Computer vision · Fall detection
Someone falls.Someone knows.
Fall detection is the computer vision technique that recognises, from body posture and movement, the moment a person falls and stays on the ground. We build systems that watch the agreed areas and send an alert to the staff on duty, who check and step in. The system observes an event, not an identity: it does not recognise faces and does not work out who the person is.
Events
- 06:17:38Person in view · not identified
- 06:17:39Sudden change of posture
- 06:17:40Person on the floor · alert
- 06:17:58Shift carer confirms
An unseen fall means lost time
When a person falls in an area with nobody around, help can be delayed for a long time, even if the area is filmed. Cameras record, but nobody watches every image, and the person who has fallen cannot always call for help.
- At night or on reduced shifts, staff cannot be present in every area
- Emergency buttons only help if the person is wearing one and is able to press it
- Camera footage is usually looked at after the event, not while it is happening
- Lone workers in isolated areas can go unnoticed for a long time after an accident
Who it is for
- Care homes and care centres, for common areas, corridors or rooms, with the agreed consent
- Warehouses and halls where people work on reduced shifts or in isolated areas
- Companies with lone workers on night shifts, in plant rooms or in security roles
- Building managers with little-used areas, such as stairwells, underground car parks or service corridors
Step by step
How the system recognises a fall, step by step
The animation follows a corridor where a person falls by the handrail, from the camera image to the staff member's confirmation. The five steps below explain each stage.
Events
Events
- 06:17:38Person in view · not identified
Events
- 06:17:38Person in view · not identified
- 06:17:39Sudden change of posture
Events
- 06:17:38Person in view · not identified
- 06:17:39Sudden change of posture
- 06:17:40Person on the floor · alert
- 06:17:58Shift carer confirms
Events
- 06:17:38Person in view · not identified
- 06:17:39Sudden change of posture
- 06:17:40Person on the floor · alert
- 06:17:58Shift carer confirms
Step 1 of 5
The image of the corridor
The camera is mounted high with a wide angle, so that the body is seen as fully as possible, including the floor around it. A usable frame has even lighting and few obstacles between the camera and the monitored area. The images are analysed locally, on a device inside the building.
Two ways of recognising a fall
From pose
The model estimates the body skeleton in every frame and follows the tilt of the torso and its position relative to the floor. It works well when the whole body is visible, for example in wide corridors or open communal areas.
From trajectory
The system follows only the box around the person and learns how its shape and position change over a short sequence of frames. It copes better when the body is partly hidden or the camera is far away, but it needs examples from your own space.
What you receive at hand-over
- The person detection and fall recognition models, based on pose or trajectory, exported for the chosen local device
- The monitored zones drawn on each camera, with the time-on-floor interval and thresholds agreed with your team
- Alerts linked to staff phones or, typically, to the existing call system, where it allows integration
- The interface where alerts are confirmed or marked as false, together with the event log
- A measurement report on falls simulated safely in your spaces and on everyday movements that can resemble a fall
- Documentation, staff training and the image blurring and retention rules, configured
No strings attached: tell us what the cameras should see and we'll tell you honestly whether it can be done.
Let's talkWhat the fall detection system can do
- 01
Estimates body posture
The model follows the general position of the body, for example standing, sitting or lying down, without recognising the person.
- 02
Recognises a falling movement
It compares consecutive frames to tell a sudden fall from ordinary sitting down or bending over.
- 03
Checks whether the person gets up
The alert can be sent only if the person stays on the ground beyond a set interval, to reduce unnecessary alarms.
- 04
Watches only the agreed areas
Analysis runs on the zones drawn on the image, such as aisles or corridors, and the rest of the frame can be ignored.
- 05
Sends an alert to staff
The alert reaches the person on duty with the location and time, and the image can be blurred or limited to what is needed to check.
- 06
Keeps an event log
Each alert and the staff response are recorded, for review and for tuning the system.
How it works
From a fall to staff stepping in
The steps marked in yellow are approved by a person.
- Input
- AI
- Action
- Result
- Human in the loop
- 01
Input
Image from the area
A camera facing the agreed area sends images for on-site analysis.
- 02
AI
Posture and movement
The model estimates posture and recognises the pattern of a fall.
- 03
Action
Confirmation over time
It checks whether the person stays on the ground beyond the set interval.
- 04
Result
Alert to staff
The person on duty receives the location, time and minimal visual evidence.
- 05
Human in the loop
Check and help
Staff check the situation and give the help needed.
