With the development of society, it brings more invisible pressure on the physiology and psychology of the elderly group, and dance recreation, as a new type of psychological relief and detoxification, has gained the attention of people from all walks of life. In this paper, Kinect sensor technology is utilized to collect and track the depth data information of human skeleton joints and display it in three-dimensional space. A graph convolutional neural network is introduced to extract the human skeletal sequence related to action features, preprocess the skeleton information, and repair the data related to the occluded skeletal points and action information. 3D CNNs dance movement recognition algorithm is proposed on the basis of 2D convolutional neural network, which facilitates the capture of movement information in multiple consecutive frames. The 3D convolutional network and Kinect sensor are utilized to mine exploitable recreational dance materials and embedded in the elderly care model to explore the effectiveness of dance recreation on the health of the elderly through the combination of experimental and practical approaches. Using the method of this paper, the completeness of the extraction of the dance movements is above 95%, which shows that the dance recreation movements are less affected by the complexity of the background when they are tracked. Through the dance recreation intervention after the elderly in the back direction, d = -0.74, right front direction, d = -0.68 movement efficiency used to reduce the trend of time is greater. It shows a significant effect effect on the physical health of the elderly, which further illustrates the role of providing new evidence on the physical health of the elderly under the dance recreation intervention.
