KamiVision's three-layer system, an edge chip, cloud AI filter and human review of a blurred clip, reports 99.5 percent alert accuracy by the time alerts reach users.
KamiVision says its camera-based KamiCare system needs no action from a fallen person, unlike wearable alarms that fail when users forget to press them or the battery is flat.
Providers are paid to fill rooms rather than for better care, so KamiVision pitches its system as an insurance policy that protects building revenue from injurious falls.
AI fall detection is accurate, but elder care economics resist it
KamiVision is repurposing camera AI from its home-security business to detect falls in elder care. Combining on-device detection, cloud filtering and human review makes the alerts reliable enough to act on. The harder problem is economic, because care providers are paid to fill rooms and a prevented fall leaves no visible return. Detection only improves care when a human workflow turns each alert into help.
Takeaways
Layered review pushes alert accuracy to 99.5 percent. An edge chip flags a possible fall and a cloud AI model strips out false positives. A human then checks a blurred, encrypted clip. False alerts were common until 2023 or 2024, when tens of thousands of recorded falls trained the model.
Cameras work where wearable alarms fail. Wearables depend on users remembering to press a button and keeping the battery charged. Link says his own grandmother would not remember in an emergency. A camera system needs no action from the person who falls.
Return on investment is hard to prove. Providers are paid to fill rooms rather than for better care, and prevented falls leave nothing visible. Link frames KamiCare as insurance for the building, since US families pay $4,000 to $10,000 a month and may leave after injurious falls.
Reviewing fall clips makes facilities proactive. Facilities first treat falls as unavoidable. Link says 96 percent of falls cause no hip fracture or injury, and most are people lowering themselves to the floor. One memory care resident sat down to put on shoes, so staff arranged dressing help.
Fast alerts help staff prioritise long rounds. KamiCare notifies staff within 60 to 90 seconds when someone is on the ground. Rounds in a 50-room building can take two and a half to three hours, so the alert acts as a buffer rather than a guaranteed response time.
Private homes are much harder than facilities. Facilities have uniform studio or one-bedroom rooms and consistent lighting, almost like a test lab. Homes bring varied lighting, pets and blind spots and may need five or six devices. KamiVision has explored radar, robot vacuums, mini drones and a heart-rate ring.
Freedom from venture capital enables lower prices. Millions of dollars from the home-security business funded KamiCare, so KamiVision answers to no VC investors and can undercut competitors. KamiHome has over a million paid users and three or four million free subscribers, giving experience with large data volumes.
What KamiVision is working on next
Showing ROI by cutting the time staff spend on fall reporting.
Using medical data to flag residents becoming fall risks before an emergency visit. This is a hypothesis and is not yet deployed.
Tap a name to go there · tap a line to see why they are linked
On this episode 2
Show 1
Considering a SIM card for a future consumer version, since rural homes and some facilities have unreliable Wi-Fi.
Expanding beyond about 500 consumer families and care homes in the US, the UK and Japan.
How privacy is handled
Facilities get consent from residents or their power of attorney before installation.
KamiVision does not livestream or monitor rooms. It captures only the fall and its cause, which Link says video shows better than infrared or heat-map sensors.