Background
Urban public spaces are essential for enhancing quality of life — these spaces constitute a critical component of sustainable cities. However, a key challenge arises from the disconnect between how planners/designers create these spaces and how people actually use them. Conventional participatory methods can bridge this gap but are resource-intensive, while immersive VR/AR tools remain costly and inaccessible. Recent advances in generative AI enable rapid visualization but still demand technical expertise, limiting public engagement.
Research Questions
This study aims to address the following questions through an interactive platform that integrates computer vision, graph theory, and generative AI:
1.What is the relationship between the spatial characteristics of urban public spaces and human activity?
2.Compared to designs generated solely from text prompts, can activity-based design generation more effectively translate early-stage public input into spatial proposals that are both contextually and activity-appropriate?
Methodology
The methodology uses semantic segmentation, constructs graphs linking annotated activities to spatial elements, and identifies activity–space patterns. These patterns are then converted into semantic spatial masks and refined through diffusion models with spatial conditioning, producing contextually appropriate design proposals.
Key findings
Our graph-based analysis shows that activity patterns are closely linked to spatial features —such as linear/open spaces, vegetation richness, and spatial depth —which correspond to different activity intensities and needs (e.g., pavements support walking/jogging, grass fields enable diverse uses, shaded areas encourage rest). More importantly, designs generated from activity-specific graph patterns were more contextually appropriate than those derived solely from text prompts.
Potential impact
Overall, the study contributes a novel method to predict activity patterns from scene composition and to generate activity-based, inclusive design proposals. The proposed platform also serves as an early-stage design tool for enhancing public participation and bridging the gap between professional expertise and community needs in sustainable urban development.