Connecting AI Fall Detection With Care Workflows at KamiVision.

[00:00:04] What has to happen after an AI system detects a fall for the alert to be classed as care? The answer involves people, connectivity, privacy and an understanding of exactly what happened before somebody even reached the floor. Today I'm joined by Mike Link, Chief Operating Officer at KamiVision, to discuss computer vision in residential care and the work involved in adapting it for our own homes.

[00:00:32] We're going to talk about how the company combines device detection, cloud filtering and human review, along with the questions buyers should ask about accuracy claims in any system. And he will explain why room layout, lighting and everyday behaviour can change what a system needs to recognise. For care providers and business leaders assessing physical AI, today's conversation is about the process surrounding the technology.

[00:00:59] An alert on the screen is a starting point, but the value depends on whether the people receiving it can understand the event and respond appropriately in practice. But enough for me, let me introduce you to my guest now. We're going to talk about technology, its role in residential care homes and how it can improve looking after the elderly. So thank you for joining me on the podcast today. Can you tell everyone listening a little about who you are and what you do?

[00:01:28] Yeah, my name is Mike Link. I'm the COO of KamiVision and we operate an AI vision business, which means AI and cameras. We started probably 10 years ago with home security using AI to do, you know, people detection, dog detection, package detection. And that thing grew pretty quickly and then allowed us to build other things.

[00:01:51] One of which we'll talk about today is the elder care business where we can detect falls and let families know, staff know when somebody's on the ground in a room, which is pretty cool. And then the other thing which we're building is enterprise security.

[00:02:05] So, you know, in retail stores, in commercial real estate, using existing cameras to layer AI over to do either security or intelligence, maybe in like a grocery store, seeing where people spend the most time or maybe in a warehouse to do detections of breaking and entering or maybe inventory management or possibly even like a warehouse. And then the other thing is shelving, uh, deteriorating and breaking.

[00:02:34] So, um, yeah, we, we, we kind of do it all. Incredibly cool. And I've got to ask, cause I always love diving into the origin story of my guests. You talk about home security, package detection and retail store, uh, security. What was the first, what was it that first showed you that elder care needed a different approach to monitoring safety and, and also supporting independence? There's got to be a story there. Was there a moment that inspired that? Not necessarily.

[00:03:03] So I never intended to get into this. I actually studied, uh, neuroscience and wanted to become a neurologist or like a surgeon or something and ended up in India for a while studying alternative medicine, um, called Ayurveda. And when I came back to the States, I'd always done sales and I'd always had businesses, um, and I needed some money. So I ended up consulting a different company that was doing AI in the elder care space.

[00:03:32] And that's how I first learned about the challenge with elder care and that, you know, people fall and nobody finds out about it till like three, four hours later often, even in a facility. And it was an interesting problem. Um, and then, you know, we had some success there and then Kami had hit me up and we're doing something similar. And I thought that it was pretty cool that they were able to do similar technology or build similar technology, but for a fraction of the price.

[00:04:00] Um, and, and, and modify the technology very quickly because we didn't have any VC or any, you know, uh, outside investor to answer to because the home security business had done so well. We had, I wouldn't say like infinite money, but you know, basically millions of dollars that we can invest and then not have to answer to a VC, which is pretty unique.

[00:04:23] Um, a lot of the companies in the space, you know, the way that technology typically gets built and now it's a little different with this whole AI credits thing, but you know, a VC or an outside investor will come in, give some money unless the founder has some cash. Um, so in this case, the founder had some cash and was able to build the technology super quickly. And then because, um, nobody else was, you know, knocking on the door for the check, we could have a much lower price than some of our competitors.

[00:04:51] So, um, the challenge was that, you know, the technology in elder care is very expensive because of the reasons I just said. Um, and the other challenge was that it was hard to install and actually took, um, a lot of capital from a lot of, uh, communities and facilities up front to actually install the technology. The other thing was that in home, there was nothing that existed for consumer.

