基于YOLOv11n-Pose的人体姿态检测方法OA
Human Pose Detection Method Based on YOLOv11-Pose
为了应对复杂场景下人体姿态估计中关键点丢失以及多尺度目标识别难题,提出了一种改进的YOLOv11n-Pose模型架构.通过采用部分卷积(PConv)优化C3k2 模块中的Bottleneck,显著增强了网络对多尺度特征的提取与融合能力.此外,采用 RepViT Block模块替代传统卷积结构,降低了参数量,同时提高了模型对复杂姿态的识别精度.引入CPAM注意力机制,有效强化对人体关键部位的特征捕获与语义分析能力.结果表明,改进的YOLOv11n-Pose模型在检测精度、召回率和平均精度上均优于原始YOLOv11n-Pose模型,对不同尺度目标识别效果显著提升;同时,参数量和模型大小分别缩减10.1%和 8.2%.
To address the challenges of key point loss and multi-scale object recognition in human pose estimation in complex scenarios,an improved YOLOv11n-Pose model architecture is proposed.By using Partial Convolution(PConv)to optimize Bottleneck in the C3k2 module,the network's ability to extract and fuse multi-scale features has been significantly enhanced.In addition,using RepViT Block module instead of traditional convolutional structure reduces the number of parameters while improving the recognition accuracy of the model for complex poses.Finally,the introduction of CPAM attention mechanism effectively enhances the ability to capture features and analyze semantics of key parts of the human body.The experiment shows that the improved model outperforms the original YOLOv11n-Pose model in detection accuracy,recall rate,and average accuracy,and significantly improves the recognition performance of targets at different scales.Meanwhile,the number of parameters and model size decreased by 10.1%and 8.2%,respectively.
刘玉成;王甜甜;刘美;朱一鑫
吉林化工大学 信息与控制工程学院,吉林 吉林 132022||广东石油化工学院 自动化学院,广东 茂名 525000吉林化工大学 信息与控制工程学院,吉林 吉林 132022||广东石油化工学院 自动化学院,广东 茂名 525000广东石油化工学院 自动化学院,广东 茂名 525000广东石油化工学院 自动化学院,广东 茂名 525000
信息技术与安全科学
YOLOv11n-Pose姿态检测关键点检测部分卷积
YOLOv11-Posehuman posture detectionkey point detectionpartial convolution
《广东石油化工学院学报》 2026 (1)
68-73,6
评论