Summary

Over the past few years, we have developed and tested computational methods that use street-level imagery and AI (computer vision + NLP) to extract interpretable cues (e.g., lighting, visibility, neglect, activity proxies) and generate scalable 'perceived safety/fear of crime' indicators that align with people’s ratings and correlate with crime patterns.

A key gap is the lack of scalable, consistent tools that connect “what feels unsafe” in the built environment to specific cues and practical interventions, especially beyond approaches that rely on recorded crime statistics alone. Our work aims to fill this by producing an interpretable perceived-safety layer from street-level imagery, including human-readable explanations and uncertainty, and validating it through user group testing. This matters because it can help decision-makers prioritise place-based improvements more fairly and transparently, and it enables more credible evaluation of whether environmental changes actually improve perceived safety.

The programme’s rapid delivery model, emphasis on co-created research, and opportunity to join a wider community of interest align strongly with our ambition.