This work studies a real-time motion recognition system for table tennis based on an ESP32 sensor, which is suitable for table tennis enthusiasts to recognize the motion of table tennis. When they are playing table tennis, the system can collect the triaxial acceleration signals and transmit the signals to the computer end in real time. Then, the original acceleration signal is pre-processed, including noise reduction and window segmentation, time domain characteristics of the pre-processed acceleration signal are extracted and the feature dimension is reduced by using PCA algorithm. Finally, the decision tree algorithm is employed to build the optimal learning model for the classification and display of motion. The experimental results show that the system can recognize and display the real-time recognition and count of four kinds of table tennis motion, and the recognition accuracy reaches 97.32%.