In this study, we introduced an augmented reality and deep-learning-based system for the digital preservation and interactive learning of traditional embroidery skills. Recognizing the vulnerability of traditions in preserving intricate intangible cultural heritage, this system integrates multisource data from Kinect, Leap Motion, inertial measurement units, and force-sensitive resistor sensors to capture precise embroidery actions. An extended Kalman filter is employed for robust multisensor data fusion and accurate motion trajectory estimation. The system was developed on the basis of a combined graph convolutional network–long short-term memory model, which showed a 95% accuracy in recognizing diverse needlework techniques by effectively capturing both spatial and temporal features of embroidery movements. This real-time recognition ability in an immersive augmented reality interface provides learners with dynamic visual guidance, step-by-step instructions, and performance feedback. The system overcomes the limitations of traditional preservation methods and offers a scalable, interactive, and effective platform for the transmission and documentation of traditional craftsmanship.