<![CDATA[Person detection in a crowded environment is very challenging due to excessive occlusion, scale variation, and heavy overlap between individuals. This paper proposes a lightweight Mix-Gated attention framework to enhance the feature representation capacity of YOLOv11s for dense person detection. The proposed framework incorporates two complementary channel-wise and spatial attention mechanisms under a flexible channel-wise gating strategy, while preserving the original YOLOv11s architecture and real-time inference ability. Unlike previous studies, which primarily employed either a single attention mechanism or fixed combinations of multiple attention modules, the proposed framework introduces an adaptive Mix-Gated feature selection mechanism that dynamically learns the contribution of complementary attention modules through trainable channel-wise gating. Furthermore, this study systematically evaluates ten dual and triple attention configurations under identical training and evaluation conditions, providing a comprehensive and fair comparison of their effectiveness. this study evaluated the performance on four popular benchmark datasets - CrowdHuman, WiderPerson, MOT17Det and MOT20Det - using Precision, Recall, F1-score, mAP@50 and mAP@50-95 as evaluation metrics. The experimental results show that all proposed Mix-Gated configurations outperform the baseline YOLOv11s model. Mix-Gated-CBAM+SE+SA has the best detection accuracy and efficient inference performance among Mix-Gated configurations. These results indicate that adaptive fusion of complementary attention mechanisms can improve the feature representation and person detection performance in crowded scenes while not sacrificing computational efficiency.]]>
