Fall detection algorithm based on improved YOLOv8: DEW-YOLO
Chi Wang, Lou Li · 2025
Falls are one of the leading causes of death due to injury among the elderly over 65 years old. Existing fall detection algorithms face challenges such as low accuracy and high model complexity in complex environments (e.g., light and angle changes). To address these challenges, this paper proposes a fall detection algorithm based on an improved YOLOv8 model: DEW-YOLO. By introducing the Deformable Convolutional Network (DCNv2) to enhance the C2f module in YOLOv8, the flexible convolutional kernel improves the model's recognition ability in dynamic, complex scenes. An efficient multi-scale attention module (EMA) is added to the neck network to further enhance feature extraction through weighted fusion of features at different levels. At the same time, the WIoU v3 loss function is used to optimize the weighted overlapping area, effectively reducing boundary frame positioning errors. Experimental results show that, while the number of parameters and computational cost of DEW-YOLO increase by only 6.3% and 4.9%, respectively, the indices [email protected] and [email protected] reach 97.3% and 73.2%, which are 0.9% and 1.5% higher than the baseline model, respectively, significantly improving detection accuracy.