1
- https://doi.org/10.5302/j.icros.2025.24.8009
탐지 거리 및 시선각 변화에 따른 균열 검출 성능 분석
- Jul 31, 2025
- Journal of Institute of Control, Robotics and Systems
- Sung-Min Hong +3 more
In disaster environments, mobile robot–based crack detection has received growing attention for autonomous structural inspection. However, most existing models are evaluated from fixed, frontal viewpoints, with limited investigation into performance degradation caused by variations in detection distance and viewing angle. This study quantitatively analyzes the detection performance of a YOLOv8-based model using real crack images captured under varying distances (60–210 cm) and viewing angles (0 °, ±30 °, ±45 °, and ±60 °). Evaluation metrics include mean average precision, intersection over union, and confidence score. The results indicate that changes in viewing angle have a greater effect on detection performance than changes in distance. In particular, performance drops sharply at oblique angles of ±60 °, with detection failures observed under certain conditions. These findings highlight the limitations of current crack detection models regarding rotational invariance and robustness to perspective distortion. They underscore the need for viewpoint correction techniques and more diverse training datasets to improve reliability in real-world disaster scenarios.