BFRB Aware

Mobile app as part of TOM France Hackathon 2023 (Team 6)

Requirements

In order to run an Androïd, the app requires JDK 11 to be ran.

Installation

• npm install

Running the app

• npm run android
• npm run ios
• npm run web

State of the app

React Native App

This repository contains a working React Native app. It displays the status of the user In crisis / Not in crisis as well as a list of contact, guidelines and history of crisises.

The application fetches a json wich contains:

• heartrate: contains the heartrate value with its associated timestamp
• position: contains a [x,y,z] vector which contains the movement values with its associated timestamp.

Currently, we are using the heartrate to flag a crisis if there is a difference of more than 10 between two values.

Technical architecture

Work to be done

The app currently does not use the position values. It could be done to identify the movement of the hand and detect the crisis more precisely.

Algorithms used in the app

Parsing

Raw data extracted from the smart watch are in ./rasp-api/input_data. The parsing algorithm to extract x, y, z and the heart rate with their timestamp is in ./rasp-api/algorithm/parsing.py. It extracts data into a JSON file in ./rasp-api/algorithm/json_outputs. You can all the parsing algorithm like this :

python3 parsing.py <path_to_input_file>

Crisis detection

A crisis is characterised by a higher stress level and a specific motion.

Heart rate detection

To detect a higher stress level, we decided to look at the heart rate. We are monitoring heart rate. Any increase of 10 bpm is considered a crisis and is detected through detectHeartRateIncrease, any decrease of heart rate back to base level is considered the end of a crisis and is detected through detectHeartRateDecrease. Both algorithm are in ./detectCrisis.js.

Motion detection

To detect the motion, we first explored identifying a pattern and isolating it to compare it to live data. But motion are never exactly the same, so we couldn't identify a clear pattern.

Inspiring ourselves from the gesture detection project on a Bangle.js watch (https://docs.edgeimpulse.com/experts/prototype-and-concept-projects/gesture-recognition-with-banglejs-smartwatch), we used Edge Impulse to train a neural network to detect the motion we want to identify(https://studio.edgeimpulse.com/public/252177/latest).

This neural network can be exported as a tensorflow model. Unfortunately we were not able to integrate it in the mobile app during the hackathon.