Abstract: To meet the needs of experimental teaching in multi-sensor measurement and action recognition, a human motion capture experimental system based on distributed inertial sensors is developed. The system consists of a distributed inertial sensing subsystem and a human motion analysis platform, covering the complete experimental workflow from hardware deployment and configuration, data acquisition, attitude estimation, and 3D scene construction and visualization to action recognition model training and evaluation. To verify the system’s effectiveness, typical industrial operations are used to construct a dataset containing seven action classes, including hammering, sweeping, painting, and welding. Sliding windows are applied separately within the training, validation, and test sets to generate model input samples. The experimental results show that the system supports multi-node inertial data acquisition, 3D skeleton reconstruction, action recognition model training, and online recognition. The action recognition model achieves an average accuracy of 92.29% on the independent test set. Using this system, experimental projects in data acquisition, attitude analysis, and action recognition are designed to support parameter comparisons, result interpretation, and system integration, providing a configurable environment for comprehensive experiments in related disciplines.