<![CDATA[This study developed a real-time road damage detection system on an edge device to address the inefficiency and subjectivity of manual monitoring. Following the Engineering Design Process (EDP), the system uses a YOLOv11 model—trained in PyTorch and converted to ONNX for optimization—to classify four types of road damage on a Raspberry Pi 5, integrated with a GPS module for automatic coordinate recording. Field evaluation showed that after optimization, the system achieved an average processing speed of 3.66 FPS with an inference latency of 277.69 ms, while maintaining efficient resource usage (45.06% RAM, 65.05% CPU load) and an mAP@50 of 0.554. These results demonstrate the feasibility of an autonomous, accurate, real-time road damage detection system on a low-cost edge computing platform, producing geospatial data ready to support operational decision-making in road maintenance.]]>
Copyrights © 2026
