Link to our official documentation for the MIT 2023 CRE[AT]E CHALLENGE

Navigating this Repository

This is where we organize the code for our project. This is a README file in the repository. It informs you what this repository is all out. Repositories track the code and its history. Code can be managed and changed in different versions. To nagivate changes and different versions, Github uses a system of managing branches and utilizing pulls and commits Most updated branch is the main branch

Overall Process:

For running machine learning

1. Upload FallAllD2.csv and machine learning.ipynb to Google Colab
2. Run machine learning.ipynb to create the machine learning model. Right now models are already created so you don't need to do this step. For running prediction script
3. run sensordata.py to run the prediction script. it will automaticaly collect data from the raspberry pi's sensehat

For Connecting Raspberry Pi

link to instructions

1. ctrl p (microsoft)

2. command shift p (linux)

3. remote ssh either connect or add new host

   • username@ip address

4. input password

5. connect to linux

6. enter password

7. on raspberry pi terminal:

   • connect to wifi
   • ifconfig

8. to run on local terminal:

   • open terminal type:
   • python3 name of file
   • to stop file:
   • ctrl c

Files and Descriptions:

all models

folder that holds all the models

• teacher models
  1. teacher_model4_device2_aggregate is the current teacher model AND model being tested on sensordata.py
  2. teacher_model0_device2_aggregate is the teacher model with all three sensors

• student models they all hold aggregate data from device 2 (wrist data) they are numbered with from 0-6. a number corresponding to how many sensors are configured. Below is the pseudo code snippet that you can also find in machine learning.ipynb (in python) that explains the layout

  0: ['Device','Acc_x','Acc_y','Acc_z', 'Gyr_x', 'Gyr_y', 'Gyr_z', 'Bar_x', 'Bar_y']

  1: ['Device','Acc_x','Acc_y','Acc_z', 'Gyr_x', 'Gyr_y', 'Gyr_z']

  2: ['Device','Gyr_x', 'Gyr_y', 'Gyr_z', 'Bar_x', 'Bar_y']

  3: ['Device','Acc_x','Acc_y','Acc_z', 'Bar_x', 'Bar_y']

  4: ['Device','Acc_x','Acc_y','Acc_z']

  5: ['Device','Gyr_x', 'Gyr_y', 'Gyr_z']

  6: ['Device','Bar_x', 'Bar_y']

FallAllD:

a folder that holds the raw data from FallAllD dataset. It is stored in csv files

M. Saleh, M. Abbas and R. L. B. Jeannès, "FallAllD: An Open Dataset of Human Falls and Activities of Daily Living for Classical and Deep Learning Applications," in IEEE Sensors Journal, doi: 10.1109/JSEN.2020.3018335.

scripts

a folder that holds the description and code for the data extraction from the FALLALLID dataset

data_extract.py

extracts data from all csv files from FallAllID and turns it into a panda dataframe on a csv file for machinelearning.ipynb this was modified from FallALlID's data extraction file.

Plot_FallAllD_Record.m

file from FALLALLID website that provides an example of the data extraction (From their original website)

Sensor and Sound Description

a text file that describes the methodology of the sensors and sound in our product

machine learning.ipynb

machine learning modeling on jupyter notebook if your device doesn't have GPU, you can upload the jupyter notebook onto Google Colab In different branches there are different models. Right now there are 4 models:

1. Simple Neural Network Model
2. LSTM Model with Kaggle Data
3. LSTM Model with FallAllD data
4. LSTM Model with Knowledge Distillation [MAIN BRANCH]

   sensordata.py

   where Accelerometer and Gyroscope data is collected using raspberry pi's sensor then, the data is run through the model and predictions are made

   test_student_model.py

   testing python file to debug our student model

   What is the MIT Create Challenge?

Our Project

Summary

Create an assistive piece of technology that can sense when a senior is in danger, and contacts for help

The base product works without wifi connection The entire product works will work with wifi connection

Major Components

There are three major components 1) Sensors that detect falling movement 2) A device that connects the sensor to a communicative application 3) The program

to articulate: as of now we are planning to program either in Raspberry Pi 4 + Python

Sample Projects

Please look at these sample projects to understand what we are doing:

Our recent slideshow Understanding Rasberry Pi

• Intro Project w/ Raspberry Pi
• Raspberry Pi Documentation
• Raspberry Pi Official Website

Citations

M. Saleh, M. Abbas and R. L. B. Jeannès, "FallAllD: An Open Dataset of Human Falls and Activities of Daily Living for Classical and Deep Learning Applications," in IEEE Sensors Journal, doi: 10.1109/JSEN.2020.3018335.

FallAllID Dataset