As China’s population continues to age, the issue of falls among the elderly has increasingly become a focal point of public health concerns. Traditional fall detection methods often underperform in complex scenarios, whereas deep learning-based object detection algorithms have shown greater poten-tial. This paper proposes a novel deep learning model, ylanNet, specifically designed for fall detection tasks, which combines the strengths of YOLOv8 and GELAN. By integrating CSPNet and ELAN architectures, ylanNet en-hances feature extraction capabilities and computational efficiency, making it adaptable to tasks of varying scales. In experiments, ylanNet outperforms YOLOv5, YOLOv8, YOLOv10, and RT-DETR on fall detection datasets, particularly excelling in terms of precision (0.833), mean average precision (mAP, 0.817), and mean average precision at different IoU thresholds (mAP50–90, 0.540). The results demonstrate that ylanNet achieves superior detection performance while maintaining efficiency, making it suitable for real-time fall detection tasks. Future work will focus on further improving the generalization ability of ylanNet and exploring its potential applications in other fields.