Developing a Voxel-Based Semantic Segmentation Technology for Human Behavior Representation in an Atypical Architectural Design

This study aimed to develop a semantic segmentation technology for simulating human behavior in an atypical architectural design. This technology extracts potential behavioral patterns from uniquely designed atypical shapes. In contrast to traditional semantic segmentation technologies in architecture, which focus on extracting standardized objects, such as furniture and building components, this study explores the possibility of inducing behavior through atypical architectural shapes. To this end, potential behaviors in an atypical architectural space were extracted and classified using case studies and analyses. Subsequently, a technology was developed to automatically extract and visualize atypical architectural forms aligned with the classified human behaviors. This was accomplished using the application programming interfaces of two commercial tools for atypical architectural design: Rhino and Grasshopper. When designing an atypical building using Rhino, the developed tool, ActoViz, was executed to voxelize it. ActoViz automatically interprets the positional relationship of a voxel and performs semantic segmentation. This technology not only supports architects in intuitively understanding the potential for inducing behavior by executing it in real time during the atypical architectural design process but also emerges as a foundational tool for advancing precise human behavior simulations, extending beyond current path search–based approaches.