What fall detection needs
- Cameras with a clear view of the areas of interest, mounted so that the whole body is visible as often as possible
- Real images from the space, including ordinary movements that can look like a fall: bending, sitting on the floor, picking things up
- A list of the watched areas, the excluded ones and the people who receive alerts on each shift
- Written decisions on informing people, consent where applicable, image retention and access
- On-site processing close to the cameras, so images do not leave the building and alerts do not depend on the internet
Possible integrations
- IP cameras with an RTSP stream
- ONVIF-compatible cameras
- NVIDIA Jetson edge devices
- Local servers with GPUs
- Nurse call and staff call systems, where they offer an integration interface
- Notifications via mobile apps, SMS or messaging apps
- REST API and webhooks
- MQTT
- IP cameras with an RTSP stream
- ONVIF-compatible cameras
- NVIDIA Jetson edge devices
- Local servers with GPUs
- Nurse call and staff call systems, where they offer an integration interface
- Notifications via mobile apps, SMS or messaging apps
- REST API and webhooks
- MQTT
Each integration's compatibility is checked during the analysis phase, for the systems and versions you use.
Control and transparency
What stays with people
- Checking every alert and deciding what help to give
- Setting the watched areas, especially in private spaces such as bedrooms or bathrooms, where cameras may be unsuitable
- Informing people, obtaining consent where required and following data protection rules
- Reviewing false alerts and missed falls, so the system can be tuned based on them
Limits and operating conditions
- The system only sees what the camera sees: a fall behind an object, outside the frame or in darkness may be missed
- Ordinary movements, such as bending, sitting on the floor or physical exercise, can sometimes be mistaken for a fall
- Several people close together or overlapping in the frame make posture estimation harder
- It is not a medical device and does not replace staff, supervision or other existing alarm systems
How we evaluate fall detection
We agree the indicators together before launch and track them afterwards. The values depend on the process, the data and the volume.
- Missed falls
- Tested with falls simulated safely, using mats and volunteers, in the real areas, checking whether an alert was raised
- False alerts
- Alerts that do not match a real fall, counted over a period of operation and grouped by type of movement
- Time to alert
- The interval between the fall and staff receiving the alert, measured on the simulated falls
- Stability across conditions
- Results compared separately for day and night, busy and empty areas, and different camera angles
How we put fall detection into service, in 4 steps
- 01
We agree the areas and the rules
Together we choose the watched areas, who receives alerts and the rules for informing people and keeping data.
- 02
We check the cameras
We review the angle, distance and lighting, so people are as fully visible as possible in the chosen areas.
- 03
We test with simulated falls
We run safe tests in the real spaces, measure missed and false alerts and tune the thresholds.
- 04
We integrate and monitor
We connect alerts to staff channels and regularly review events together with your team.
Use cases
The scenarios below are conceptual examples, designed to illustrate realistic applications. They are not implementations delivered by WEB-AZ.
- Elderly care
Care home corridors at night
Corridor cameras flag a person who has fallen and does not get up, and the nurse on shift receives the alert on their phone with the exact location and checks straight away.
Conceptual example - Logistics
Warehouse aisles on the night shift
In a warehouse with few staff at night, a fall in an isolated aisle reaches the shift supervisor as an alert, and they go over or call their colleague.
Conceptual example - Facility management
Plant rooms with lone workers
In a building's plant rooms, a technician working alone is covered by an alert to the control room if they stay on the ground after a fall.
Conceptual example
Useful guides
Before you decide, read
Frequently asked questions
What is AI fall detection?
It is a computer vision technique in which a model analyses camera images, estimates body posture and movement and recognises the moment a person falls and stays on the ground. It then sends an alert to staff, who check and step in.
Does the system recognise the person who fell?
No. Fall detection works with body posture and movement, not with faces or other identifying features. We do not use facial recognition or biometric identification, and the images in the alert can be blurred or limited to what is needed to check.
Can it be used in residents' rooms or bathrooms?
These are very private spaces, and video cameras may be unsuitable there. The decision is made together with management, the people concerned or their representatives and a data protection specialist, and in many cases a different solution, such as an emergency button, remains the right one.
What if someone just bends over or sits on the floor?
The system usually checks whether the person stays on the ground beyond a set interval, so it does not send an alert every time someone bends down. We test on ordinary movements in your space and tune the thresholds, and any remaining false alerts are reviewed with your team.
Does it replace staff or emergency buttons?
No. It is additional support that helps staff learn about a fall sooner, especially in areas where they cannot be present all the time. It is not a medical device, and supervision and other alarm systems remain necessary.
Do the images leave the building?
We recommend on-site processing, on a device inside the building, so that images stay there and only the alert is sent out. What is kept, for how long and who has access is agreed before going live.
Can’t find the answer? Write to us and we’ll reply directly.
Do you have areas where a fall could go unseen?
Tell us which spaces you want to cover, who should receive the alerts and which cameras are already in place. We will tell you whether fall detection makes sense there and which rules need to be set.
Related solutions
- Video event detectionVisual rules that flag entries into restricted zones, missing protective equipment or fallen objects, for human review.
- Edge AI for computer visionVideo analysis on devices installed on site: fast response, less network load and images that never leave the building.
- 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.