Research on Pedestrian Fall Detection Based on YOLOv8n

To address the challenges of finding an ideal balance between computational redundancy, accuracy, parameter count, and computational complexity in existing pedestrian fall detection algorithms, this paper proposes a lightweight pedestrian fall detection algorithm based on YOLOv8n. Firstly, a DWR-DRB module is designed to enhance the capabilities of large-kernel convolutional layers, extending the receptive field from the higher layers of the network to extract features, thereby reducing computational load and improving model performance. Secondly, a Deformable Attention mechanism is introduced to adaptively adjust the attention mechanism to focus on critical regions of the image, thereby enhancing the model's ability to recognize and localize targets. Finally., a CA-HSFPN module is constructed to improve the expressiveness of feature maps, Substantially decreasing the number of parameters and the computational complexity of the model, achieving a lightweight model. Experiments conducted on the constructed dataset show that the improved algorithm achieves a 0.7% increase in [email protected] compared to the original model, with a significant reduction in parameter count and computational load, with parameters decreasing by 33.2% and computational load decreasing by 16%, while maintaining high detection accuracy.