부산대학교 도서관

The safety inspection for guaranteeing the safety and functionality of buildings is an effective measure to prevent hazardous accidents caused by aging and inadequate management of buildings. However, there is a need for improvement as the number of inspectors who perform on-site inspections regularly is limited, and the inspection results can be significantly influenced by the skill level of inspectors and subjective evaluations. In this paper, we propose a cost-effective and objective safety inspection automation system to overcome the limitations and constraints of current building safety inspections. The proposed technique focuses on detecting indoor cracks in concrete structures and consists of on-site investigation using autonomous drones and crack detection using deep learning vision. In this paper we propose the use of low-power drones to reduce the cost of safety inspection activities and propose algorithms for autonomous on-site investigation in indoor environments without shadow areas through altitude maintenance and fixed-distance flight capabilities and collecting images along with their location information. To minimize the inspection time, the proposed system utilizes the DRAEM deep learning model, which allows effective crack detection with a small number of images, and proposes a threshold-based pixel masking technique to improve crack detection accuracy. The proposed system has been validated through tests in real-world environments.