스마트건설을 위한 영상정보 처리와 객체 탐지 연구 = A Study on Image Processing and Object Detection for Smart Construction

다국어 초록 (Multilingual Abstract)

This study proposes an integrated system for automated safety management that automatically identifies safety hazards at construction sites using drone imagery. The construction industry remains labor-intensive and is characterized by high-risk environments where major accidents such as falls, collapses, and collisions occur frequently. In particular, excavation boundaries and blasting operations at earthwork sites are significant safety hazards, yet their management has largely relied on visual inspection, limiting effective safety control. To address this issue, this study introduces a safety management system that detects excavation boundaries and on-site individuals using drone imagery.
In the boundary detection process, initial outlines were extracted using Canny edge detection, followed by morphological closing to ensure boundary continuity. Principal Component Analysis (PCA) was then applied to quantify the predominant orientation of irregular excavation boundaries. Experiments were conducted at different flight altitudes (50 m, 70m , 100 m), camera angles (60°, 90°), and flight directions (north-south, east-west). Results showed that vertical imagery at 90° produced approximately 2% error compared with GNSS-RTK (Global Navigation Satellite System-Real-Time Kinematic) survey data, while at 50m altitude, the error was only 0.05 m compared to the reference value. These findings demonstrated higher accuracy than conventional edge detection methods and confirmed applicability to practical tasks such as earthwork volume estimation and slope stability assessment.
For human detection, the YOLOv5 model was applied. Initial training used the VisDrone2019-DET public dataset, followed by transfer learning with approximately 1,600 real construction site images collected for this study, thereby enhancing field applicability. As a result, the YOLOv5l model achieved mAP@0.5 = 0.455 and maintained robust detection under diverse conditions and complex backgrounds. Resolution-based analysis revealed that with 1280-pixel input, YOLOv5l yielded a detection rate of 85.7%, YOLOv5m achieved 83.3%, and YOLOv5s achieved 85.7%. At low resolution (640 pixels), detection rates decreased to 55-71%. These results provide practical guidelines for determining the optimal image resolution and model configuration in drone-based safety monitoring.
In summary, this study developed an integrated safety management system that combines boundary detection and human detection, thus distinguishing it from previous research. Furthermore, when linked with BIM (Building Information Modeling) and digital twin technologies, the proposed system can support verification of deviations from design plans, identification of unauthorized access to hazardous zones, and evaluation of safety compliance, thereby demonstrating its potential to evolve into a smart construction safety management platform. Looking ahead, if sensor fusion, lightweight algorithm design, and institutional framework development advance in parallel, the proposed system could become a core safety management technology in the era of smart construction.

목차 (Table of Contents)

• I. 서 론 1
• 1.1 연구배경 및 목적 1
• 1.2 연구내용 2
• 1.3 선행연구 동향 3
• 1.4 본 연구의 방향과 차별성 6
• II. 스마트건설 이해 및 필요성 8
• 2.1 스마트건설 개념 8
• 2.1.1 스마트건설의 정의 8
• 2.1.2 스마트건설을 활용한 건설안전의 정의 17
• 2.2 스마트건설 현장 적용 27
• 2.2.1 스마트건설 기술 분류 27
• 2.2.2 스마트건설 기술 적용 사례 34
• 2.3 스마트건설 한계 및 개선목표 46
• 2.3.1 스마트건설 기술 적용의 한계 46
• 2.3.2 스마트건설 확산을 위한 개선 전략 50
• III. 영상정보 취득 및 처리 방법 52
• 3.1 영상정보 취득을 위한 공간정보 개요 52
• 3.1.1 공간정보와 측량의 기초 52
• 3.1.2 주요 측량 기술과 장비 55
• 3.1.3 정사영상과 사진측량의 개념 57
• 3.2 현재 사용중인 영상 취득 59
• 3.2.1 CCTV(closed-circuit television) 영상 취득 60
• 3.2.2 스마트폰 영상 취득 62
• 3.2.3 스마트건설 안전장비 영상 취득 63
• 3.3 드론사진측량을 이용한 영상 취득 및 처리 65
• 3.3.1 드론의 개요 65
• 3.3.2 영상처리를 위한 드론사진측량 73
• 3.4 항공영상의 기하학적 특성과 정사보정 81
• 3.4.1 항공사진의 기하학적 원리 81
• 3.4.2 촬영 요소 상호관계와 해상도 84
• 3.4.3 항공사진의 촬영 유형과 왜곡 특성 87
• 3.4.4 정사보정과 공간정보 정합 91
• 3.5 에지 검출을 위한 영상처리 이론적 배경 93
• 3.5.1 디지털 영상처리 개요 93
• 3.5.2 영상강조의 개념 97
• 3.5.3 에지 검출 알고리즘 이론 102
• 3.6 영상정보 AI(Artificial Intelligence) 처리 방법 109
• 3.6.1 컴퓨터비전 객체 탐지 알고리즘의 개요 109
• 3.6.2 객체 탐지 알고리즘 모델 분석 113
• IV. 건설공사 개선을 위한 영상정보 분석 123
• 4.1 에지 검출로 토공사 개선 123
• 4.1.1 드론 및 스마트폰 등으로 영상 취득 123
• 4.1.2 영상정보 에지 검출 130
• 4.1.3 토공사 경계와의 실제거리 분석 134
• 4.2 AI 탐지로 안전사고 예방 155
• 4.2.1 공개 데이터셋 활용 평가 159
• 4.2.2 구축 데이터셋 활용 평가 196
• 4.2.3 건설공사 활용 가능성 평가 222
• V. 결 론 224
• 5.1 연구결과 224
• 5.2 연구의 한계점 226
• 5.3 발전방향 및 의견제시 227
• 참고문헌 230
• Abstract 267