Overview

This is an Android-based application for real-time human activity recognition using IoT sensors (Thingy, RESpeck). It leverages machine learning to classify activities and monitor real-time data.

---

Requirements

• Android SDK: API 34+ (Android 13 or higher)
• Minimum SoC: Snapdragon 625 or higher
• Minimum Storage: 1GB free space
• Minimum RAM: 512MB

Installation Guide

Step 1: Download the App

Method 1: Transfer APK from Computer

1. Download {PDIoT_Group_X1.apk} from the Learn page.
2. Locate the file (Windows: C:\Users\YourUserName\Downloads).
3. Connect your phone via a USB cable.
4. Set connection mode to File Transfers (MTP mode).
5. Copy the APK file to your phone’s storage.

Method 2: Download APK Directly on Phone

1. Open a browser on your phone.
2. Download the APK from the GitHub repository: PDIoT Group X1 APK Release

Step 2: Install the APK

1. Locate the APK file using File Explorer.
2. Tap to install.
3. If prompted, allow installation from unknown sources.
4. Bypass Play Protect warning if necessary.

---

Features

1. Login System

• Register a new account (username, email, password).
• Login with an existing account.
• Reset password if forgotten.

2. Main Menu Functions

• Pair Sensors – Connect IoT devices.
• Live Predict & Record – Monitor real-time activity data.
• Gather Raw Data – Save sensor data for analysis.
• View History – Access past recorded data.

3. Sensor Pairing

• Bluetooth Pairing: Connect Thingy or RESpeck sensor.
• QR Code Scanning: Scan sensor ID for quick pairing.

4. Live Activity Recognition

• Uses machine learning models to predict real-time gestures and breathing activities.
• Displays recognized activities as text and graphical icons.
• Saves results as CSV files for future use.

5. Historical Data Viewing

• Browse saved activity records.
• Integrated file viewer for CSV data.
• Delete unnecessary records.

---

Technical Details

Classification Model Performance (On-device)

Task	Accuracy
Task 1	96.48%
Task 2	85.71%
Task 3	72.03%

Supported Activities

• Walking, Running, Sitting, Standing
• Lying Down (Various Positions)
• Stair Ascending/Descending
• Coughing, Hyperventilating
• Miscellaneous Movements

Software Architecture

• Model-View-Controller (MVC) Structure
• Programming Language: Kotlin
• Database: SQLite (future upgrade to Firebase)

Machine Learning Model

• CNN-based model with a 4-layer architecture.
• Uses Batch Normalization, ReLU Activation, and Max Pooling.
• Evaluated with Leave-One-Subject-Out Cross-Validation (LOSOXV).

---

Future Improvements

• Expand classification to more human activities.
• Integrate cloud-based machine learning for faster model training.
• Migrate to Firebase for real-time database syncing.