https://github.com/devyansh22156/alertdrive-android-application
AlertDrive is a real-time Android application that monitors driver fatigue using live camera feeds, employing a lightweight EfficientNet‑B0 + GRU deep learning model served via a Flask backend to deliver rapid and accurate drowsiness alerts.
https://github.com/devyansh22156/alertdrive-android-application
android-application computer-vision deep-learning flask kotlin-android python
Last synced: 4 months ago
JSON representation
AlertDrive is a real-time Android application that monitors driver fatigue using live camera feeds, employing a lightweight EfficientNet‑B0 + GRU deep learning model served via a Flask backend to deliver rapid and accurate drowsiness alerts.
- Host: GitHub
- URL: https://github.com/devyansh22156/alertdrive-android-application
- Owner: devyansh22156
- Created: 2025-04-28T21:21:34.000Z (about 1 year ago)
- Default Branch: master
- Last Pushed: 2025-04-28T21:38:04.000Z (about 1 year ago)
- Last Synced: 2025-04-30T05:04:39.474Z (about 1 year ago)
- Topics: android-application, computer-vision, deep-learning, flask, kotlin-android, python
- Language: Kotlin
- Homepage:
- Size: 529 KB
- Stars: 1
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# 🚗 AlertDrive: Driver Drowsiness Detection Android App
**AlertDrive** is a real-time Driver Drowsiness Detection application designed to help prevent road accidents caused by driver fatigue.
It uses a deep learning model deployed via a Flask backend and an Android app built with Kotlin, Jetpack Compose, and CameraX.
## 📸 Features
- Real-time driver monitoring via phone camera.
- Predicts **"Non-Drowsy"**, **"Drowsy"** and **"No Face Detection"** or states based on facial cues.
- Displays prediction confidence and extracted features.
- Alarm System, gets activated when driver feels **"Drowsy"**
- Minimalistic and intuitive UI for quick usability.
- Lightweight deep learning model combining EfficientNet and GRU for fast and accurate predictions.
## 🏗️ System Architecture
- **Backend (Flask Server)**:
- Receives camera frames from the app.
- Preprocesses and predicts driver state using a CNN+GRU model.
- Sends prediction label, confidence score, and feature vector back to app.
- **Frontend (Android App)**:
- Built with Kotlin and Jetpack Compose.
- Uses CameraX API for real-time frame capture.
- Automatically sends frames periodically to backend for prediction.
- Displays results to the driver.
## 🛠️ Setup Instructions
### 1. Clone the Repository
```bash
git clone https://github.com/devyansh22156/DriverDrowsinessDetection-Android-Application.git
cd DriverDrowsinessDetection-Android-Application
```
### 2. Set up the Flask Backend (Server)
Navigate to the `App_Server` directory:
```bash
cd App_Server
```
Install the required Python packages:
```bash
pip install -r requirements.txt
```
Run the Flask server:
```bash
python app.py
```
> **Important**:
> - Make sure your server is running on the **same network** (WiFi or Hotspot) as your Android phone.
> - Note your **IP Address** (for example, `192.168.1.5`) — you will need it in the next step.
### 3. Set up the Android App (Frontend)
Open the `app` directory in **Android Studio**.
- Open `app/src/main/java/com/example/driverdrowsinessdetection/ApiClient.kt`
- **Update the `BASE_URL`** variable with your computer's IP address and port (default Flask port is 5000):
```kotlin
private const val BASE_URL = "http://:5000/"
```
**Example:**
```kotlin
private const val BASE_URL = "http://192.168.1.5:5000/"
```
### 4. Build and Run the App
- Connect your Android device via USB with **developer mode** and **USB debugging** enabled.
- Press **Run** (▶️) in Android Studio to install the app on your device.
- Alternatively, **build an APK** and install it manually:
- Build → Build Bundle(s) / APK(s) → Build APK(s)
- Install the generated APK on your Android phone.
### 5. Using AlertDrive
- Open the app, login (dummy login currently), and navigate to the detection screen.
- Press **Start Auto Detect**.
- The app will automatically capture frames every 0.5 seconds and display predictions.
- If the driver is detected as **drowsy**, appropriate feedback can be triggered.
## 📚 Dataset Used
- **NTHU Driver Drowsiness Detection Dataset (NTHUDDD2)** — 66,500 images annotated for drowsy/alert conditions, with various lighting conditions and occlusions like glasses.
## 📈 Model Details
- **Spatial Feature Extractor**: Pretrained EfficientNet-B0 (ImageNet weights).
- **Fine-Tuned model weights**: https://drive.google.com/file/d/1gO5yMIWkIXpJ8abINwDpvyhF6pEXJJEz/view?usp=sharing
- **Temporal Feature Model**: Gated Recurrent Unit (GRU) layer on extracted spatial features.
- **Training**:
- Optimizer: Adam
- Loss: Binary Cross-Entropy
- Early stopping and data balancing techniques used.
- **Achieved Accuracy**: ~99.65% on validation set.
## 📱App Screenshots
Below are key app screens
### Login screen
### Dashboard screen
### Drowsy Detected
### Non-Drowsy Detected
## 🚀 Future Improvements
- Add interactive alert dashboard (e.g., vibrations, sounds).
- Explore Transformer-based temporal modeling.
- Integrate fatigue risk scoring over longer driving sessions.
- Extend support to wearable devices (like smartwatches).
## 🧑💻 Authors
- **Anikait Agrawal (2022072)**
- **Devyansh Chaudhary (2022156)**
- **Dhawal Garg (2022160)**
## 📄 License
This project is for academic and research purposes only.