关键词: Kinect, 相机标定, 深度图像, 卷积神经网络, 人机交互

Abstract: In order to better output various pose risk assessment methods quickly and accurately, this paper proposes to use Kinect v2 and convolutional neural network to identify the joint angles of the human body and output the pose risk assessment score. First, two cameras are calibrated by using the checkerboard calibration algorithm of sub-pixel corner extraction. Second, the improved bilateral filtering is used to denoise the depth image. At the same time, the convolutional neural network is used to identify the two-dimensional position of the human joint, and the depth information is used to obtain the actual three-dimensional coordinates. The joint angle is obtained, and finally the posture risk assessment score is outputted. The accuracy of Kinect recognition angle and posture evaluation scores is verified by two experiments, which indicates that the identification and evaluation accuracy of the proposed method is high and there is a certain degree of generalization.

Key words: Kinect, camera calibration, depth image, convolutional neural network, human-machine interaction