🚨 Human Fall Detection System

YOLOv8-based static image fall detection system with support for real-time video stream and offline video file processing.

📋 Table of Contents

• 🚨 Human Fall Detection System
  • 📋 Table of Contents
  • 🎯 Introduction
  • 📊 Model Performance
    • Class-wise Performance
      • 🟢 Normal State
      • 🔴 Fall State
      • ✅ Performance Highlights
  • 💾 Installation
    • Method 1: pip Installation (Recommended)
    • Method 2: Install from Source
  • 📖 Usage Examples
    • Jupyter Notebook Example
  • 📈 Evaluation Results
    • Performance Metrics Charts
    • Detection Examples
    • Video Detection Demo
  • 🙏 Acknowledgments
  • 📮 Contact

🎯 Introduction

This project is a deep learning-based human fall detection system that achieves high-precision fall detection using a fine-tuned YOLOv8 model. The system can be applied to elderly care, hospital monitoring, smart home scenarios, and other contexts to promptly detect fall events and trigger alerts.

📊 Model Performance

Evaluation results based on test dataset:

Metric	Value	Description
mAP@0.5	90.61%	Primary performance indicator
mAP@0.5:0.95	39.63%	Stricter evaluation standard
F1-Score	88.60%	Harmonic mean of precision/recall
Overall Precision	90.13%	Accuracy of all detections
Overall Recall	87.11%	Proportion of detected targets

Class-wise Performance

🟢 Normal State

• AP@0.5: 86.22%
• Precision: 90.75%
• Recall: 80.36%

🔴 Fall State

• AP@0.5: 95.00%
• Precision: 89.51%
• Recall: 93.86%

✅ Performance Highlights

• Excellent fall detection sensitivity (93.86% recall)
• Low false alarm rate (<10% false positives)
• Outstanding overall detection performance (mAP > 90%)

💾 Installation

Method 1: pip Installation (Recommended)

pip install human-fall-detection

Method 2: Install from Source

git clone https://github.com/TomokotoKiyoshi/Human-Fall-Detection.git
cd Human-Fall-Detection
uv sync

📖 Usage Examples

Jupyter Notebook Example

See example_usage.ipynb for interactive examples.

📈 Evaluation Results

Performance Metrics Charts

Confusion Matrix	PR Curve	F1 Curve
Precision Curve	Recall Curve

Detection Examples

Actual detection performance on the test dataset:

Ground Truth Labels	Model Predictions

Video Detection Demo

System's real-time video processing demonstration:

The video demonstrates the system's real-time detection capabilities:

• 🟩 Green box: Normal state
• 🟥 Red box: Fall detected
• Real-time confidence scores displayed

🙏 Acknowledgments

• Ultralytics YOLOv8 - Excellent object detection framework
• UAH Fall Detection Dataset - High-quality fall detection dataset

📮 Contact

• Project Homepage: https://github.com/TomokotoKiyoshi/Human-Fall-Detection
• Issues: Issues

---