Fall Detection for Smart Living using YOLOv5

TL;DR

This work tackles real-time fall detection in smart homes by employing a lightweight YOLOv5mu detector trained on a compact, augmented dataset. Despite starting from only 60 annotated frames, the approach leverages Roboflow-based data augmentation and a carefully resized input to achieve a high $mAP_{50}=0.995$ and strong per-class performance, suitable for edge deployment. The results demonstrate robust detection across occupant states and offer insight into model efficiency (339 layers, 25.07M parameters, 64 GFLOPs) and training dynamics. The authors advocate future expansions including posture analysis and multisensor fusion to further improve accuracy and practical applicability in varied home environments.

Abstract

This work introduces a fall detection system using the YOLOv5mu model, which achieved a mean average precision (mAP) of 0.995, demonstrating exceptional accuracy in identifying fall events within smart home environments. Enhanced by advanced data augmentation techniques, the model demonstrates significant robustness and adaptability across various conditions. The integration of YOLOv5mu offers precise, real-time fall detection, which is crucial for improving safety and emergency response for residents. Future research will focus on refining the system by incorporating contextual data and exploring multi-sensor approaches to enhance its performance and practical applicability in diverse environments.

Figures (8)

• Figure 1: Smart Home Movement Dataset
• Figure 2: Image resized to 640x640 pixels
• Figure 3: Original vs. Grayscale image
• Figure 4: Images with hue adjustments
• Figure 5: Saturation applied on dataset