MetroGuard

Traditional CCTV systems across rapid transit platforms are entirely forensic—passively recording slips, medical collapses, and platform edge intrusions for post-incident review after a tragedy has already occurred. In bustling transit networks like Pune Metro Line 1 & Line 2, platform crowding and high passenger turnover demand an autonomous system that intervenes before a train arrives.

Our Solution: MetroGuard AI is a local-first, operational safety copilot that transforms standard CCTV feeds into an active intervention system. Built on a modular "Eyes & Brain" architecture:

The Eyes (Local Edge Vision): Powered by YOLOv8-Pose and ByteTrack, the system executes real-time (25+ FPS) skeletal landmark tracking directly on local hardware. By computing torso tilt kinematics ($\ge 60^\circ$ relative to the vertical axis) and ground-foot contact points, it reliably distinguishes routine yellow-line waiting from critical platform falls and track bed collapses. The Spatial Matrix: Dynamically correlates boundary dwell times with simulated CBTC (Communications-Based Train Control) train arrival telemetry to scale threat tiers from Caution to Critical Emergency. The Brain (Cognitive Copilot): Sanitized numerical telemetry is processed by OpenAI GPT-4o-mini to generate instant 3-bullet Standard Operating Procedure (SOP) dispatch briefings for station controllers and platform marshals, backed by a resilient offline rule-based fallback. Autonomous Action: The system triggers real-time localized bilingual Public Address (PA) speech warnings (English & Marathi/Hindi) and asserts simulated Emergency Train Hold (ETH) signal interlocks when seconds count. With MetroGuard AI, transit operators gain an intelligent, privacy-preserving copilot designed for zero platform fatalities.