딥러닝 기반 이미지 자동인식 기술을 활용한 사무집기 자동인식과 정보관리 시스템과의 연동방안 (Approach to Interworking between the Deep Learning-based Object Detection of Office Furniture and Appliances and Information Management System)

서지정보

· 발행기관 : 한국퍼실리티매니지먼트학회
· 수록지 정보 : 한국퍼실리티매니지먼트학회지 / 12권 / 2호 / 73 ~ 80페이지
· 저자명 : 김진성, 송재열, 김하얀, 이진국

초록

This paper aims to propose an approach to auto-recognition of office objects using non-professionally taken indoor pictures based on the deep learning technique. Recently, artificial intelligence(AI) has been applied in broad fields of industry. Especially, the deep learning-based image recognition and object detection technologies are attaining a high level of accuracy close to human capability. In addition, its source technology has been open to the public, allowing people to use it according to their own purpose. According to the AI trend, deep-learning also has been actively studied in the field of building technology. This paper describes an approach to utilizing image recognition at the phase of facility management (FM) and the process for verifying the possibility of adapting up-to-date technologies in FM fields. The procedure of this study includes data collection, data preprocessing, deep learning-based model training and office indoor image auto-recognition test for detecting office objects. The target office objects include office desks, office chairs and electronic devices, which are most commonly seen in the office space. Over 200 indoor images including target office objects are collected from the domestic office furniture firms' catalog for model training and testing. The results of test with new indoor image are analyzed as the factor of accuracy and similarity of detected office objects. This paper also depicts the potential to apply the auto-recognition technique with BIM(Building Information Modeling) for supporting CAFM (Computer-Aided Facility Management).

영어초록

This paper aims to propose an approach to auto-recognition of office objects using non-professionally taken indoor pictures based on the deep learning technique. Recently, artificial intelligence(AI) has been applied in broad fields of industry. Especially, the deep learning-based image recognition and object detection technologies are attaining a high level of accuracy close to human capability. In addition, its source technology has been open to the public, allowing people to use it according to their own purpose. According to the AI trend, deep-learning also has been actively studied in the field of building technology. This paper describes an approach to utilizing image recognition at the phase of facility management (FM) and the process for verifying the possibility of adapting up-to-date technologies in FM fields. The procedure of this study includes data collection, data preprocessing, deep learning-based model training and office indoor image auto-recognition test for detecting office objects. The target office objects include office desks, office chairs and electronic devices, which are most commonly seen in the office space. Over 200 indoor images including target office objects are collected from the domestic office furniture firms' catalog for model training and testing. The results of test with new indoor image are analyzed as the factor of accuracy and similarity of detected office objects. This paper also depicts the potential to apply the auto-recognition technique with BIM(Building Information Modeling) for supporting CAFM (Computer-Aided Facility Management).

참고자료

· 없음