This paper addresses the issue of preserving and digitizing ethnic dance movements by analyzing the characteristics of ethnic dance movements, Kinect-based motion capture technology, and skeleton-based motion recognition methods. It proposes a deep learning-based method for recognizing typical ethnic dance movements. Using Kinect sensors to collect data on typical ethnic dance movements, a dataset of typical ethnic dance movements was constructed, and 3D CNNs were used for recognition. Finally, strategies for the protection, inheritance, and digital development of ethnic dances are proposed, and effective pathways for the digital preservation of minority ethnic dances are explored. The results indicate that both the detection of two-dimensional joints and the extraction of three-dimensional joint information can to a certain extent meet the requirements for three-dimensional human motion reconstruction. Additionally, the results of the motion capture system setup and three-dimensional human motion reconstruction are also satisfactory, with experimental errors around 3%. Compared to traditional methods, the motion capture joints and angles under the proposed method are closer to the Kinect standard values, and the motion capture trajectories have the smallest error compared to the Kinect method. Furthermore, the recognition accuracy rate of the proposed method remains above 95%, with a maximum accuracy rate of 99.76%, demonstrating that the proposed method has certain feasibility and application prospects in the preservation and inheritance of ethnic dances.