基于轻量级网络的实时人体关键点检测算法

为提升人体姿态估计在移动终端设备上的运行速度与实时性，提出一种改进的人体关键点检测算法。通过将MobileNetV2轻量级主干网络与深度可分离卷积模块相结合加速特征提取过程，使用精炼网络进行多尺度人体关键点预测，并利用融合网络整合多个尺度的预测结果得到最终人体关键点检测结果。实验结果表明，与传统CPM算法相比，该算法在网络模型参数量和浮点运算量明显减少的情况下PCKh@0.5仅下降0.1个百分点，具有较高的检测精度和较好的实时性。

Real-Time Human Keypoint Detection Algorithm Based on Lightweight Network

To improve the real-time performance and speed of human pose estimation on mobile devices,this paper proposes an improved human keypoint detection algorithm.The algorithm uses the depthwise separable convolution module and MobileNetV2,a light-weight backbone network,to accelerate feature extraction.Then the refine network is used to predict the keypoint of human body on multiple scales.The results at different scales are fused by using fusion network to obtain the final keypoint detection result.Experimental results show that compared with the traditional CPM algorithm,the proposed algorithm significantly reduces the amount of parameters and floating point operations while the PCKh@0.5 is decreased only by 0.1 percentage point.It has higher detection accuracy and better real-time performance.