[00:05:15] So, you know, even though there was a lot of these companies selling into communities and facilities where like mom might go to a nursing home, um, there wasn't any technology for somebody at home. And I think the big shift was people wanting to not go into facility because it's not the nicest thing when you have to leave home and, you know, move all your things. And then you're basically in like, uh, an adult daycare. Yeah.

[00:05:42] Anybody that's got an elderly parent, grandparent, or family member, they will know about fall detection systems, things that could be worn on the wrist or, or around the neck. But for, for me, your experience here, where do those current traditional fall detection systems, where do they tend to fail in homes and residential care settings? Well, those are definitely good options, but the problem is if somebody doesn't remember to use them or if they need to be charged, then they're kind of useless.

[00:06:11] Um, and the idea is that, you know, somebody who's older doesn't have to remember to press something or ask for help in an emergency because, you know, I don't know if you have a mom or dad or had grandparents who maybe needed extra help. But I know for sure my grandmother would not remember, you know, in an emergency situation, she's not going to think, oh, let me press my button for help. You know, she's going to freak out and be panicking and then be stuck. So, yeah, there's nothing wrong with them.

[00:06:38] I think it's a good, they were a good start, but I think now the technology has evolved to a point where it's requiring less human assistance from the person in need. And that's the idea is that, you know, the person who needs extra care doesn't need to say anything, doesn't need to do anything. They could just exist, um, which prevents a lot of like user error. And then we can see based on how they live, what kind of help they need.

[00:07:02] And of course, Sammy, uh, KamiVision sites 99.5% accuracy, which is incredibly impressive. But how is that measured and what should buyers ask before trusting any accuracy claim in a real care environment? Yeah, that's a good question. So the 99% comes from the three layer detection system. So we have an edge chip on the device that detects the fall itself.

[00:07:28] So there's AI on the chip in the sensor. So that's the first layer. The second layer is after it is detected on the device itself, it goes to the cloud. And we have an AI model in the cloud that will filter out false positives. And then we have a third layer, which is actually a human layer. And it's the, the, the clip of a fall is blurred and encrypted.

[00:07:54] And a human can then see just the blurred clip, no personal identifiers or anything like this, and say yes or no, this is or isn't a fall. Um, and because the cloud filter and the edge chip get rid of a lot of the false positives, by the time it gets to the human, there's only a handful left. And then they can say, okay, you know, this is or isn't a real notification by the time it gets the user. It's not 9.5% accurate.

[00:08:22] And as it's, um, we're talking about cloud and connections and things. And a lot of elderly people won't have an internet connection. Is it SIM related? How's that, how is that connection managed? It's not SIM related. It's wifi based. We've thought about putting a SIM card in the consumer version in the future. Um, we haven't even fully released the consumer version. We've, uh, send it out probably to like 500 families to use.

[00:08:47] And, you know, that's how we've been, you know, building and, and kind of modifying the model because we have it in tons of care homes, um, in the U S in, in, uh, a bunch in the UK, some in Japan. And so the, the way that we have now started to modify that for in home, um, is by having families use it in home. Um, so right. You know, as of now you can't even go, you know, buy the consumer version.

[00:09:15] It's only in, um, community and residential settings, but yeah. Yeah. The, the SIM card is definitely the thing that we've looked at because in a lot of homes, especially in rural places in America, or, you know, I would assume also in the UK, um, wifi can be an issue and connectivity can be an issue. So, and even in some of the facilities, like they don't want to, uh, have wifi go down and then, you know, potentially miss something. But, um, at the same rate, like a lot of these facilities do have wifi now. So it hasn't really been too much of an issue.

[00:09:43] So when you've put all these in, what changes when computer vision has to operate reliably across millions of live devices, rather than just a controlled pilot, what have you seen? Has anything surprised you here? Nothing really. I mean, if you think about it, there's already millions of connected devices with Kami home. Um, we have more than a million paid users. There's probably three or 4 million, uh, free subscribers to the Kami home platform.

