基于关键点检测的坐姿识别方法

Sitting Posture Recognition Based on Keypoints Detection

针对直接应用深度学习分类算法所得模型泛用性较差的问题,提出了一种分两个步骤完成人体坐姿识别的方法。所提方法首先提取图片中人体上身关键点的坐标信息,在关键点检测环节采用高分辨率主干网络,并进一步改进了模型结构,在下采样环节中引入SE(Squeeze-and-Excitation)注意力机制,加强了空间位置特征的表达,取得了更高的检测平均精准度;然后采用随机森林算法对关键点进行坐姿分类。实验结果表明:所提方法识别准确率可以达到94%以上,并且在陌生场景下有更好的泛用性,能适应实际应用中复杂的人物环境。

Aiming at the problem of poor generalizability of the model obtained from the direct application of deep learning classification algorithm,this paper proposes a two-step method to complete the human sitting posture recognition.In the proposed method,firstly the coordinate information of the keypoints of the upper human body in the picture is extracted,a high resolution backbone network in the keypoints detection link is adopted,and the model structure is furtherly improved.The Squeeze-and-Excitation(SE)attention mechanism is introduced in the down-sampling link to strengthen the expression of spatial positional features,achieving a higher detection average accuracy.Then the Random Forest algorithm is used to classify the keypoints for sitting posture.Experimental results show that the recognition accuracy of the proposed model can reach over 94%,and it has better generalization in unfamiliar scenarios,which better adapts to environments with complex characters in practical applications.