In order to solve the problem that traditional sports training methods rely heavily on the manual guidance of coaches, the innovative research on artificial intelligence-assisted sports training from the perspective of the integration of body and beauty was put forward. The sports training assistant system uses Kinect V2 as the real-time motion acquisition sensor, and uses the separation strategy to separate people from the background of sports venues, thus reducing the calculation amount of data. The complexity of motion recognition is reduced by simplifying the human body into 18 skeletal joints, and multi-target tracking algorithm is used to capture the position data of joint points. The VGG convolutional neural network is used to transform the two-dimensional joint data into human posture maps, and the human posture maps at different times are used as training samples of the stacked model, and the real-time motion recognition model of sports training is obtained by training and parameter optimization in the way of supervised learning. The experimental results show that, as can be seen from the figure, the precision curves of training data and verification data gradually recombine towards 0.86 by increasing the training times, which indicates that the scheme proposed in this paper has good stability and robustness.