🚁 PrediFly - YOLO-Kalman-GRU Pipeline for Drone Detection & Prediction

📝 Description

PrediFly is an advanced drone detection and path prediction system that combines YOLO object detection, Extended Kalman Filtering for state estimation, and GRU-CNN hybrid neural networks for trajectory prediction. The system can detect drones in real-time video, track their movement with high precision, and predict their future path.

✨ Features

🔍 Real-time drone detection in video feeds
🔄 Smooth trajectory tracking with Extended Kalman Filtering
🧠 Path prediction using hybrid GRU-CNN ensemble models
📊 Comprehensive error analysis for different drone motion patterns
📱 Support for multiple drone types and movement patterns
📦 Easy integration with existing surveillance systems

🛠️ Technology Stack

Category	Technologies
Programming
Computer Vision
Deep Learning
Data Visualization
Data Processing

🧮 Algorithms & Models

1️⃣ Detection & Tracking

Algorithm	Purpose	Implementation
YOLOv8	Real-time object detection	Detects drones in video frames with high precision
Extended Kalman Filter	State estimation & smoothing	Tracks drone position, velocity, and acceleration
Non-Maximum Suppression	Multiple detection handling	Removes duplicate bounding boxes

2️⃣ Path Prediction

Algorithm	Purpose	Implementation
Bidirectional GRU	Temporal sequence modeling	Processes drone trajectory history
1D CNN	Feature extraction	Extracts motion patterns from position data
Attention Mechanism	Focus on relevant timesteps	Improves prediction accuracy for complex maneuvers

📋 Project Pipeline

graph TD
    A[Drone Video Input] --> B[Preprocessing]
    B --> C[YOLOv8 Detection]
    C --> D[Extended Kalman Filter]
    D --> E[State Estimation]
    D --> F[Hybrid GRU-CNN Model]
    F --> G[Path Prediction]
    G --> H[Performance Evaluation]

📊 Results

Performance on Different Motion Types

Motion Type	Mean Distance Error	Angular Error	RMSE	Confidence
Frog Jump Motion	2123.92 px	69.00°	2127.05 px	0.733
Downward Motion	2252.24 px	101.53°	2255.59 px	0.838
Upward Motion	2290.33 px	88.74°	2291.48 px	0.868

Detection Performance

Metric	Value
Average mAP (IoU=0.50:0.95)	0.891
Precision	0.924
Recall	0.887
Inference Time	22ms per frame

🚀 Getting Started

Prerequisites

Python 3.8+
CUDA-compatible GPU (for real-time performance)
Required packages: see requirements.txt

Installation

# Clone the repository
git clone https://github.com/yourusername/predifly.git
cd predifly

# Create and activate virtual environment
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

# Download pre-trained models
python download_models.py

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

👥 Contributors

Mayank Mittal (@mayank-mittal)

🙏 Acknowledgements

Prof. Hari Kumar Kandanth for project guidance
Ultralytics for YOLOv8 implementation
Drone Dataset Consortium for providing training data

mayankmittal29/PrediFly-YOLO-Kalman-GRU-Pipeline-for-Drone-Detection-Prediction