Research On Lightweight Human Fall Detection Method

YOLOv8 has achieved good results in the fall detection task, but there are still some problems with a large amount of computation. Based on this, this paper proposes a lightweight human fall detection algorithm CPDS-YOLO. Firstly, to reduce the number of parameters and the running time of the model, a new CSPPC is proposed to replace C2f based on the lightweight PC(PartialConv). This modification significantly reduces the computational cost and number of model parameters while enhancing detection speed. Secondly, an efficient Partial Self-attention module (PSA) is utilized to enhance the global representation learning ability of the algorithm and improve the model detection precision at a low computational cost. Then, an ultra-lightweight and efficient dynamic sampler (DySample) is designed, which significantly improves the detection accuracy. Finally, the lightweight Slim-Neck replaces the original feature fusion layer, which can reduce the computational cost and effectively fuse the feature information of different stages. Experimental results show that, compared with the original YOLOv8n on the Fall Detect dataset, the precision of the improved fall detection algorithm is increased by 2.8% and the computational cost (GFLOPs) is reduced by 22%, which proves the effectiveness of CPDS-YOLO.