| Title |
YOLO-based Automatic Axle Event Detection for High-Speed Weigh-In-Motion Systems |
| Authors |
박준영(Park, Jun Young);김종우(Kim, Jong Woo);조윤범(Jho, Youn Beom);정영우(Jung, Young Woo) |
| DOI |
https://doi.org/10.12652/Ksce.2026.46.1.0095 |
| Keywords |
자동 이벤트 검출; 고속축중기; 딥러닝; 객체 탐지 Automated event detection; High-speed Weight-In-Motion; Deep learning; Object detection |
| Abstract |
Accurate axle event detection in High-Speed Weigh-In-Motion (HS-WIM) systems is a fundamental requirement for the reliable estimation of individual axle loads and the accurate identification of overweight vehicles. However, conventional threshold-based axle event detection approaches are inherently sensitive to sensor offset drift, environmental noise, and limited responsiveness to low-weight axle loads, which frequently results in false and missed detections. To address these challenges, this study presents an automated axle load segment detection method in which time-series axle load signals acquired from HS-WIM systems are converted into two-dimensional graph representations and analyzed using a YOLOv8-based object detection framework. By learning the characteristic shapes of axle load segments directly from signal representations, the proposed method enables robust axle event detection without reliance on predefined threshold values or absolute signal magnitudes. The proposed model was trained and evaluated using real-world measurement data collected from a public road testbed under actual traffic conditions. The validation results demonstrated high detection performance, with Precision, Recall, and mAP@0.5 values of approximately 0.999, 1.000, and 0.995, respectively. Furthermore, an axle load segment detection accuracy of approximately 95 % was achieved on an independent test dataset. These results indicate that the proposed YOLO-based axle event detection method provides a reliable and practical solution for axle load measurement and overweight vehicle identification in HS-WIM systems. |