In order to solve the problem of traditional tennis serving technique teaching, subjective, and experience-based teaching methods, the author proposes a method to extract tennis training movements based on artificial intelligence sensor video analysis. This method requires the use of ordinary cameras and computer technology; video analysis technology is used to guide tennis teaching and training. The result obtained is as follows: in the experimental group and the control group, at the end of the teaching stage, the movement characteristics of each link of the body are quite different, which can reach 15%-20%; when the experimental group and the control group did the “scratching back” movement during the teaching period, at the end of teaching, the students in the experimental group were close to high-level tennis players in their movement skills, and the movement gap was only about 5%. It is proved that the method proposed by the author provides an objective and scientific basis for measuring the referee’s penalty level. In the grass-roots tennis teaching and training, it has important value and significance and has a great promotion prospect.