Project: Disperse Version: 0.1 Status: Draft Date: 2025-12-15
Large public gatherings in urban environments carry significant crowd-safety risks. Event planners, urban designers, and emergency responders lack accessible tools to visualise crowd flow and identify risk hotspots from spatial data alone — without requiring on-the-ground sensors or expensive simulations.
Disperse is a browser-based tool that lets users click any location on a satellite map, mark obstructions (focal points, vehicles, barriers), and receive an AI-generated crowd density heatmap overlaid on the satellite image. The system runs road segmentation followed by a crowd prediction model to estimate pedestrian flow risk.
- Interactive satellite map (Mapbox) — click to select analysis location
- Draw obstructions on canvas: focal points, vehicles, barriers
- Road segmentation via Keras model
- Crowd density heatmap prediction via Keras model
- Heatmap overlaid on grayscale satellite image
- Export results as image / PDF
- Multiple location comparison view
- Adjustable obstruction parameters (severity, scale)
- Real-time crowd data ingestion
- 3D or building-interior analysis
- Mobile app
- How should model confidence thresholds be surfaced to users?
- Should the backend support GPU inference for faster predictions?
- What is the intended deployment environment (local-only vs. hosted)?