An Intelligent Road Damage Detection Method Based on YOLOv7-Tiny Framework.

Road health condition heavily relies on daily road maintenance work. Intelligent road damage detection method is considered one of promising techniques. The current road damage detection (RDD) methods face challenges such as high computational complexity, slow detection speed, low detection rate, and difficulty in meeting the requirements of real-time detection on mobile devices. To address these issues, this paper proposes an intelligent road damage detection system based on the You Only Look Once version seven-tiny (YOLOv7-tiny). We collect a road damage dataset named RDD and trained it using YOLOv7-tiny and other similar object detection networks. YOLOv7-tiny achieved a 55% improvement in mAP@0.5 compared to Retinanet, and a 51% improvement in F1 score compared to Faster-RCNN. The experimental results show that our proposed RDD based on YOLOv7-tiny exhibits better performance and is more suitable for deployment on lightweight edge devices for road damage detection.