ObjectVision is a fully offline Android application powered by a custom-trained YOLOv8 Nano model that detects objects in real-time at up to 60 FPS with 94.5% accuracy — and goes beyond detection by delivering AI-powered nutritional and ingredient analysis for recognized products.
🎓 B.Tech Final Year Project — ITM SLS Baroda University, 2026
🚀 Getting Started • ✨ Features • 📸 Screenshots • 🛠️ Tech Stack • 🤖 AI Model • 👥 Team
| Feature | Description |
|---|---|
| 🔍 Real-Time Neural Scan | Live camera detection at up to 60 FPS with green bounding boxes and confidence scores |
| 🧠 Custom YOLOv8 Nano Model | Trained on a proprietary product dataset — not generic COCO data |
| 📊 Neural Analytics Dashboard | Inference speed (avg. 212ms), total scans, CPU load, and 94.5% accuracy stats |
| 📋 Detection Logs | Full timestamped history of every detected object with confidence scores |
| 🥤 AI Product Analysis | Energy output, processing level, vegetarian/vegan status, and ingredient composition |
| 🧪 Nutrition Facts | Detailed carbs, fat, protein, sodium and allergen info per detected product |
| ⚙️ System Configuration | Toggle High Precision Mode, Offline Processing, Haptic Feedback, Battery Optimization |
| 🔒 100% Offline | All inference runs on-device — no internet, no data sent to the cloud |
| 🌙 Dark UI | Sleek dark theme built with Jetpack Compose for a modern, eye-friendly experience |
| ⚡ Battery Smart | Auto-reduces FPS when battery drops below 20% |
| 🌅 Splash / Boot | 🔍 Neural Scan (Live) | 📊 Neural Analytics |
|
 |
 |
| 📋 Detection Logs | ⚙️ System Config | |
 |
 |
| Product Analysis | Ingredients Composition | Nutrition & Allergens |
 |
 |
 |
| Product Analysis | Ingredients Composition | Nutrition & Allergens |
 |
 |
 |
├── Language → Kotlin (100%)
├── UI Framework → Jetpack Compose
├── AI Model → YOLOv8 Nano (custom-trained, Ultralytics)
├── Model Training → Python + PyTorch
├── Runtime → TFLite / ONNX (on-device inference)
├── Build System → Gradle (Kotlin DSL)
├── Architecture → MVVM + Jetpack Compose state management
└── Platform → Android 8.0+ (Oreo and above)
Key Tools & Libraries:
- 🤖 YOLOv8 Nano — Ultra-lightweight YOLO variant optimized for mobile edge computing
- 📱 Jetpack Compose — Declarative, reactive Android UI with smooth camera frames
- ⚙️ TFLite Runtime — Fast on-device neural network inference
- 🔬 Custom Dataset — Proprietary annotated product image dataset
- 🧱 Kotlin Coroutines — Non-blocking camera and inference threading
Camera Input
↓
Frame Preprocessing
↓
YOLOv8 Nano Model (TFLite)
↓
Bounding Box Detection
↓
Confidence Score Filtering
↓
Product Data Mapping
↓
Jetpack Compose UI Rendering
| Metric | Value |
|---|---|
| 🎯 Detection Accuracy | 94.5% on custom dataset |
| ⚡ Avg. Inference Latency | ~212 ms |
| 🎬 Max Frame Rate | 60 FPS |
| 📦 Model | YOLOv8 Nano (AI_MODEL_v4.2) |
| 🌐 Internet Required | None — fully offline |
ObjectDetection_system/
│
├── app/
│ └── src/
│ ├── main/
│ │ ├── java/ # Kotlin source (Activities, ViewModels, Composables)
│ │ │ ├── screens/ # Neural Scan, Analytics, Logs, Settings, Analysis
│ │ │ ├── model/ # Data classes (ScanSession, ProductData, Settings)
│ │ │ └── inference/ # YOLOv8 TFLite inference engine
│ │ ├── assets/ # YOLOv8 .tflite model weights + product database
│ │ ├── res/ # Layouts, drawables, strings
│ │ └── AndroidManifest.xml
│
├── build.gradle.kts # Root Gradle config
├── settings.gradle.kts # Project settings
└── gradle.properties
- ✅ Android Studio (Hedgehog or later)
- ✅ JDK 11 or above
- ✅ Android device / emulator running Android 8.0+
- ✅ Minimum 3 GB RAM recommended for smooth inference
- ✅ Device with NPU or GPU preferred for optimal performance
1. Clone the repository
git clone https://github.com/Hemanshu4949/ObjectDetection_system.git
cd ObjectDetection_system2. Open in Android Studio
File → Open → Select the cloned folder
3. Add the YOLOv8 model weights
Place your yolov8_model.tflite file in:
app/src/main/assets/
4. Sync Gradle & Run
Sync Gradle → Run on Device or Emulator
⚠️ A physical device is strongly recommended — the camera inference pipeline performs significantly better on real hardware than an emulator.
If you want to retrain the model on your own dataset:
# Install Ultralytics
pip install ultralytics
# Train YOLOv8 Nano
yolo task=detect mode=train model=yolov8n.pt data=your_dataset.yaml epochs=100 imgsz=640
# Export to TFLite
yolo export model=best.pt format=tflite| Setting | Description | Default |
|---|---|---|
| ⚡ High Precision Mode | Prioritize accuracy over speed | OFF |
| ☁️ Offline Processing | No data leaves the device | ON |
| 📳 Haptic Feedback | Vibrate on each object detection | OFF |
| 🔋 Battery Optimization | Lower FPS when battery < 20% | ON |
| Challenge | Solution |
|---|---|
| Dataset Creation | Manually collected and annotated a custom product image dataset |
| Mobile Optimization | Chose YOLOv8 Nano for the best accuracy-speed tradeoff on mobile |
| Real-Time Processing | Used Kotlin Coroutines + non-blocking Compose updates to maintain 60 FPS |
| Model Deployment | Optimized PyTorch → TFLite conversion to minimize precision loss |
- 📲 Barcode and QR code scanning for extended product databases
- 🗃️ Expanded custom dataset for more product categories
- ⌚ Wearable device integration
- 🎙️ Voice assistant support for accessibility
Follows the Agile Iterative SDLC Model, enabling continuous improvements across:
- AI model accuracy iteration
- UI/UX enhancement cycles
- Performance benchmarking sprints
git checkout -b feature/your-feature-name
git commit -m "Add: your feature description"
git push origin feature/your-feature-name
# Open a Pull Request 🎉⭐ If you found this project helpful, give it a star!
Made with ❤️ using Kotlin, Jetpack Compose & YOLOv8