of Science:

• Civil Engineering
• Computer Science
• Artificial Intelligence

Background:

• Road damage poses significant risks to infrastructure longevity and public safety.
• Early detection of pavement distress is crucial for effective road maintenance and repair strategies.
• Street view imagery offers a promising, scalable solution for comprehensive road condition monitoring.

Purpose of the Study:

• To introduce the Street View Image Dataset for Automated Road Damage Detection (SVRDD), the first public dataset of its kind.
• To provide a robust dataset for training and evaluating automated road damage detection algorithms.
• To facilitate advancements in deep learning for pavement condition assessment.

Main Methods:

• Development of the SVRDD dataset using 8,000 street view images from Baidu Maps.
• Annotation of over 20,000 road damage instances across five districts in Beijing.
• Training and performance evaluation of ten established object detection algorithms on the SVRDD dataset.

Main Results:

• Demonstrated the effectiveness of various object detection algorithms in identifying pavement damages using the SVRDD dataset.
• Established a benchmark for performance assessment of automated road damage detection systems.
• Highlighted the potential of street view imagery for large-scale pavement monitoring.

Conclusions:

• The SVRDD dataset represents a significant contribution to the field of automated road damage detection.
• The dataset provides essential data for advancing deep learning models in pavement engineering.
• This resource supports future research and development of intelligent transportation systems for road maintenance.