하이피델리티 시뮬레이션과 생성형 AI기반의 적응형 건설안전 AI합성 데이터 재생성 시스템 개발
과학기술정보통신부 2026.03 - 2030.12
This research project aims to develop an adaptive synthetic data generation system for construction safety AI by integrating high-fidelity simulation and generative AI tenenologies. The project focuses on addressing the structural shortage of real accident data in construction sites and the domain gap between synthetic and real-world environments. The proposed framework introduces a closed-loop learning architecture (Real → Simulation → Real) in which AI model errors are analyzed, new training data are generated accordingly, and the model is iteratively improved.[연구 목표]
건설안전 AI설능 향상을 위한 적응형 합성데이터 생성 기술 개발
하이피델리티 시뮬레이션 기반 위험 시나리오 데이터 생성 환경 구축
생성형 AI기반 합성 데이터 현실성 보정 기술 개발
AI 오류 패턴 분석을 통한 데이터 재생성 및 재학습 엔진 개발
Real-Sim-Real순환 구조 기반 적응형 데이터 생성 파이프라인 구축
실환경 테스스베드를 활용한 성능 검증 및 기술 안정화
[연구내용]
High-Fidelity Construction Simulation Environment
Development of a realistic virtual construction environment using game engines (Unity/Unreal) combined with BIM and reality-captured data. The environment reproduces complexx accident scenarios involving workers, equipment, and environmental conditionss.
Generative AI-based Realsm Enhancement
Application of diffusion and GAN-based models to improve the visual realism of synthetic data and reduce domain gaps between and real-world data.
Adaptive Synthetic Data Regeration Engine
Development of a feedback-driven system that analyzes AI model errors and automatically generatives new thing data focusing on vulnerable conditions.
Integrated Verification and Development Platform
Construction of a uified platform that integrates simulation, generative AI, and performance feedback to generate, evaluate, and optimize training datasets for construction safety AI.
[연구 지원 기관]
과학기술정보통신부
[연구 기간]
Mar. 2026 - Dec.2030
