README.md

Har_PytorchAndroid

This is the official repository for the Android Part of the HAR app

This Repo is part of the porject/community Mobile applications for non intrusive medical surveillance using Human Activity Recognition Technology. This REPO in particular aims to combine the different HAR approaches with an intelligent post processing of the recognized activities and implement the whole process in a usable android application.

• PytorchScholarAndroid-GitHub-Structure --> For now the most important parts can be found in:

1. The Data-Loading from the UCI in the Notebook of HAR_LSTM and HAR_CNN
2. The RNN-Model in the HAR_RNN_CNN (the training as har_rnn.py and the saved model as har_rnn_state.pth)
3. The Android Application in the HAR-PyTorchAndroid-Repo

• How to run the project

1. Clone this REPO
2. IMPORT the whole project into Android Studio
3. Run it on your mobile phone

--> Downloadable .APK will be uploaded soon
UNFORTUNATELY the conversion from pytorch to android hasn't quite worked until now!

• What are the various command line parameters which could be changed and how to change them --> For now only a basic implementation using LSTMs for the HAR without any Post-Processing is implemented

1. In the future one can choose between different deep learning approaches for the HAR-Algorithm
2. In addition the Post-Processing algorithm will allow taylored alarms for different critical medical conditions depending on the specific persons that needs surveillance.

---

The App - a visual

Normally the Gyroscope also works (In this case the gyroscope of the specific smartphone doesn't seem to work)