Falling Detection of Toddlers Based on Improved YOLOv8 Models

Area of Science:

• Computer Vision
• Artificial Intelligence
• Child Safety Technology

Background:

• Toddlers falling from heights pose significant health risks.
• Existing surveillance systems lack specialized fall detection for toddlers.

Purpose of the Study:

• To develop and evaluate a real-time toddler falling alarm system.
• To improve early detection and rescue of toddlers after falls.

Main Methods:

• A two-stage approach using GELAN-integrated YOLOv8 for body and head feature extraction.
• Generalized Hough Transform (GHT) to calculate the 'safe zone' and analyze spatial location.
• Comparison of spatial data with body (K) and head (H) centers of mass for position status detection.

Main Results:

• The proposed framework demonstrated superior precision, recall, mean average precision, and F1 scores compared to YOLOv3, v5, v6, and v8.
• Achieved an accuracy of 96.33% in detecting toddler falls.
• Met the Frames Per Second (FPS) requirement for real-time surveillance.

Conclusions:

• The developed toddler target extraction and falling alarm system is highly accurate and efficient.
• This technology offers a significant advancement in proactive child safety monitoring.
• The system provides a reliable solution for real-time fall detection in domestic environments.