An efficient algorithm for pedestrian fall detection in various image degradation scenarios based on YOLOv8n

Area of Science:

• Computer Vision
• Artificial Intelligence
• Biomedical Engineering

Background:

• The increasing elderly population necessitates robust pedestrian fall detection systems.
• Current systems struggle with accuracy in adverse conditions like rain, snow, or low light.
• Urban environments present unique challenges for reliable fall detection.

Purpose of the Study:

• To develop an enhanced pedestrian fall detection algorithm, PCE-YOLO, for improved accuracy and performance.
• To address limitations of existing systems in complex and degraded environmental conditions.
• To provide a real-time solution for elderly fall detection.

Main Methods:

• Proposed PCE-YOLO algorithm based on YOLOv8n, incorporating a Chain-of-Thought Prompted Adaptive Enhancer (CPA-Enhancer) module.
• Optimized Cross Stage Partial Bottleneck with 2 Convolution Block (C2f) for reduced computational load.
• Implemented Inner Extended Intersection over Union (Inner-EIoU) loss for enhanced bounding box regression.
• Validated using a dataset of 7,782 pedestrian fall images and three degraded datasets.

Main Results:

• PCE-YOLO achieved a 4.52% improvement in mean Average Precision (mAP) over YOLOv8n on original and degraded datasets.
• The algorithm demonstrated a high processing speed with 210.5 frames per second (FPS).
• Significant enhancement in detection accuracy and speed across various challenging environments.

Conclusions:

• PCE-YOLO offers a robust and accurate solution for real-time pedestrian fall detection, especially for the elderly.
• The algorithm's performance in degraded conditions makes it suitable for real-world applications.
• PCE-YOLO represents a significant advancement in computer vision for public safety and healthcare.