http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
평면 스케치 딥러닝 학습모델 구축과 공간디자인 활용 - 평면 스케치 인식 기반 설계초기 BIM 모델 자동생성 모듈 개발 중심으로 -
조단규 ( Cho Dahngyu ),이진국 ( Lee Jinkook ) 한국공간디자인학회 2021 한국공간디자인학회논문집 Vol.16 No.3
(Background and Purpose) As the construction industry requires the use of technologies related to the fourth industrial revolution, the field's data-based digital technology is highlighted. Currently utilized data in architecture exists in various forms, making it difficult to manage and use the accumulated design information for other reasons. Therefore, this study proposes developing and utilizing technologies that recognize floorplan sketches, one of the analog media, and convert them into digital media, BIM, that computers can understand. (Method) This study is conducted in two parts: 1) floorplan sketch recognition, 2) utilization of recognized floorplan sketches. In the floorplan sketch recognition phase, we propose an approach that applies Generative Adversarial Network-based deep learning techniques to transform human-drawn floorplan sketch styles into one unified form. Specifically, to recognize the floorplan sketches, the deep learning model learns various forms of floorplan sketches and the styles it needs to transform in pairs. The consistent style has essential architectural elements labeled on it : wall, floor, door and window objects. The study uses apartment floorplan images as learning data for floorplan sketch recognition deep-learning model. The apartment floor plan dataset provided by the Ministry of Land, Infrastructure and Transport Seumteo(e-ais) consists of scanned floorplan images of apartments, houses, and row houses in Seoul. In the phase of the utilization of perceived floorplan sketches, BIM models were generated based on the recognized floorplan sketches. They were subdivided and vectorized by object to extract coordinates, and this enabled BIM models to be obtained by using them as input values for Revit API. (Results) The floorplan sketches labeled with essential architectural elements were the output for the floorplan sketch recognition module, and it was successfully printed out. For the analysis of the study, the predicted image was compared by the ground truth image and showed success. For the second phase, the BIM model prototype made on the Autodesk Revit platform created a fast and accurate 3D BIM model when the recognized floorplan sketches were given input. The created model is low in detail, but it contains the geometry information and the property information, making it an easier way for future use. (Conclusion) This research and development is a sub-development of the overall intelligent design system. It identifies whether computers can recognize one of the design information, floorplan sketches as human beings, and suggests its potentials. Subsequent studies are believed to obtain a BIM model with a high level of detail if recognition is achieved by scaling out of the current recognizable target range.