Transport hub video analytics uses software to watch the CCTV feeds a station, airport, bus terminal or port already has, and to surface the moments that need a person: crowding building at a stairway, a queue spilling past a security lane, someone entering a track area or staff-only door. Operators still verify and decide. The value is earlier warning and faster review, without replacing cameras or identifying faces.
Key takeaways
- Falls from the platform onto the track account for about 37% of passenger fatality risk on Britain's railway (RSSB, 2024/25), and trespass is the leading cause of rail-related deaths in the US.
- Crowding rarely causes harm at "average" volumes. It causes harm during short peaks that a control room watching dozens of tiles can easily miss.
- Unattended items are a disruption problem as much as a security problem: each one can stop services while staff work through a structured check such as the UK HOT principle.
- Video analytics should help staff apply existing protocols (HOT, the 4 Cs, crowd management plans) faster. It should not replace them.
- The use cases in this guide do not require facial recognition, which matters under the EU AI Act and UK ICO guidance on proportionality.
- Start with a short pilot on a few high-risk cameras, defined owners and a clear review routine before scaling across the estate.
Why do transport hubs need video analytics now?
Passenger volumes are growing. Britain's railway recorded 1.6 billion passenger journeys in 2024/25, up 6% on the year before, according to RSSB's Annual Health and Safety Report 2024/25.
The camera estate is rarely the gap. Attention is: an operator can actively watch a handful of feeds, while incidents develop in seconds at a platform edge. See our piece on reactive vs proactive video monitoring for operations.
Transport hub video analytics is an intelligence layer that runs over existing CCTV and flags configured events, with camera, zone, time and clip attached, so a human can verify and act. It is not a new camera system, and it does not take enforcement decisions on its own.
What are the main safety risks in stations, airports and terminals?
Three risks recur: crowd density, the platform edge and track, and unattended items. Airports and ports add airside perimeter and ground operations.
How dangerous is the platform-train interface?
The platform-train interface (PTI) is the zone where passengers step between platform and train, including the platform edge and the gap. In 2024/25, three of the five passenger fatalities in stations or on trains in Great Britain happened at the PTI, and all three involved falls from the platform onto the track, according to RSSB. RSSB estimates that falls from the platform to the track account for around 37% of all passenger fatality risk.
Slips, trips and falls dominate non-fatal harm. RSSB reports that more than 3,300 station slips, trips and falls were logged into its Safe Insights system in 2024/25.
How big is the track trespass problem?
Trespass is any unauthorized entry onto railway property, including tracks, tunnels and lineside areas. The Office of Rail and Road's Rail Safety statistics for 2024-25, published October 2025, recorded 22 trespass fatalities across British networks, up from 12 the previous year. On the mainline, 17 trespassers died (excluding suicides), including ten struck by trains and four electrocuted.
In the United States, the Federal Railroad Administration states that trespassing is the leading cause of rail-related deaths, with more than 500 trespass fatalities each year (page updated April 2025). Station ends, tunnel portals and depot gates are where hub operators can act.
When does a crowd become unsafe?
Crowd density is the number of people per square meter in a given area. Crowd scientist G. Keith Still notes that UK event guidance typically assesses safety at around 2 people per square meter, that the UK Green Guide (written for sports grounds) uses 4.7 per square meter for standing areas, and that about 5 per square meter is the upper limit for standing spaces.
Transit planners often use Fruin levels of service, which grade pedestrian space from A (free circulation) to F (flow breakdown with many stoppages). In its train station crowding audit, the NSW Audit Office records that Transport for NSW and Sydney Trains use Fruin levels A to F, and that levels E or F are treated as acceptable only for short periods or emergencies.
The harder problem is measurement. The NSW Auditor-General's 2020 report found Sydney Trains rated platform overcrowding a high strategic risk but did not directly measure crowding, relying mainly on train dwell time data. Between January 2016 and October 2019, 31 station incidents were attributed to crowding, 14 of them resulting in injury. Sydney Trains was trialling CCTV and cellular signal detection at Wynyard and Redfern to monitor crowding in real time.
Why are unattended items so disruptive?
An unattended item is a bag, case or package left without its owner nearby. Most are lost property; staff must quickly separate those from the few that are suspicious.
What is the HOT principle?
The HOT principle is the UK's structured check for deciding whether an unattended item is suspicious. ProtectUK, the UK counter-terrorism policing guidance service, sets out three questions:
- Hidden: has the item been deliberately concealed?
