AI surveillance framework detects suicide risk in metro stations with 83.2% accuracy

A team of researchers including Safwen Naimi, Wassim Bouachir, Guillaume-Alexandre Bilodeau, and Brian Mishara have introduced the first interpretable framework for Suicide Risk Assessment (SRA) in metro stations using AI-powered video surveillance. Published on arXiv and accepted at IJCAI 2026, the framework formalizes SRA as a distinct computer vision task. Unlike prior approaches that focus on isolated subtasks or attempt direct intent inference, this system integrates person tracking, activity recognition, semantic segmentation of the platform, and trajectory-driven risk heatmap modeling. By accumulating behavioral evidence over time and space, it jointly reasons about passenger behavior, spatial context, and temporal dynamics.

The framework was benchmarked on real surveillance data and achieved 83.2% ROC-AUC, demonstrating its practical viability. The interpretable design is crucial for deployment in sensitive public safety contexts, as it allows human operators to understand the reasoning behind risk assessments. The authors highlight that this work not only advances AI for social good but also opens new research directions for interpretable AI systems in safety-critical applications. Metro stations, where early identification of high-risk situations can enable timely intervention, stand to benefit significantly from such technology.