Abstract:
The decay of road infrastructure, particularly potholes and cracks, requires an efficient road condition assessment system to ensure timely maintenance, minimize risks, and reduce accidents. Traditional manual inspections are labor-intensive, time-consuming, and prone to inaccuracies, often resulting in delayed repairs and increased hazards. "RoadRadar – An AI-Based Real-Time Road Condition Assessment System" addresses these challenges using advanced computer vision segmentation models, such as YOLO (You Only Look Once) by Ultralytics, for real-time detection and segmentation of potholes and cracks. RoadRadar operates through two distinct yet interconnected modules. The first is a realtime module powered by Raspberry Pi 4 combined with a Google Coral TPU, designed for continuous on-the-go monitoring and detection of road anomalies, this uses a quantized YOLOv8n segmentation model. The second is a web-based module that allows users to upload videos or images for segmentation and detection of cracks and potholes, this contains multiple YOLO segmentation models ranging from a nano to an extra large YOLOv8 segmentation model. A diverse dataset of over 3000 road images, collected from Pakistani roads under varying weather conditions and environments, has been used to train these models, ensuring their effectiveness in real-world scenarios. This dual-module system increases RoadRadar’s versatility, offering both mobile and stationary applications for comprehensive road monitoring. Facilitating timely and accurate upkeep interventions, it contributes to road safety and infrastructure management.
