Verification of the effect of UAV shooting altitude on detection accuracy in asphalt crack detection using deep learning Eita Uotani(a), Mitsuharu Tokunaga(b)
a)Graduate Student, Department of Civil and Environmental Engineering, Kanazawa Institute of Technology, Japan
b)Professor, Department of Civil and Environmental Engineering, Kanazawa Institute of Technology, Japan
Abstract
Accurate identification of early deterioration in asphalt pavement and the implementation of appropriate preventive maintenance are critical issues in reducing life cycle costs (LCC). However, at present, regular inspections are conducted in only about 80% of prefectures and approximately 20% of municipalities, and standardized methods for data collection and management remain underdeveloped. This study aims to improve the efficiency and quality of pavement inspections by combining unmanned aerial vehicles (UAVs) with artificial intelligence(AI).A machine learning model was developed to automatically detect cracks from UAV-captured images of asphalt pavement. The study evaluated how detection accuracy is affected by imaging conditions such as flight altitude, camera angle, and lighting, as well as by model architecture and training methods. Analysis under multiple conditions revealed a tendency for detection accuracy to decline as flight altitude increases. However, the introduction of model optimization techniques and data augmentation was found to effectively suppress this decline.As a result, high-accuracy crack detection was demonstrated to be feasible even from relatively high altitudes, enabling more extensive and efficient pavement inspection. This approach is expected to contribute to the advancement of road management practices.
Keywords: Deep Learning, YOLO, Asphalt Crack, UAV altitude