[00:10:11] So we already manage a huge amount of data. Um, so it's basically the same thing with the other system, the Kami care, elder care in communities and facilities system. And what kind of feedback have you had from some of those nursing homes or elderly homes? Have you had any feedback from around the world? I know you said the UK, Japan, US. So how's that feedback? The feedback is always kind of the same.

[00:10:38] It's kind of shock and surprise because when we go to sell the solution to some of these communities and facilities, they're like, well, we don't really have an issue with falls. They just kind of happen. And, you know, how is detecting the event going to help us prevent it? They kind of feel like it's reactive. Um, and I think the thing that's surprising to them is by seeing what happened in a reactive fashion, they're actually able to be proactive because most of the time these people aren't hurting themselves.

[00:11:05] Like 96% of falls aren't resulting in some kind of hip fracture or injury or event that makes them need emergency services. So the majority of falls that are detected are people putting themselves on the ground because they don't have the strength or the confidence to walk and do the things they need to do. So what happens a lot of the time is either somebody puts themselves on the ground or they scoot off the bed onto the ground.

[00:11:33] And then what happens in a facility is if a staff member comes in, imagine you have a memory care patient who's got dementia or a memory care resident who has dementia and you find them on the ground. They can't remember why they are on the ground. And then as somebody running the facility, you have to document this as a fall. But in fact, it wasn't a fall. And then when they can see the clip of the fall, they go, oh, they just scooted onto the floor. They actually wanted to put their shoes on, but they sat on the floor to put their shoes on because they didn't have the strength to stand up.

[00:12:01] And now they can send a nurse in there when they need to put their shoes on or when they need to dress or change or whatever the thing is. So it's these kind of like environmental changes, maybe in like the room layout or activity changes in their routine where maybe a staff member can help them is where the realizations happen. And then the other thing is you can start to see as somebody's declining because they'll start to go to the ground, right? It's not necessarily a fall, but it's detecting the person on the ground, which allows you to

[00:12:28] see the meaningful change and then, you know, make a change in the way that you care for them. And I'm curious, how do you reduce false alarms while still detecting the accidents that require a rapid human response? Tell me a bit more about that. And I would imagine that's somewhat of a challenge or can be. It definitely was. I think for the first handful of years, we had a lot of false events because the model was being trained. But, you know, if you imagine, you know, tens of thousands of falls, um, it's probably

[00:12:59] you know, maybe 2023, maybe 2024. Um, when we were able to actually hit this accuracy number, because we had enough falls that the data was being trained on. Do you need AI agents that you can trust? Well, with an AI data layer providing real time connection within your data platforms, you can trust your agents to provide accurate solutions.

[00:13:25] So scale your business by trusting your agentic AI accurately getting the work done for you. Trust its capabilities with Denodo. And you can do that by simply visiting denodo.com to learn more. And how do you think the tech should differ between living independently at home and a resident in a managed care setting? I know you've only got the one at the moment, but you're obviously looking at both. Is there a big difference there? There is kind of a big difference.

[00:13:55] I think some of the challenges are a little bit different. If you think about in a facility, it's very controlled versus in home. You have a lot of different objects, people, lighting. Um, so in a way, the community is kind of like a test facility. You know, the lighting is typically the same. All the rooms are the same. Typically it's a one bedroom or a studio. Whereas a home, maybe you have four bedrooms, maybe you have, you know, dim lighting, or

[00:14:23] maybe you have really bright lighting, or maybe you have a dog running around. And these are all things that the model has to differentiate. So I think that's the next kind of challenge is how do you make fall detection in home just as accurate? The other thing is the visual sensor isn't able to see everything always in the way that you set up in the home. Um, and sometimes you need like five or six devices. So the thing we've been looking at is like, what kind of ancillary device will we be able

[00:14:52] to connect with the visual sensor that we can like say, okay, the visual sensor can't see behind the couch over here, but there's wifi radar mapping technology where we can see if a body is somewhere in the room. Um, so maybe when the camera or the visual sensor or, you know, something else wouldn't be able to see in one area, maybe we could use a different type of technology to detect the other area.

