With the development of global economy and culture and the transformation of China’s cultural development strategy, the traditional sports culture of nationalities is undergoing a great change. In this paper, for the protection and inheritance of traditional sports skills of the She ethnic group, two Kinect sensors are firstly used to monitor the target human body simultaneously to solve the problem of inaccurate recognition of human behavior caused by occlusion. The human body model is used as the target, and the bone tracking technology is utilized to collect position information of the human body joints under the coordinate system of the two sensors respectively. Secondly, the joint point data is unified by global coordinate transformation, and the complete human posture model is derived by data fusion to extract the human feature region. Design the DLSTM-GCN action recognition network model, which extends the TCN network hierarchy on the basis of the ST-GCN algorithm and adds the residual mechanism to better capture the short-term and long-term time-dependent relationship between the actions of She traditional sports skills, and define the action evaluation formula to assess the She traditional sports skills. The experimental results show that the fusion study of human skeleton data is completed through the collection of coordinate points and the verification of simulation experiments, and the classification ability of DLSTM-GCN model on the She traditional sports skills movements is verified in the She dataset, as well as the enhancement effect of the modules on the classification accuracy, and the effect enhancement accuracy is improved from 63.55 to 98.76, which is remarkable. Finally, the traditional strategy of inheriting sports skills is proposed based on the results.