딥러닝 기반 아파트 단위 평면 빅데이터 분석 연구

Abstract

When dealing with Korean residential spaces, studying apartments is necessary for socio-cultural context. How did the apartments affect people's lifestyles? Researchers analyzed the spatial structure of the unit plan of the apartment to figure out the answer. This paper aims to inspire residential space designers to create novel and satisfying apartments and support the floor plan retrieval and generation system that considers residents' demand by analyzing Korean apartments' unit plans from various perspectives. We extracted five types of information from the 34,132 unit plans and 5,344 floor plan images and performed four analyses. First, we extracted the following information in the floor plan images using the custom-developed computer-aided image-based system: silhouette, number, and area of rooms, direct and adjacent connectivity. Second, utilizing the extracted information, the following four analyzes were conducted: 1. The analysis of the relationship between the silhouette and the layout 2. The ratio of the area of the rooms by type 3. The relationship between the area and the direct connectivity of each room 4. The modularization pattern of the bedroom. As a result, we obtained objective and advanced results through a multi-faceted survey of large-scale datasets of apartment unit floor plans distinguished from those in the existing literature. The following results were conducted: the layouts of Korean apartment unit plans were not uniform; there was a unique ratio of room area that is mainly preferred in Korean apartments; rooms had their principal uses and standardized roles for each type; and we noticed the intention to secure the privacy of the main bedroom through the modularization of the bedroom. Furthermore, based on the results, we presented the unique spatial characteristics that appeared in Korean apartments.