[00:15:21] Um, so yeah, we've been looking at a handful of things like radar. Um, we've looked at potentially using like robot vacuums or like mini drones in the home, um, as well as wearable devices. Uh, we looked at a ring, um, in 2025 to potentially see if we could detect when somebody's heart rate like drops significantly, um, in, in home, because maybe this would be indicative of all,

[00:15:49] but there's a bunch of other devices that we've looked at to try to cover spots that we might miss in a home that we could easily get in the care setting with, with one type of technology. And if we do have anybody listening in, uh, maybe they're a care provider or work in the industry for, and they're looking for the ROI on this, what is the business case for a care provider and what results show the tech and any technology is improving outcomes rather than just adding another dashboard that they need to monitor? Yeah, that's a great question.

[00:16:19] I think this has been one of the biggest challenges in growing the business and the residential care setting because the sad truth is that people don't get paid more to provide better care. They get paid to fill the rooms. Um, and you, you come to realize this when there's a question of ROI with this type of technology, because anybody who goes to a facility goes to a facility because they can't live at home independently. And if that's the case, they are a fall risk.

[00:16:48] Um, I would say in an independent living setting where somebody's doing their own thing and maybe has, you know, a check every couple of days or doesn't have checks and just calls for assistance when needed, you know, maybe it's not super necessary in that kind of setting, but in an assisted living or in a dementia care setting, um, every resident really should have this because this is an insurance policy for the building. You know, imagine somebody falls and you don't get there to the morning as the facility, right?

[00:17:17] You face huge scrutiny from the family that, Oh, I pay you monthly. I don't know how much it is now in the UK, but in the, in the U S it's anywhere between 4,000 and $10,000 a month. And families are paying this per month and they go, well, you didn't have staff checking on my mom throughout the night, you know? Um, so I think the, the real ROI comes when someone has faced that scrutiny.

[00:17:46] When they've already experienced a family who's been really disgruntled or has had, or they've had a fall in a facility where they haven't gotten there for a long time. I think that's when they start to realize the ROI. Um, I think the other thing is that the less people have injury in a facility, the less likely they are to move out. And this is one of the things that we found is that, uh, typically like what I said, the falls aren't resulting in an injury at first, but when they can see the falls at first and

[00:18:15] can get ahead of them, they could prevent falls with injury. And that significantly decreases the, the individual moving out because often what happens after a couple of falls, the family will choose to move the resident out into a new setting because they feel that the facility or the care, the care staff can't care for their mom. Because imagine you get a call at, you know, three, 4am say, Hey, your mom fell. We're taking her to the emergency room. We don't know what happened. You know, you're probably going to get fed up on, you know, you're going to be like, well, what do I do now? Right.

[00:18:43] Like I've trusted this facility to care for my mom or dad and they can't. Um, so yeah, I would say that the other thing is like a lot of people moving in or going to a care facility, they go there because of falls at home. Um, like my grandfather, for example, went to a facility because he was having falls at home. And when facilities have something to say, Hey, you know, we actually have this Kami care system when your mom or dad falls, you know, it notifies us within 60 to 90 seconds that they're on the ground.

[00:19:12] We can get there and care for them and see what happened and communicate this to you. It's a huge selling point. And it's not like a, we guarantee every single time we're going to get there in this amount of time, but the technology is a buffer so that when a care staff is walking around a 50 room building and it takes them two and a half, three hours to do their rounds that if somebody is on the ground, they'll get notified and they can prioritize that because they know that this person needs help.

[00:19:39] Um, so I think, you know, the challenge is there's kind of an intangible ROI to some extent because the ROI is preventing falls, right? So if we're getting rid of all the falls, then what you'd say, well, where's the ROI? Nothing's changed. There's no, like the goal is to prevent something bad from happening. So if the bad thing doesn't happen, there's no ROI, right?

