Nobi is a device that monitors an older persons movements 24/7. It knows if they may be ill, need to stay hydrated, or have fallen. In the case of a fall, Nobi immediately alerts the caregiver and emergency services and unlocks the door when help arrives.
The Nobi smart lamp’s mission is to enable older adults to live at home comfortably for as long as possible, by using fall prevention and fall detection tech. It has other care and comfort functions to make independent living possible. It blends smart home and elderly care technology to provide these services to users.
In, Episode 36 of 15 Minutes With The Doctor, Vinay speaks to Roeland Pelgrims, co-founder of Nobi. Roland was previously on the board of directors for a large care company in Europe and CEO of a smart home company. He was able to fuse his knowledge from these two fields to create a solution that could help tackle how to provide the right in-home care for older adults. Roeland explains how massive leaps in machine learning and processing power in the last decade have allowed the Nobi lamp to provide a high level of tech that significantly decreases an older person’s risk of injury from falls. Today, we share an unreleased section of the episode where Roeland shares how Nobi is using AI and image processing in fall recognition:
Vinay: I’ll come to the price in a second, but I wanted to touch base on what you said about the AI. I presume you had to have a lot of images and training for the artificial intelligence in terms of what a fall looks like in an image processing kind of way…. how many images did you use, and how long did it take?
Roeland: Millions of images.
That’s a process of several months. Our AI has various levels of analysis. So, it is about both detection, object recognition, and of course, it is about behavioural patterns. Then more specifically, falls. At first sight, a fall might seem a relatively simple phenomenon, and if you would study it from the perspective of let’s say, the average 30 year old, it probably is. It involves a person standing up and then accelerating massively, and then hitting the ground. So ending in a relatively horizontal pose. But, when you start analysing for behaviour amongst the elderly then you see a huge variety of cases.
AI sounds very high tech, which it probably is as an end result. But, training AI is still pretty basic and manually laborious. In essence, it involves a person looking at a screen and clicking on ‘it’s a fall,’ or ‘it’s not a fall,’ and then the computer trying to keep up until it reaches a reliability score of above 99%. Since we are in the realm of potentially life-saving technology, we want to have very very high reliability – So complete absence of false positives and false negatives. That pushes us then to train and constantly retrain the model.
We will, by the way, retrain continuously so every detected false positive will be a new learning moment for the model. We will make sure that there is no false positive experience for our users – When the lamp sees a fall, we then have 30 seconds where the lamp asks the user if there has been a fall? This second line anonymised verification has two advantages. The first one is that our users will not be confronted with a false positive and the subsequent actions that the device can make for the user. The second one is that we are in a continuous training mode. Only when we notice that the levels of false positives have dropped to absolute zero, will we switch modes and move to a sample base. Until then, every single occurrence will be double-checked to ensure that we can deliver on our core mission of answering the claim that there will be no fall that goes undetected.
Vinay: I’m really interested in the fact that you’ve already used millions of images and that it’s 99% accurate. I like that because of the way the system is developed; you very quickly get a real-life validation by asking the user whether they’ve fallen or not. That further validates the algorithm and can learn further from the new images coming in. As we said earlier, there’s so many types of falls, and each fall is quite complex inself.