- Obviously suspicious: are there visible wires, batteries, tape or liquids, or was it found after suspicious behavior?
- Typical: is it the kind of item you would expect in this location?
If the item is judged suspicious, staff move to the 4 Cs: Confirm, Clear, Communicate and Control. ProtectUK's guidance includes cordon distances of at least 100 meters for a small item such as a rucksack, 200 meters for a large item or small vehicle, and 400 meters for a large vehicle. It also advises against using radios or phones within 15 meters of the item. A 100-meter cordon in a busy concourse closes platforms, gate lines or check-in rows.
What does a single unattended bag cost in disruption?
Paris transport operator RATP reports that around four unattended bags a day were found on its rail networks between January and April 2022, and that each one interrupts traffic for about an hour on average while checks are completed.
Reporting is growing too. The UK government said in September 2025 that reports to British Transport Police's 61016 text service have risen more than eightfold since the "See it. Say it. Sorted" campaign launched in 2016, to 255,088 a year. The same announcement included almost £17 million for Network Rail to improve police access to station CCTV.
In the US, the Department of Homeland Security's "If You See Something, Say Something" campaign began with New York's MTA after 9/11 and went national in 2010, with transit stations among its partners.
More reports mean more items to triage. Video answers the first HOT question fast: when did the item appear, and did its owner walk away or just step to a ticket machine?
Why can't a control room watch every feed?
Large interchanges and airports run hundreds or thousands of cameras, while operators also handle radio calls and help points. A person past the platform edge can be on a tile nobody is watching. With analytics, operators review a short list of flagged events with the clip attached instead of watching everything. See your next security operations hire might be an AI agent, and the deployment details in our practical guide to real-time threat detection.
What can transport hub video analytics detect?
Kote, Kotelab's platform, connects to existing IP cameras and RTSP-capable NVR/DVR streams, and surfaces configured events for human review. For transport, the Kotelab transport page lists:
- Passenger flow: queue lengths at check-in and security, congestion hotspots, terminal flow patterns, boarding gate crowding and peak traffic periods.
- Restricted zones and perimeter: unauthorized access to restricted zones, perimeter breaches, after-hours activity in secure areas and breaches of staff-only areas.
- Ground operations: baggage handling activity, aircraft turnaround zones, loading and unloading, service vehicle movement and dwell, and operational delays or blockages.
Operators can also query footage in plain language. The transport page's example is "Show baggage handling delays near Gate C", which returns a service vehicle blockage and slowed baggage belt activity with timestamps. Our explainer on natural-language video search covers how this works. For an unattended item, a control-room lead could ask something like "Show who was near the bench on Platform 4 between 14:00 and 14:20" to support the HOT check.
Linking staff-door or airside gate events to video helps verify whether a breach is a tailgate; see access control and video correlation explained.
Incident type, camera signal and responder
- Platform crowding at stairs or escalators. What the camera can show: Rising density and congestion hotspot at a defined zone; Typical first responder: Station supervisor, crowd management team; Kotelab fit: Listed: congestion hotspots, crowding.
- Check-in or security queue overflow. What the camera can show: Queue length past a set point, peak period building; Typical first responder: Terminal duty manager, security lane lead; Kotelab fit: Listed: queue lengths at check-in and security.
- Boarding gate crowding. What the camera can show: Gate area density ahead of boarding; Typical first responder: Airline or gate agent, terminal ops; Kotelab fit: Listed: boarding gate crowding.
- Person in track area, tunnel portal or platform end. What the camera can show: Entry into a zone configured as restricted; Typical first responder: Control room, signaller, BTP or transit police; Kotelab fit: Listed: restricted-zone entry.
- Staff-only door or airside breach. What the camera can show: Person entering a staff-only or secure area, after-hours activity; Typical first responder: Security officer, airside operations; Kotelab fit: Listed: staff-only breaches, perimeter breach.
- Perimeter fence breach (airside, depot, port yard). What the camera can show: Movement across a perimeter line; Typical first responder: Security patrol, airport police; Kotelab fit: Listed: perimeter breach, intrusion.
- Unattended item on a concourse or platform. What the camera can show: When it appeared, who left it, where they went; Typical first responder: Staff applying HOT and the 4 Cs; Kotelab fit: Search supports review; ask any vendor how dedicated left-object detection is tested.
- Baggage or ramp delay. What the camera can show: Belt activity slowed, service vehicle dwell near a stand; Typical first responder: Ground handling supervisor; Kotelab fit: Listed: baggage handling, service vehicle dwell.