[00:20:04] So the thing that we've been doing now is modifying the system to try to save staff time and save other things on a, on the way that the business operates so that we could actually show the ROI. For example, helping to fill out reporting so that staff, when they actually do get a fall, they could do way less work, you know, um, starting to take different, uh, you know, we haven't deployed this to facilities and communities, but the hypothesis is that we

[00:20:34] could take a lot of different medical data, you know, core mobilities and medication lists and vitals and maybe activities and say, okay, well, this person is becoming a fall risk. Right. And then we could say, well, we've actually detected X number of things that your staff might not have detected and been able to make a change that helped prevent something. And then we can document this in the system and say, Hey, listen, this is actually a huge ROI, right? Like we caught this thing that would have happened because of this medication change

[00:21:04] and now prevented an ER visit, which is a huge ROI, you know? So, uh, you know, that, that's kind of the, the, the ROI thing. I don't know if that makes sense or not, but. Yeah, it really does. And also for people listening, maybe considering something like this, are there any privacy consent or even staff training questions that typically come up, uh, an organization need to resolve

[00:21:28] or get in place before thinking about deployment, any big conversations there or FAQs that you get a lot? Yeah, we, we have all the facilities get consents from the resident or the power of attorney. Um, the other thing is that we stress that we don't live stream. We don't, we're not looking into the room. We're not monitoring into the room. Um, the visual, the visual sensor is only meant to see the cause of the fall and exactly how it happened.

[00:21:56] Um, for example, there's a lot of infrared or, uh, you know, non video based sensors out there. But the problem is, is you can't confirm whether somebody's hit their head or not hit their head. If it's a stick figure in the clip that you get, or if it's a infrared sensor where you could just see a heat map or something like this, you can't see if they were reaching for something. You can't see objects in the room. Um, so really like the video is a huge value add.

[00:22:24] Um, but it does cause concerns about like live stream or recording or things like this. And this is why we get consents. But when the staff and the family members know that the only thing captured is the fall and the cause of the fall. And the reason why, um, and the benefits to this, it usually there's, there's no type of, uh, pushback. Um, but yeah, the consent is definitely required. Well, for anybody listening, would like to find out more information. As you said, you serve people around the world right now.

[00:22:51] If they want to dig a little bit deeper, uh, connect with you or your team, or just have a browser. Well, it looks like how it works, et cetera. Where should they go? Um, they could go and search, uh, Kami care on Google and, uh, it'll come up and, you know, they're happy to reach out to us and we can give a demonstration of how the technology works or set up a meeting or anything like this. So, um, that's probably the best way.

[00:23:16] Well, I cannot thank you enough for coming on today and sharing a, a grounded non-hyped breakdown of what it takes to execute and scale physical AI and modern environments. I love what you're doing and for all the right reasons too. So I will add links, videos to everything that we've talked about today. I urge people to check those out in the show notes, but more than anything, Mike, thanks for coming on and bringing all this to life and a language everyone can understand. Really appreciate your time. Yeah. Thanks for having me. I appreciate it.

[00:23:46] One of the big points I'm taking away today is that detecting an event and improving care, they're all connected through a human workflow. Someone will always need to receive a notification, understand its context and decide what help is required. And also appreciate the discussion about how different a home can be from a residential care home. Lighting, furniture, connectivity and areas a sensor cannot see all that will impact deployment.

[00:24:14] So for anyone evaluating a system, I would want to understand how performance is measured, what happens when something is missed and how consent and access are managed. And I think Mike's account today gave us so many useful questions to bring to those conversations and he answered them all around Camivision too. But he also brought it to life with some examples of how event information can inform changes to care routines.

[00:24:41] A big thank you to Mike Link today for joining me and sharing the story here and the human behind the tech too. But over to you, share your experiences. What would you need to see before trusting a computer vision alert as part of a care workflow for someone? techtalksnetwork.com. That's where you'll find everything. 4,000 interviews, eight podcasts. And you can get a hold of me by connecting with me on LinkedIn, Instagram, X, just at Neil C. Hughes.

[00:25:10] Send me a DM there. Send me an audio message from the website. Whatever it is, I want to hear from you. So, well, I have taken up far too much of your time. I'll be back tomorrow, but bye for now.