Analysis and Design of Discrete Crowd Behavior in Architecture

This research investigates discrete crowd behavior through a multi-stage analytical pipeline evolving from linear to circular cyber-physical system. The pipeline comprises four stages: Data Extraction (converting surveillance footage to 3D trajectories using YOLOv5/YOLOv9 + StrongSORT), Simulation (generating synthetic crowd behavior in Unreal Engine with parameters like interest scores and density responses), Visualization (rendering heatmaps and trajectory maps), and Analysis (comparing real and simulated data for validation). Inspired by Professor Ikeda's framework, this was reconceptualized as a circular cyber-physical system where crowd phenomena in physical space are captured as monitoring data, integrated through visualization in cyberspace, analyzed for patterns, and fed back into simulation—creating reciprocal feedback loops between physical observation and digital modeling.

This research has established robust crowd behavior observation and simulation systems but lacks feedback mechanisms to physical space, creating an incomplete cyber-physical loop. Current architectural interventions—movable walls, robotic partitions, digital signage, and billboards—demonstrate technical capabilities but suffer from overt control, scripted behaviors, limited social affordance, and heavy infrastructure requirements. This creates an opportunity to develop more empathetic, socially literate interventions that signal and invite rather than instruct and impose.

The Sakura effect—using planted individuals who influence behavior through presence and social mimicry—offers a non-intrusive intervention model that aligns with DBA research's focus on discrete crowd behavior. Unlike conventional methods, it embeds behavioral cues within the spatial fabric itself rather than imposing external control. It resolves key limitations through three advantages: no directiveness, no infrastructural burden, and social naturalness, completing the cyber-physical loop through embodied participation.

Human Sakura agents face ethical, performance, and scalability challenges, prompting the use of robotic visitors—socially present, mobile entities that influence through presence rather than instruction. Operating under experimental control and ethical transparency, they offer four advantages: real-time adaptability to changing conditions, feasible crowd experience improvement without infrastructure, efficient space utilization by activating underused areas, and rich behavioral data from embodied perspectives unavailable to passive surveillance systems.

This project analyzes the complex and discrete behavioral patterns of crowds in urban public spaces where pedestrians move freely, examining their relationship to spatial form. By using automatically recognized data from surveillance camera footage, it aims to refine behavior simulations based on dynamic visual information models and apply the results to spatial design.

Visualization of pedestrian flow using surveillance camera data

Discrete algorithm–based crowd simulation

This project aims to analyze the complex and discrete behavior of crowds in urban public spaces with free movement in relation to spatial shape. We aim to refine behavior simulations using dynamic visual information models based on automatic recognition data from surveillance camera images, and apply them to spatial design.

Creating a visual Activity heatmap in Grasshopper based on the captured data.

Data Collection and Visualization Flowchart and the used applications. The Computer Vision data is visualized in 3D as a colored heatmap

Overlaying the video camera footage and the reconstructed 3D model