Research and Implementation of YOLOv8+Transformer in Human Fall Detection Image Recognition

Full abstract

Read the full abstract

Human fall detection is a crucial research direction in the fields of intelligent security and elderly care monitoring. Traditional detection methods suffer from issues such as low accuracy, poor real-time performance, and privacy leakage. To address these problems, this paper proposes a human fall image recognition method that integrates YOLOv8 and Transformer. First, to tackle the limited sample size in public fall datasets, various data augmentation strategies like image rotation and flipping are employed to expand the dataset and enhance the model's generalization capability. Second, the Swin Transformer module is introduced into the YOLOv8 backbone network, leveraging its sliding window self-attention mechanism to improve the model's ability to capture global contextual information. Finally, multi-scale feature fusion is achieved through a feature pyramid network, thereby enhancing the detection accuracy for fall targets of varying scales.