If you need dedicated left-object alerts, ask any vendor how they behave in busy concourses with seating, trolleys and cleaning carts, and how false alarms are filtered before reaching a radio channel.
How do you stay proportionate on privacy in public spaces?
The EU AI Act, Article 5, lists prohibited AI practices that apply from 2 February 2025. Point (h) bans law enforcement use of real-time remote biometric identification in publicly accessible spaces, except in narrow cases that need prior judicial or independent authorization.
In the UK, the ICO's video surveillance guidance says surveillance must be lawful, necessary and proportionate, and singles out facial recognition in public spaces as particularly intrusive. The ICO notes the guidance is under review following the Data (Use and Access) Act, so check for updates.
The practical point: crowding, queue length, restricted-zone entry, perimeter breach and baggage delays are about where people and objects are, not who they are. None of the use cases in this guide require facial recognition. Kotelab customers configure retention, video is encrypted with AES-256, and Kotelab does not sell or use customer video for advertising. Details sit in the Kotelab Legal & Trust Center.
Good practice: update your data protection impact assessment before adding analytics to public-area cameras, switch on only detections someone owns, keep flagged clips no longer than your evidence process needs, and keep a human in the loop for every action affecting a passenger.
How do you run a transport hub video analytics pilot?
Kotelab's onboarding follows three stages: define the operational need, review the camera estate, then configure and evaluate with operators. In a hub:
- Pick three to five problem zones from your incident log, such as a busy platform stairway, a security queue, a staff-only door and a platform end.
- Name the owner for each event. Crowding goes to the station supervisor, track-area entry to the control room, staff-door breaches to security. No owner, no alert.
- Review the cameras for angle, lighting, platform-edge coverage and RTSP access on your NVR or VMS.
- Configure zones and thresholds with operators, agreeing what "too crowded" means for each zone using your existing crowd guidance.
- Run in review mode for two to four weeks. Operators mark each flagged event useful, noise or missed, checked against radio logs.
- Wire alerts into existing procedures such as HOT and the 4 Cs, crowd management plans and dispatch routines.
- Scale only where the pilot showed faster verification or fewer missed events.
Comparing vendors? Our guide to the best AI video analytics software for existing cameras lists the questions to ask. If you also run park-and-ride or depot chargers, see EV charging station cable theft and vandalism.
Frequently asked questions
What is transport hub video analytics?
Transport hub video analytics is software that analyzes live CCTV feeds in stations, airports, bus terminals and ports to flag events such as crowding, queue overflow, restricted-zone entry and perimeter breaches. It runs over existing cameras and sends flagged events with a clip to staff, who verify and decide what to do.
Can video analytics detect crowd density on a train platform?
Yes, analytics can flag congestion hotspots and rising crowding in defined zones such as stair feet, escalator landings and platform sections. The thresholds should come from your own crowd management plans and guidance such as Fruin levels of service, set with the station team. Staff then decide whether to hold gate lines, make announcements or restrict access.
How does CCTV help with unattended items?
CCTV lets staff answer the HOT questions faster by checking when an item appeared, who left it and whether they walked away. Natural-language search can pull up the relevant minutes of footage without scrubbing through recordings. The decision to clear an area, and how far, stays with trained staff and police under protocols such as ProtectUK's 4 Cs.
Does transport hub video analytics require facial recognition?
No. Crowding, queue length, restricted-zone entry, perimeter breaches and ground operations delays are about positions and movement, not identity. Avoiding identification keeps these use cases away from the stricter rules on remote biometric identification under the EU AI Act and the ICO's concerns about facial recognition in public spaces.
Do we need to replace our cameras?
Usually not. Kote connects to existing IP cameras and RTSP-capable NVR/DVR streams from brands including Hikvision, Dahua, Axis, Bosch, Hanwha Vision and Avigilon. A camera review during the pilot shows which views are usable.
Can it monitor airside perimeters and ground operations?
Yes. Kotelab's transport use cases include perimeter breaches, staff-only area breaches, after-hours activity, baggage handling activity and service vehicle dwell. Operators can ask "Show baggage handling delays near Gate C" and get time-stamped events back.
See what your existing cameras can already tell you
Kote, Kotelab's agentic physical security platform, turns those feeds into flagged events for crowding, restricted zones, perimeter breaches and ground operations, so your team spends its attention where it counts. Learn more about Kotelab for transport hubs, or talk to our team about a pilot on your highest-risk platforms, queues and gates.