Staffing by experience
Without continuous data on peaks, passages and queues, staff allocate resources by experience and rely on retrospective video review.
Intelligent Station Incident Detection Solution
Turn station CCTV into actionable information
SIDS reads existing CCTV feeds in parallel and turns them into passenger counts, equipment status and incident records for control-room review and operational analysis.
What continuous video alone cannot tell operators
Incidents found by watching screens
End-door openings, falls and entrance congestion rely on visual detection; fatigue can delay intervention.
Faults found through reports
Passengers or patrols often first report faulty feeds, ticket machines or gates. Late detection affects passenger service.
Video without operational records
Nearly 400 feeds run continuously, mainly for retrospective evidence, without daily, weekly or monthly flow, alarm and equipment records.
Parallel integration and deployment
Integration path
| Stage | Description | Status |
|---|---|---|
| Video sources | Nearly 400 existing camera feeds at entrances, gates, escalators and end doors | Existing, unchanged |
| Core switch | Tap a network connection from CCTV to obtain feeds in parallel | Existing, unchanged |
| Firewall | Isolate detection traffic from the existing network | New |
| Detection server | Run detection services and output flows, status and incidents | New |
| Control-room frontend | Large-screen display, audiovisual alerts and video review | New |
Two-host deployment
| Location | Workload |
|---|---|
| Equipment-room server | Video ingestion and all detection processes |
| Control-room host | Information aggregation, streaming, web frontend and speakers |
Alerts sound in the control room where duty staff can hear them. The hosts use configurable communication addresses without algorithm changes.
Software architecture
Each detection function runs in a separate multithreaded process to isolate faults. Containerised streaming starts with the system.
Only the firewall, detection server and control-room frontend are added. Existing network topology and cameras remain unchanged.
Six detection categories
Rain and water ingress
Identify rainfall, pooled water and indoor leaks to support flood response and passenger guidance.
Platform end doors
Flag unexpected openings, count night-work entries and exits, and prompt a clearance check at the agreed time.
Escalator boarding and exit safety
Detect falls, oversized luggage and congestion at landings so staff can locate the affected escalator.
Automated video rounds
Cycle through camera routes and record blank or distorted feeds instead of checking every screen manually.
Passenger assistance and ticketing incidents
Flag prolonged lingering near gates, unavailable ticket machines and situations that may require assistance for staff to verify.
Crowd monitoring
Measure regional flows, flag surges and grade platform queues to inform staff deployment.
The six categories and 14 subcategories are enabled as needed. Set areas, triggers and priorities separately for each task.
Measuring passenger flow
Coverage areas
Cover entrances, gates, escalators, transfer passages, restroom access and platforms. All cameras in the platform queue area contribute to grading against three reference lines.
Counting intervals
Count every 10 minutes from 07:00–09:00 and 17:00–19:00, and every 30 minutes off-peak.
Surge detection
Calculate separate weekday and weekend means over the past month. Flag a surge when flow exceeds the applicable mean by 20%.
Report output
Export daily, weekly and monthly Excel reports, recording incidents and lingering separately for operational review.
Alert-to-video review
Open the live feed directly from the timestamp, location, category and still image.
Alarm handling in the control room
- 1Detect: create an incident record.
- 2Prioritise: sort by two severity levels, then by time.
- 3Display: keep incidents visible without interrupting work with modal windows.
- 4Notify: combine category-specific voice messages with a flashing indicator.
- 5Retain evidence: keep detection video, the original feed and a 2,000-frame clip containing the incident.
- 6Confirm: compare both views and review suggested actions. Level-one alerts persist until acknowledged.
- 7Report: weekly and monthly Excel logs retain alerts and suggested actions for handover and review.
Control-room views
Authorised station browsers access four tabs in a 1920×1080 web interface. The demonstration views do not show identifiable passenger footage.
Acceptance criteria by function
- Camera communications and picture faults
- >99%
- Passenger flow, end doors and station clearance
- >95%
- Equipment status, flooding and escalator events
- >90%
- Time from detection to display
- <5 s
These figures are design and acceptance requirements, not measured site results. Verify enabled functions separately under project acceptance conditions.
From passenger data to operating decisions
A transfer station conducted an initial analysis of nearly 400 feeds. These observations are specific to that station and period.
| Comparison | Finding | Operational implication |
|---|---|---|
| Entrances G and H | G handled about 1.6 times H’s entry and exit traffic; exits predominated in the morning, entries in the afternoon | Prioritise advance deployment of guidance staff at G |
| South and north hall escalators | Traffic peaked at 08:00–10:00; south hall flow was about 6.5 times north hall flow | Station assistance staff in the south hall and shorten maintenance intervals |
| North and south transfer passages | The north passage carried about twice as many people as the south | Advertising space in the north passage has higher commercial value |
Extend station by station: connect feeds, configure scenarios, accumulate records and refine operations.
Last updated