Traditional folk dance represents a significant form of intangible cultural legacy in China, The use of modern technology to identify and capture folk dance movements facilitates their understanding and inheritance. This paper collects folk dance motion data via Kinect to build a typical dance movement dataset, and adopts an attention-enhanced spatio-temporal graph convolutional network to implicitly learn skeletal sequence features for dance motion recognition. Furthermore, it proposes a sensor-based auxiliary training method. By constructing a 3D human skeleton joint model, the system reconstructs learners’ movements to assess positional accuracy, thereby enabling assisted dance training and supporting heritage preservation.The test results show that compared with other motion recognition systems, this system has the smallest error between the motion recognition angle and joint positioning and the Kinect standard value, and the recognition accuracy rate is as high as 99.1%. A thorough examination validates that the suggested sensor – based supplementary training approach allows for the precise acquisition and automated identification of folk dance motions, which attests to its efficacy.
