DMB-YOLOv11: a novel fall detection algorithm based on YOLOv11 for complex scenarios

Abstract

A novel fall detection algorithm is proposed based on YOLOv11 to address the issues of occlusion sensitivity, small target missed detection, and limited deployment costs in fall detection in complex scenarios. The algorithm introduces a dynamic multi-scale (DM) attention mechanism that integrates dynamic channel attention and multi-scale spatial attention after the SPPF module in the backbone network. And a cross dimensional position perception module is added to enhance the modeling ability of the model for occluded targets and human pose information. At the same time, a dynamic large convolution kernel selection strategy is adopted to optimize multi-scale feature extraction in scenarios. A bidirectional feature pyramid network (BiFPN) is introduced in the feature fusion stage to enhance the interaction between deep and shallow features to improve the representation ability of multi-scale fall postures. Experimental results show that compared with the baseline model YOLOv11n, the proposed algorithm improves precision by 2.2%, recall by 2.9%, mAP@0.5 by 5.1%, and mAP@0.5:0.95 by 3.2%. The proposed method shows better detection performance while maintaining lightweight characteristics.