Suicide Risk Assessment from AI-powered Video Surveillance: An Interpretable Framework for Prevention in Metro Stations

Researchers have developed an interpretable AI framework for assessing suicide risk in metro stations using surveillance video analysis, achieving 83.2% ROC-AUC by combining person tracking, activity recognition, and trajectory analysis. This work addresses a critical public health challenge by enabling early identification of high-risk situations that could facilitate timely intervention.

This research represents a significant advancement in applying computer vision and machine learning to suicide prevention in high-risk public spaces. The framework moves beyond simple behavior classification to create a comprehensive risk assessment system that synthesizes multiple data streams—person tracking, activity recognition, platform geometry, and temporal dynamics—to identify individuals at imminent risk.

The work addresses a genuine public health crisis; metro stations worldwide experience preventable suicides annually. Traditional monitoring relies on human observation, which is inherently limited by fatigue, inconsistency, and the unpredictability of critical moments. By automating risk detection, the system could enable security personnel to intervene before tragedy occurs, making it potentially life-saving infrastructure.

However, the deployment of such systems raises significant ethical and privacy concerns. The 83.2% ROC-AUC, while respectable, means approximately 17% of cases are misclassified—potentially including false positives that could lead to unnecessary interventions or false negatives that miss at-risk individuals. Privacy advocates will scrutinize how surveillance data is collected, stored, and used, particularly given the sensitive nature of mental health assessment.

The framework's interpretability is crucial for building trust and accountability. Unlike black-box systems, stakeholders can understand why specific individuals trigger alerts, reducing algorithmic bias concerns. Future work should focus on validating this approach across diverse populations and geographic contexts, establishing clear protocols for human oversight, and developing robust safeguards against misuse of surveillance capabilities.

• →Interpretable AI framework combines multiple computer vision tasks to assess suicide risk in metro stations with 83.2% ROC-AUC accuracy.
• →System incorporates person tracking, activity recognition, platform segmentation, and trajectory analysis rather than attempting direct intent inference.
• →Technology addresses critical public health need but raises significant privacy and ethical concerns requiring careful governance.
• →Framework's explainability design enables human oversight and reduces algorithmic bias in high-stakes mental health assessments.
• →Deployment success depends on validation across diverse populations, clear intervention protocols, and robust safeguards against surveillance misuse.