VLM output
A handful of masked passengers stand in an orderly line along the platform screen doors of an underground metro platform, most looking at phones while waiting for a train. Over the four frames the group shifts slightly and one more person joins the queue; a train is visible stationary behind the closed platform doors and no incident occurs.
- Platform Edge Line CrossedNo
- Passenger Fall Or IntrusionNo
- Crowd DensityLow
- Waiting Passengers Count8
- Queue OrderlyYes
- Screen Doors ClosedYes
- Train At PlatformYes
- Face Mask UsageMost
- Staff PresentNo
Customer challenges
Thousands of CCTV streams make continuous manual monitoring impractical
Safety and operational incidents often depend on manual observation, patrols, or passenger reports
Track intrusion, unattended objects, and unsafe behavior require immediate attention
Crowd congestion and queue conditions are difficult to monitor consistently across stations
Flooding and environmental conditions can disrupt station and network operations, as well as passenger safety.
CCTV, VMS, alarms, and operational information are often managed across separate systems
Traditional video analytics are typically built for individual use cases, making new scenarios costly and time-consuming to deploy
Transit agencies need to modernize operations while maximizing existing CCTV investments
Customer benefits
Establish 24/7 proactive AI monitoring across metro environments
Shift from manually watching CCTV to AI-assisted monitoring
Detect potential safety and operational issues earlier
Shorten Detection → Verification → Response time
Reduce control-center workload associated with continuous CCTV monitoring
Transform existing cameras into intelligent operational sensors
Rapidly introduce new use cases through VLM-based analytics
Scale from a single station to multiple lines and network-wide deployments
Key features
AI Passenger & Crowd Intelligence
Monitors crowd density, platform and concourse congestion, queues and abnormal crowd formation, with real-time peak-hour alerts.
AI Safety & Security Detection
Detects track and restricted-area intrusion, unattended objects, unsafe or abnormal behavior, falls and person-down events, and monitors construction and restricted zones.
AI Infrastructure & Environment Monitoring
Detects flooding, fire and smoke, camera obstruction, blur and health issues, and other infrastructure and environmental anomalies through 24/7 AI inspection of critical areas.
VLM-Powered Event Detection
Defines detection scenarios in natural language for complex event and scene understanding, enabling rapid deployment of new AI use cases with flexible rules per station and environment.
Smart Incident Management
Delivers real-time alerts with automatic event video capture and evidence retention, AI event classification, centralized camera, station and event management, an AI-assisted alert plus human verification workflow, and integration with existing VMS and control-center systems.
Before & after
| Before | After — with Linker AI Nexa |
|---|---|
| Before Operators manually monitor CCTV | After — with Linker AI Nexa AI continuously monitors and identifies relevant events |
| Before Incidents depend on manual or passenger reporting | After — with Linker AI Nexa AI proactively detects and alerts operators |
| Before Video is primarily reviewed after an incident | After — with Linker AI Nexa Relevant video evidence is surfaced in real time |
| Before Crowd conditions rely on visual observation | After — with Linker AI Nexa AI continuously monitors crowd and queue conditions |
| Before Safety risks may be discovered after escalation | After — with Linker AI Nexa AI identifies potential risks earlier |
| Before CCTV primarily serves as a surveillance system | After — with Linker AI Nexa CCTV becomes a real-time operational sensing network |
| Before Individual AI models for individual scenarios | After — with Linker AI Nexa VLM enables faster expansion of new use cases |
| Before Information is fragmented across systems | After — with Linker AI Nexa AI events are centrally managed |
| Before Reactive operations | After — with Linker AI Nexa Proactive, AI-assisted operations |