kaiyizhang-uo / 7163

Introduction

Hello, we are CSI7163 group 8: Jiacheng Hou (300125708) and Kaiyi Zhang (300070775).

This project applies two deep learning models on Human Activity Recognition (HAR) dataset [1][2], which is available in the UCI Machine Learning Repository. The two deep learning models are Graph neural network (GNN) and Long Short-term Memory (LSTM). We also convert the GNN and LSTM models to .tflite and deploy them on Android. Our application can predict a person's activities in real-time, including WALKING, WALKING_UPSTAIRS, WALKING_DOWNSTAIRS, SITTING, STANDING, LAYING.

Contributions

Jiacheng:

• Plot Data Visualisation
• Train a GNN model
• Deploy the GNN model on Android

Kaiyi:

• Train an LSTM model
• Deploy the LSTM model on Android
• Android application UI

Required Packages

implementation 'androidx.appcompat:appcompat:1.3.0'
implementation 'com.google.android.material:material:1.4.0'
implementation 'androidx.constraintlayout:constraintlayout:2.0.4'
testImplementation 'junit:junit:4.13.2'
androidTestImplementation 'androidx.test.ext:junit:1.1.3'
androidTestImplementation 'androidx.test.espresso:espresso-core:3.4.0'
implementation 'org.tensorflow:tensorflow-lite:+'

References

• [1] HAR Dataset. UCI Machine Learning Repository: Human Activity Recognition using smartphones data set. (2012). Retrieved March 8, 2022, from https://archive.ics.uci.edu/ml/datasets/human+activity+recognition+using+smartphones
• [2] Roobini, M. S., & Naomi, M. J. F. (2019). Smartphone sensor-based human activity recognition using deep learning models. Int. J. Recent Technol. Eng, 8(1), 2740-2748.