Real-Time Outdoor Verification Engine

High-performance computer vision backend + cross-platform mobile app

---

🎯 Overview

Ventus is a real-time outdoor scene verification system that combines a high-performance C++ computer vision engine with a cross-platform mobile application. The system verifies whether users are actually outdoors by analyzing selfie photos in real-time.

Use Case: Wake-up accountability app that requires an outdoor selfie to prove you're awake—with computer vision ensuring authenticity.

---

🚀 CV Engine (C++, OpenCV, TensorFlow Lite, gRPC)

The core of Ventus is a custom-built computer vision engine optimized for low-latency mobile inference.

Performance

Metric	Value
Inference Latency	<50ms per image
Throughput	500+ verifications/day
Scene Classification	95%+ accuracy
Outdoor Categories	40+ scene labels

Architecture

┌─────────────────────────────────────────────────────────────────┐
│                     Ventus CV Engine                            │
│  ┌───────────────┐  ┌──────────────┐  ┌─────────────────────┐  │
│  │ Preprocessing │──▶│ Scene CNN   │──▶│ Verification Logic │  │
│  │   (OpenCV)    │  │  (TFLite)    │  │                     │  │
│  └───────────────┘  └──────────────┘  └─────────────────────┘  │
│         │                                       │               │
│         └───────────────────┬───────────────────┘               │
│                             ▼                                   │
│                    ┌─────────────────┐                          │
│                    │  Face Detector  │                          │
│                    │   (BlazeFace)   │                          │
│                    └─────────────────┘                          │
└─────────────────────────────────────────────────────────────────┘
                              │
                         gRPC API
                              │
                              ▼
┌─────────────────────────────────────────────────────────────────┐
│                   Flutter Mobile App                            │
│         Camera Capture → Upload → Display Results               │
└─────────────────────────────────────────────────────────────────┘

Key Features

• Custom CNN Model — Trained on 50,000+ images across 51 scene categories
• Multi-Factor Verification — Scene classification + face detection
• Optimized Inference — TensorFlow Lite with INT8 quantization
• gRPC Streaming — Supports batch processing for high throughput
• Cross-Platform — Runs on Linux, macOS, and embedded devices

Outdoor Scene Labels (40+ categories)

sky, outdoor, nature, tree, forest, park, street, road, sidewalk,
building_exterior, garden, beach, ocean, mountain, hill, field,
meadow, lake, river, waterfall, sunrise, sunset, cloud, sun, rain,
snow_outdoor, desert, canyon, cliff, bridge, parking_lot, playground,
stadium, campus, courtyard, patio, balcony, rooftop, trail, path

📖 Full Engine Documentation →

---

📱 Mobile App (Flutter)

Cross-platform mobile client with social accountability features.

Features

• 🔔 Smart Alarms — Customizable wake-up times with grace windows
• 📸 Camera Integration — Native camera for selfie capture
• 🔥 Streak Tracking — Calendar views and progress visualization
• 👥 Social Accountability — SMS notifications via Twilio
• 🔐 Secure Auth — Amazon Cognito authentication

Tech Stack

Layer	Technology
Framework	Flutter 3.35
State	Riverpod
Navigation	Go Router
Local Storage	Hive
Auth	AWS Cognito
Storage	AWS S3

---

📂 Project Structure

Ventus/
├── engine/                      # C++ CV Engine
│   ├── src/                     # Source files
│   │   ├── preprocessing.cpp    # OpenCV image processing
│   │   ├── scene_classifier.cpp # TFLite inference
│   │   ├── inference_engine.cpp # Verification pipeline
│   │   └── server.cpp           # gRPC server
│   ├── include/                 # Header files
│   ├── proto/                   # gRPC service definitions
│   ├── models/                  # TFLite model files
│   ├── tests/                   # Unit tests
│   └── CMakeLists.txt           # Build configuration
│
├── app/                         # Flutter Mobile App
│   ├── lib/
│   │   ├── core/                # Services, routing, constants
│   │   ├── features/            # Feature modules
│   │   └── shared/              # Shared widgets
│   ├── amplify/                 # AWS backend config
│   └── assets/                  # Images, fonts
│
└── README.md

---

🛠️ Getting Started

CV Engine

# Install dependencies (macOS)
brew install opencv grpc protobuf cmake

# Build
cd engine
mkdir build && cd build
cmake ..
make -j$(nproc)

# Run server
./ventus_server --port 50051 --model models/scene_classifier.tflite

Mobile App

cd app
flutter pub get
flutter run

---

📊 Benchmarks

Tested on Apple M1 MacBook Air:

Operation	Time
Image Decode (JPEG)	3ms
Preprocessing (resize, normalize)	5ms
Scene Classification	25ms
Face Detection	12ms
Total End-to-End	~45ms

---

🧪 Testing

# Engine unit tests
cd engine/build
ctest --output-on-failure

# Flutter tests
cd app
flutter test

---

🤝 Contributing

Contributions welcome! Please open an issue or submit a PR.

---

📄 License

MIT License — See LICENSE for details.

---

📧 Contact

Henry Fritz — GitHub

Project Link: https://github.com/hfritz34/Ventus