Improved Fall-down Detection Algorithm: FE-YOLO
Yuchen Zhang, Hongyan Ma, Zongyuan Xie · 2024
To address the concerns regarding dataset completeness and detection speed for common fall injuries in the elderly, a comprehensive dataset has been established. Additionally, a lightweight fall detection model named FE-YOLO, which is based on YOLOv8, has been developed. This model replaces the network’s internal feature extraction module with a more efficient Fasternet Block module and introduces an enhanced detection head that utilizes data sharing and grouped convolution techniques. When compared to the YOLOv8n model, the FE-YOLO model demonstrates a reduction in parameters by 43.1%, a decrease in GFLOPs by 51.9%, an improvement in model inference speed by 9.70%, and an increase in FPS by 10.5%. Notably, these enhancements do not significantly impact the average accuracy of the model. The efficiency of the FE-YOLO model renders it suitable for deployment on devices with limited computing power.