基于深度迁移学习的人体舞蹈姿态检测方法

Human Dance Posture Detection Method Based on Deep Transfer Learning

常规的人体舞蹈姿态检测方法存在多姿态动作检测精度和识别率较低,因此提出基于深度迁移学习的人体舞蹈姿态检测方法。首先,利用Kinect传感器采集人体舞蹈姿态动作三维数据。其次,基于滑动窗口间接分割原理识别动作类别。再次,利用深度迁移学习建立动作识别模型,识别人体舞蹈的特定动作和非特定动作。最后,结合人体关节位置特征,检测人体舞蹈姿态动作中左手、右手、上身及全身4类局部特征信息。实验分析可知,新的方法应用后,舞蹈姿态检测的交并比值较高,显著提升了检测准确性。

The conventional human dance pose detection methods have low detection accuracy and recognition rate for multi pose movements.Therefore,a human dance pose detection method based on depth transfer learning is proposed.Firstly,use Kinect sensors to collect three-dimensional data of human dance postures and movements.Secondly,based on the principle of sliding window indirect segmentation,identify action categories.Thirdly,using deep transfer learning,a motion recognition model is established to identify specific and non-specific actions of human dance.Finally,combined with the position characteristics of human joints,four types of local feature information in human dance posture movements,namely the left hand,right hand,upper body,and whole body,are detected.Experimental analysis shows that after the application of the new method,the intersection to union ratio of dance pose detection is higher,significantly improving the detection accuracy.