All of us have people in our lives we care about, but we can't always be near them when disaster strikes. Like many other elders, my grandmother lives alone, and if a medical emergency occurs there's nobody around to report it or help her.

1UP is a mobile application that can be installed on android smartphones without any external accessories, and runs in the background while the phone rests in the user's pocket. Whenever a user faints, a common symptom of many medical emergencies, it signals an alarm and notifies an emergency contact with an SMS message containing the location of the user.

Although sending an SMS with the phone's location is very easy (especially using RapidAPI), it's obviously not very easy distinguishing loss of consciousness from other activities using only the built-in gyroscope and accelerometers, so we decided to use machine learning in order to tackle that problem. We conducted dozens of experiments, gathering thousands of samples of everyday activities and many more samples of our team members falling to the ground.

Teaching an artificial neural network is very hard, and required a lot more samples than we could gather in a weekend. To get to the required amount, we used Unreal Engine, a video game physics simulator with built in motion capture animations, and generated even more samples.

With the required samples our team has gathered, we were ready to teach the network. That's when we ran into our most difficult obstacle: teaching an artificial neural network takes a lot of time. That is the one resource we don't have right now. With the amount of time we were given, we managed to teach the network to alert almost no false positives, and 66% of the true positive.

At the end of the day, we currently have a standalone app, that from the comfort of your pocket, can save your life, but we still have a lot more to do. Teaching the artificial neural network for a longer time is just the fist step of improving precision. We can also use the built in microphone, and even external peripherals such as Pebble. Expanding the app to other platforms such as iPhones and Windows phones is also an important milestone for the app that should be relatively easy compared to other obstacles we were able to overcome.

This is just the beginning, and we are more than excited to get 1UP out there and save lives.

Share this project:

Updates