A Fall Detection Algorithm Based on EN-YOLO

YOLO is a real-time object detection algorithm. When the YOLOv5 framework is applied to fall detection, its accuracy is relatively low in complex scenarios such as lighting changes and object occlusion. To solve this problem, this paper proposes an enhanced fall detection algorithm—EN-YOLO. It introduces NAS search technology and EfficientNet-Backbone to optimize the model structure and reduce computational complexity; adopts the MBConv convolution block to replace the traditional modules of YOLOv5, combined with residual connections to mitigate information loss; integrates the SE attention mechanism to focus on key task information; and replaces FPN with BiFPN to promote cross-scale feature interaction. Test results on the public dataset Fall-Down-Det demonstrate that the system achieves an accuracy of 98.5% in detecting fall and normal states, meeting the expected requirements.