DGC-GatedFusion: An Improved YOLOv11 Model for Low-Light RGB-D Pedestrian Detection

Pedestrian detection methods based on single RGB images often suffer from performance degradation under lowlight conditions due to insufficient illumination and image noise, making them less effective in complex lighting environments. To address this issue, this paper proposes an improved RGB-D detection model based on YOLOv11, in which a DGC-GatedFusion module is introduced to replace the first convolutional layer of the backbone. This module enables independent modeling and gated fusion of RGB and depth information, enhancing multimodal feature representation through depth-guided convolution and channel-wise weighting. Experimental results demonstrate that the proposed method achieves improvements of 22% in Precision and 12.6% in Recall, as well as increases of 19.5% and 6.91% in mAP50 and mAP50-95, respectively, compared to the original YOLOv11 model using only RGB input. Meanwhile, the model size is reduced by 0.47%, significantly lowering model complexity and computational cost while maintaining high detection accuracy. These advantages make the proposed method well suited for real-time and robust applications such as intelligent surveillance under low-light conditions.