Fall Detection for Industrial Setups Using YOLOv8 Variants
Gracile Astlin Pereira
TL;DR
This work targets industrial fall detection by evaluating YOLOv8 variants with a dedicated augmentation pipeline to increase dataset variance. The study finds that YOLOv8m delivers a strong balance between detection accuracy and computational efficiency, achieving $mAP_{50}=0.971$ while maintaining a moderate model size, whereas larger variants like YOLOv8l and YOLOv8x achieve higher precision/recall but impose heavier resource requirements. The data-augmentation strategy (Random Resize, Grayscale, Blur, Median Blur, CLAHE) helps mitigate dataset limitations from a low-frame-rate industrial video, enabling robust performance across two classes: 'Fall Detected' and 'Human in Motion'. Overall, the results support deploying YOLOv8m in resource-constrained industrial environments for reliable real-time fall detection, with YOLOv8l/x as potential options when hardware permits higher accuracy. This work advances practical, real-time safety monitoring by balancing model complexity with detection reliability in challenging industrial settings.
Abstract
This paper presents the development of an industrial fall detection system utilizing YOLOv8 variants, enhanced by our proposed augmentation pipeline to increase dataset variance and improve detection accuracy. Among the models evaluated, the YOLOv8m model, consisting of 25.9 million parameters and 79.1 GFLOPs, demonstrated a respectable balance between computational efficiency and detection performance, achieving a mean Average Precision (mAP) of 0.971 at 50% Intersection over Union (IoU) across both "Fall Detected" and "Human in Motion" categories. Although the YOLOv8l and YOLOv8x models presented higher precision and recall, particularly in fall detection, their higher computational demands and model size make them less suitable for resource-constrained environments.