Solutions / Catalogue / Edge Video Analytics

Edge Video AnalyticsReference design

Fall Detection

An on-device fall detector for care homes, rehab wards and people living alone. A reCamera or an existing IP camera feeds a pose model that tracks each person and raises one MQTT alert per fall. Video stays local, and an optional alarm panel lets staff acknowledge and escalate alerts.

Healthcare

Talk to an engineer See the hardware

Empty, clean hospital room with an adjustable bed, chair and bedside table — Photo: Wnkwdy, CC BY-SA 4.0, Wikimedia Commons. Shows the setting, not this installation.

95.8%: Fall recall on a 27-clip held-out public test set (GMDCSA-24)
77.8%: Everyday activity clips that did not trigger an alert, same test set
85.8%: Average accuracy across six configurations, same test set
1.4 s: Average time from fall to alert, same on all platforms

Results reported by the design's authors on their own test setups. Yours depend on the site, so we confirm them in a pilot.

The problem

Why it matters.

Falls in care homes or among people living alone go unnoticed until the next check.

The flow

From signal to action.

01SenseCameraA reCamera or an existing RTSP camera views the room from the side or a corner.
02DecidePose trackingThe device estimates 17 body points for each person and follows them across frames.
03DecideFall logicA state machine collects about 5 seconds of evidence before declaring a fall.
04ConnectMQTT eventOne event per fall is published, and Home Assistant picks up the sensors automatically.
05PeopleAlarm panelStaff acknowledge on the panel within 60 seconds, or the alert escalates.

Deployment architecture

How it is wired.

The devices on site, what carries their data, where the decision is made and where the result goes.

01SenseOn site

reCamera 20021 per view
reCamera Pro 2GB1 per view
IP camera (RTSP)1 per view

02ConnectCarries the data

RTSP

03DecideRuns the model

reComputer R1000 Series (4 GB RAM, 32 GB eMMC)1, optional
reComputer Super J30111 per site
reComputer Super J40121 per site
reComputer RK3576-301 per site
reComputer Industrial R2035-121 per site
TensorRT

04Your systemsWhere results land

MQTT
Mosquitto
Home Assistant
Webhook

How it works

Step by step.

01Mount the camera at the side or in a corner, 2 to 3 m from walking space, with shoulders and hips visible.
02Choose a reCamera on its own or an IP camera with a reComputer host.
03Set no-person and no-movement timeouts for each zone.
04A confirmed fall sends one MQTT event, which can drive Home Assistant or the alarm panel.
05Staff respond, and unanswered alerts escalate.

The hardware

What goes on site.

Items we stock can go straight in the bag. Everything else we source and quote with the installation.

reCamera 2002All-in-one AI camera1 per viewSourced on quote
reCamera Pro 2GBAll-in-one AI camera1 per viewSourced on quote
reComputer R1000 Series (4 GB RAM, 32 GB eMMC)Optional alarm panel host for reCamera setups1, optionalSourced on quote
reComputer Super J3011Detection host for IP cameras (Jetson Orin Nano)1 per site₹84,589
reComputer Super J4012Detection host for IP cameras (Jetson Orin NX)1 per site₹1,59,389
reComputer RK3576-30Lower-cost detection host1 per siteSourced on quote
reComputer Industrial R2035-12Hailo-8 detection host for many sub-streams1 per siteSourced on quote
IP camera (RTSP)Video source for the reComputer presets1 per viewSourced on quote

Before you install

A side or corner view 2 to 3 m from the area, aimed at walking space rather than beds or exercise areas.
No top-down mounts, long corridor shots or heavy furniture blocking the view.
A local network. Internet is only needed to forward alerts outside.
At least 10 GB free disk on Jetson hosts for the first engine build.
640x480 or 640x360 sub-streams from cameras when using the R2035-12.

Good to know

It is not a medical or life-safety device and makes no promise about missed falls.
In testing, 77.8% of everyday activity passed without an alert, so false alarms will happen and need staff review.
Recall drops to 52.9% to 58.8% on wide shots and hidden views.
The reference pose weights are AGPL-3.0, so confirm a licence or swap the model before selling it commercially.

How we deliver it

Paper to site, one team.

01Site surveyWe walk the site, or review photos and drawings, and agree what to measure and where each device goes.
02PilotOne lane, one line or one room first. We tune it on your conditions and agree what good looks like.
03Roll outWe supply the hardware in India with GST invoices, install it, and connect it to the systems you already run.
04SupportSpares, remote checks and model updates, with one team to call when something changes.

Questions

Asked before buying.

Does it work in a dark room?

The page gives no low-light figures. Test in the real room at night, or consider radar for dark spaces.

Can it link to Home Assistant?

Yes. It announces a fall sensor, state and presence sensor to Home Assistant automatically over MQTT.

Can we sell it as part of our product?

The runtime is Apache-2.0, but the reference pose model is AGPL-3.0, so you need a commercial licence or a different model.

Want this on your site?

Tell us the site and the problem. We come back with a pilot plan and a priced hardware list.

Book a call with an engineer Write to us

Want this on your site?

Industrial Visual Inspection

Parking Lot Management

Elderly Care Monitoring

Want this on your site?

Industrial Visual Inspection

Parking Lot Management

Elderly Care Monitoring