Research And Application Development Of Original Dance Movement Optimization Based On Three-dimensional Skeleton Point Grap

In recent years,with The State Council’s "Opinions on Further Strengthening the Protection of Intangible Cultural Heritage" and other documents,the protection of intangible cultural heritage has attracted more and more attention.The original ecological dance of China’s ethnic minorities is an important part of the national intangible cultural heritage.However,the diversification of culture has affected the development and inheritance of the original ecological dance.Traditional dance inheritance is a one-way and artificial way,which makes it impossible for dance trainers to get effective and timely feedback.With the rise of digitalization,it has become a trend to use digital means to inherit the original ecological dance.Compared with the early single inheritance method,digitization has the advantage of strong interest,but it has the disadvantage of no professional teacher’s guidance,which makes the standard of the action difficult to ensure.For the trainer,in the process of practicing dance,the standard of the movement not only determines the final score,longterm practice of non-standard movement is likely to cause muscle strain,sprain and so on.Therefore,in the digital practice of the original ecological dance,the movement optimization of the dance trainer is extremely important.In this paper,three ways of action evaluation,action recognition and action evaluation are used to optimize the original dance movements of the trainers,and the following work is carried out:(1)Research on primitive dance movements of ethnic minorities and construction of data sets.This paper analyzes eight common primitive dances and their typical movement characteristics.Eighteen typical movements of six kinds of primitive dances are screened and their skeletal information is collected to construct a typical movement data set of primitive dances.Aiming at the self-occlusion problem of skeleton points in the acquisition process,the existing self-occlusion motion information inpainting algorithm was improved to recover the position information of the occluded skeleton points.According to the different degree of participation of each bone point in different movements,the skeleton points that play a key role in the movement are dynamically weighted to improve the participation of key limbs and enhance the reliability in the process of data collection.(2)The improvement of action optimization research algorithm.In this paper,Graph Convolutional Neural Networks(ST-GCN),improved three-dimensional Convolutional neural networks(3D CNNS),improved Dynamic Programming(DTW)and other algorithms are used to optimize the original ecological dance movements of the trainers,and the performance of the improved algorithm is verified by experiments.Then,the improved algorithm is verified twice in the Kinect public data set.Experimental results prove the universality of the improved algorithm,which provides technical support for subsequent application construction.(3)The construction of original ecological dance auxiliary training application.In this paper,the above theory and methods are integrated.Aiming at the problems that the traditional primitive dance training method is not interesting and the trainers cannot ensure the standard of their own movements,an auxiliary training application of primitive dance movement optimization based on 3D bone point information is designed and developed by using Unity3 D software.The trainer practices the dance movement in front of Kinect,and the skeleton point information of the trainer is collected.The action data of the trainer’s skeleton point is assigned to the corresponding skeleton point of the character model in Unity3 D through virtual reality technology,and the model is driven to reproduce the trainer’s dance movement,and the movement is evaluated,recognized and evaluated according to the needs of the trainer.The action evaluation module recommended the original dance movement optimization scheme for the trainer,and the action recognition module helped the trainer to find suitable exercises.The action evaluation module enables the trainer to understand the difference between his own action and the standard action,so as to optimize the trainer’s action.