Computer vision · Events and alerts
The alert reachesthe right person.
AI video event detection means visual rules applied to camera images: who goes where they should not, whether helmets and vests are worn in the work area, whether an object has fallen onto an access route. The system sends the alert with the related image, and a person checks it and decides. It is extra support for the team, alongside certified security and fire detection systems, which it does not replace.
Dozens of cameras, a single operator
A screen split into many images cannot be watched with constant attention. Important events are rare and brief, so they are easy to miss in real time and hard to find later in hours of footage.
- Entries into hazardous or restricted zones are noticed only if someone is looking at exactly that moment
- Wearing protective equipment is checked only through occasional inspections
- Objects that fall or are left on access routes stay there until someone walks past
- Finding an event in recordings means hours of manual scrolling
Who it is for
- Health and safety officers in factories, warehouses or construction sites
- Security teams and control rooms that monitor many cameras at once
- Operators of industrial sites with restricted or hazardous zones
- Building managers who want to find out quickly about blocked access routes
What video events it can flag
- 01
Entry into a defined zone
Flags the presence of a person or vehicle in a zone drawn on the image, optionally only during certain hours.
- 02
Protective equipment
Estimates whether people in the work area are wearing a helmet or a high-visibility vest and flags likely non-compliant cases.
- 03
Fallen or abandoned objects
Notices when an object appears and stays on an access route or in an area that must be kept clear.
- 04
Smoke-like patterns
Can flag visual patterns that resemble smoke, as an additional warning to be checked, not as a certified detector.
- 05
Alert with visual evidence
Each alert includes a frame or a short clip, so the person receiving it can judge the situation quickly.
- 06
Event log
Events are indexed by type, camera and time, so they can be found without scrolling through recordings.
How it works
From frame to operator decision
The steps marked in yellow are approved by a person.
- Input
- AI
- Action
- Human in the loop
- Result
- 01
Input
Live video stream
Compatible cameras send the image to the analysis system.
- 02
AI
Detection and rules
The model recognises people, objects or patterns and checks the zone rules.
- 03
Action
Filtering and grouping
Repeated events are grouped, and those below the threshold are marked or ignored.
- 04
Action
Alert sent
The alert, together with the image, reaches the agreed channel.
- 05
Human in the loop
The operator checks
A person confirms the event, rejects it or starts the internal procedure.
- 06
Result
Documented event
The decision and the visual evidence stay in the log for review.
What data and cameras alerts require
- A list of events of interest, with clear definitions: what counts as an entry, what counts as loitering, during which hours
- Real video footage of those events and of normal situations that look similar
- A zone map for each camera and the procedures that follow an alert
- Cameras with the right angle and resolution for what needs to be seen: a helmet far away takes up only a few pixels
- An agreed alert channel and a person responsible for each type of alert
Possible integrations
- IP cameras with an RTSP stream
- ONVIF-compatible cameras
- NVRs and VMS systems, for example Milestone XProtect
- Email and SMS notifications
- Telegram, Slack or Microsoft Teams
- Webhooks to the control-room application
- MQTT
- NVIDIA Jetson edge devices
- Grafana dashboards
- IP cameras with an RTSP stream
- ONVIF-compatible cameras
- NVRs and VMS systems, for example Milestone XProtect
- Email and SMS notifications
- Telegram, Slack or Microsoft Teams
- Webhooks to the control-room application
- MQTT
- NVIDIA Jetson edge devices
- Grafana dashboards
Each integration's compatibility is checked during the analysis phase, for the systems and versions you use.
Control and transparency
What stays with people
- Confirming each alert before any intervention or conversation with the people involved
- Triggering emergency procedures, which follow internal rules, not the automatic alert
- Setting the zones, active hours and sensitivity, with a review after the pilot period
- Legal review and informing employees before any rule that concerns people's behaviour
Limits and operating conditions
- It is not a certified fire detection, alarm or machine safety system and does not replace them
- False alarms (reflections, shadows, fog, vegetation) and missed events can occur; thresholds are tuned during the pilot to balance the two
- Checking protective equipment depends heavily on distance, angle and obstructions and becomes unreliable at long distances
- At night, in rain or against the light, detection quality can drop noticeably without additional lighting
How we evaluate alerts: confirmed, missed, response time
We agree the indicators together before launch and track them afterwards. The values depend on the process, the data and the volume.
- Confirmed alerts
- The share of alerts the operator confirmed as real, recorded with each decision
- Missed events
- Known events, from planned tests or internal reports, that the system did not flag
- Time to review
- The interval between the alert being issued and the human decision, calculated from the event log
- Performance by condition
- Results analysed separately for day and night, good and bad weather, for each camera
How we implement video alerts, step by step
- 01
We define the events and the procedure
We agree what needs to be flagged, who receives the alert and what they do once they receive it.
- 02
We assess the cameras and the legal framework
We check real images and stream compatibility, how filmed people are informed and, where relevant, the impact assessment (DPIA).
- 03
We test on recordings
We run the rules on recordings and simulated scenarios, then tune the zones and thresholds.
- 04
Pilot with the operator in the loop
We activate alerts on a few cameras and adjust based on confirmations and rejections.
Use cases
The scenarios below are conceptual examples, designed to illustrate realistic applications. They are not implementations delivered by WEB-AZ.
- Warehouses
People in the forklift aisle
When someone enters the aisle marked for machinery during working hours, the shift supervisor receives an alert with the image and decides whether to step in.
Conceptual example - Construction
Helmet and vest in the assembly area
At the entrance to the work area, the system flags likely cases of a missing helmet, and the health and safety officer checks them and talks to the team, with no automatic penalties.
Conceptual example - Office buildings
Blocked evacuation routes
The corridor camera flags an object left in front of the emergency exit, and the building manager receives the notification.
Conceptual example
Useful guides
Before you decide, read
Frequently asked questions
What is AI video event detection?
It is a computer vision application that watches camera images according to agreed rules, for example entry into a restricted zone or an object left on an evacuation route, and sends an alert with the related image. A person checks each alert and decides what happens next, following internal procedures.
Can it replace the fire detection system?
No. Flagging smoke-like patterns is only an additional warning, for human review. Certified detectors and systems remain necessary under applicable regulations.
What happens with false alerts?
They are part of how any such system works, which is why we address them in the design. Each alert has visual evidence attached, so the operator can reject it quickly. Rejections are recorded and we use them to adjust the zones, thresholds or model.
Is it legal to analyse images of employees automatically?
It depends on purpose, proportionality and information, and employee monitoring is strictly regulated. We recommend rules that concern zones and safety, not people's performance, clear information for those filmed, limited retention and, where necessary, a data protection impact assessment, with legal support.
Does the system identify the person who entered the zone?
No. Detection establishes that a person entered the zone, not who they are. Facial recognition is not included and would require a separate assessment of legality, necessity and risks.
Does it work at night?
It depends on the cameras and the lighting. Infrared cameras can help, but quality is usually lower than during the day. We test on night-time recordings before activating rules for those hours.
Can’t find the answer? Write to us and we’ll reply directly.
Which events should reach you straight away?
Describe the zones and situations that concern you. We check whether they can be flagged visually and how alerts fit into your procedures.
Related solutions
- Object detectionModels that recognise the objects that matter to your process in images and pass them on as structured data.
- Operational monitoringVisual tracking of process stages and of vehicle and pallet movement in yards and warehouses, so you can see where the flow gets stuck.
- AI agent for operationsTracks internal requests, tickets and stock thresholds, prepares tasks for teams and coordinates handovers between departments, with a person responsible for every decision.