<![CDATA[Road surface deterioration poses increasing risks to transportation safety and operational efficiency, prompting the need for automated, real-time detection systems suitable for mobile deployment. This study develops and evaluates a YOLOv11-based road damage detection model implemented on Android devices, targeting four defect classes: potholes, cracks, surface waviness, and patched roads. Unlike previous YOLO-based approaches (e.g., YOLOv5, YOLOv8), YOLOv11 integrates a C2PSA attention mechanism and an anchor-free architecture, offering enhanced detection accuracy and computational efficiency critical for resource-limited environments. A total of 2,989 images were collected and annotated from public datasets and organized in standard YOLO format. Model evaluation was conducted using metrics such as AP@0.5, confidence curves, confusion matrix analysis, and latency benchmarks. YOLOv11 achieved high AP@0.5 scores of 0.849 and 0.850 for cracked and patched roads, with a real-time inference latency of 2.6 ms per image and an end-to-end latency of 3.8 ms faster than YOLOv8 in comparable mobile settings. The model was successfully integrated into an Android application, demonstrating robust performance during real-time deployment. However, the system showed reduced accuracy in detecting subtle or low-contrast defects such as shallow potholes and wavy surfaces, often due to background-texture similarity. These limitations suggest the need for improved data diversity and feature refinement. Overall, the findings confirm YOLOv11’s suitability for mobile-based road monitoring, combining speed, accuracy, and lightweight deployment. Future research should address generalization challenges under varied lighting and environmental conditions.]]>
Copyrights © 2025
