AI-Powered Road Damage Detection
Autonomous FPV drone system with real-time AI for road infrastructure maintenance.
Project Mission
Traditional road inspection is time-consuming, costly, and often inconsistent. This project combines autonomous drone technology with edge AI to deliver efficient infrastructure monitoring.
Key capabilities:
- Autonomous Flight — Pre-programmed paths with GPS tracking for systematic road coverage
- AI Detection — Real-time pothole detection using YOLOv8 deep learning model
- Edge Computing — On-board processing with Google Coral TPU for fast inference
- Live Monitoring — FPV transmission for real-time oversight and control
System Architecture
┌─────────────────────────────────────────────────────────────┐
│ FPV DRONE PLATFORM │
│ ┌──────────────┐ ┌────────────────────┐ │
│ │ Camera │─────────────▶│ Raspberry Pi │ │
│ │ Module │ │ Zero 2 WH │ │
│ └──────────────┘ └─────────┬──────────┘ │
│ │ │
│ ┌──────────────┐ ┌─────────▼──────────┐ │
│ │ Flight │◀────────────▶│ Google Coral │ │
│ │ Controller │ │ AI Accelerator │ │
│ └──────────────┘ └────────────────────┘ │
│ │
│ ┌──────────────┐ ┌────────────────────┐ │
│ │ GPS │ │ FPV TX/RX │ │
│ │ Receiver │ │ System │ │
│ └──────────────┘ └────────────────────┘ │
└─────────────────────────────────────────────────────────────┘
Current Detection Capabilities
The system currently implements pothole detection using a YOLOv8n (nano) model trained specifically for identifying road surface depressions and holes. The model is optimized for real-time detection on edge devices like the Raspberry Pi Zero 2 WH with Google Coral TPU acceleration.
Currently Detected:
- Potholes — Bowl-shaped depressions and holes in road surfaces
Technical Details:
- Model: YOLOv8n (nano variant for speed and efficiency)
- Format: TensorFlow Lite with INT8 quantization
- Inference: Hardware-accelerated via Google Coral USB Accelerator
- Performance: Real-time detection with ~10x speed improvement on Edge TPU
Future Extensions: The system architecture is designed to be extensible for additional road damage types including longitudinal cracks, transverse cracks, alligator cracks, rutting, bleeding, and weathering. Additional damage types can be added by training on appropriate datasets.
For detailed information about the AI model, training process, and technical specifications, see the AI Model Documentation.
Documentation
- Project Overview — Concept, problem statement, and approach
- Hardware Setup — Assemble and configure drone components
- Software Installation — Install AI framework and tools
- Camera Control — RC-triggered recording system
- AI & Datasets — Train and deploy detection models
- Tutorials — Step-by-step guides
Technology Stack
Hardware
- FPV Drone Frame
- Raspberry Pi Zero 2 WH
- Google Coral USB Accelerator
- Camera Module
- GPS Receiver
- Flight Controller (INAV/ArduPilot)
Software
- OS: Raspbian/Ubuntu
- AI: YOLOv8n (nano), TensorFlow Lite, INT8 Quantization
- Flight: INAV, ArduPilot
- Camera: Picamera2
- Languages: Python, C++
- ML Framework: Ultralytics, PyTorch (training), TFLite (deployment)
Supervised by Prof. Dr. Christian Baun
Frankfurt University of Applied Sciences · WS 2025/26