F2D: Fall detection with a smartwatch.

F2D is a fall detection system running on a smartwatch. Unlike traditional systems that need a base station and an alarm central, F2D works completely independently, protecting so its user at home but also outside. It uses the current context as well as the user profile to offer more accuracy in the fall detection process. Running on a standard smartwatch, F2D is less stigmatising for the user and can be offered for less than half price of existing systems on the market.

Common fall detection systems are based on a sensor detecting a strong vertical acceleration. But smoother falls (the person grabs a chair, a table, or any other object) aren't detected while some particular situations (a user that seats "brutally" on his sofa) lead to false alarms. To distinguish only the critical situations, we will take into account contextual data, like what happened before the fall, after the fall, at which place in the room, at which time of the day, or next to what piece of furniture. A user's profile, continuously updated with his habits or particular behaviours, will bring some additional data to the user's context. The second key element is that it is the first fall detection system working on an independent smartwatch. There is no base station (which limits the range), no central alarm station (easier to manage) and works on a common and standard smartwatch (less stigmatizing, quite cheap, easily extendable...). The system can be summarized by the following picture:

  • Sensors. We collect raw data from the different sensors, mainly from the accelerometer and the gyroscope.
  • Filters. We filter the raw data from the sensors in order to detect patterns corresponding to a fall. The filters have parameters that modify their behaviour. And these parameters will be set and updated dynamically according to the user's context and the user's profile.
  • Context. The current context (position, time, ...) is used in two different manners. The first is to define the variables of the filters in order to differentiate better the possible (expected) patterns. For instance, we expect to see a pattern of a user "falling" on his bed only if he is in the bedroom. The second is to help the decision module. Once a potential fall is detected by the sensors, is it likely that it is really a fall according to the current context? For this project, we assume that the position of the smartwatch (and therefore the user) is always known. It will use the Bluetooth fingerprinting technique, a solution that we develop in the framework of another research project.
  • Profile. The profile of the user (age, health problems, tendency to fall mostly during night when going to bathroom...) will be taken into account in order to set the parameters of the filters and help the decision module in case a fall is detected, working in a similar way as for the context.
  • Decision. The decision module combines the data coming from the filters, the current context and the user's profile, in order to decide whether it corresponds to a fall. If it is the case, the information is transmitted to the alarm module. For instance if the user had to get up very often during the last night, resulting in fatigue, the probability of a fall is higher; in this case the threshold between true and false alarm can be moved accordingly to be sure not the miss a real fall.
  • Alarm. When a critical situation is detected, the smartwatch uses different communication means (WiFi, 3G, SMS...) in order to inform the caretakers (family, friends, official caretakers...) about the situation. It is done in a similar way than with the traditional systems except that the smartwatch communicates directly with the caretakers (without passing by a base station and an alarm central). In case of an alarm the loudspeaker of the watch is automatically turned on at maximum volume and calls from caretakers are automatically answered, allowing so to communicate even in uncomfortable positions that could result from a fall.

Choosing a smartwatch as main device adds significant advantages over other more traditional systems, like a sensor belt. Firstly, some of the targeted users are also needing other applications, like for instance a geofencing app, and regrouping all these on a single device is of course quite practical.

Secondly, using a smartwatch allows the system to work in a bigger area. Indeed, a classical system is usually composed of a wearable device (to detect the fall) connected wirelessly to a base station (to transmit the alarm). But this works only when the user isn't too far from the base station. If the user falls in his garden or while walking towards the opposite end of the retirement home (to visit a friend), the system couldn't work correctly.

Thirdly, everything works on a single watch. Removing so the base station and the alarm centre, we can provide a solution that is simple to deploy and to manage, and that is significantly cheaper since the only hardware we need is a mass-market smartwatch (the price is about 300 CHF). And for the end-users the advantage is that the solution is much less stigmatizing, since a smartwatch is worn mostly be young and healthy people.

And finally, the commercial solutions are usually closed systems. No improvement or link with another system (fall detection or not) is usually possible. However, we observe that IT solutions are more and more used to improve the quality of life of people that don't benefit from all their faculties. The FST for instance is working on solutions like James4, a remote control for wheelchair running on a smartphone. Another example is the AAL European research program, whose aim is to provide the older adults with IT solutions. For the FST, having the full control of their software solutions (like this fall detection system) in order to combine them and let them evolve with the time represents an obvious advantage.

Project details

Project website
http://tam.unige.ch/projects/f2d.html
Start date
Tue, 04/01/2014
End date
Sat, 10/31/2015
Funding
Project participants
Project participants teams
shadow