Exploration and Innovation of Art Education Based on Internet VR Technology

Abstract

The application of virtual reality (VR) technology in education has become increasingly widespread, particularly in art education, where enhancing teaching effectiveness through pose recognition and motion correction holds significant importance. This study introduces MoveNet pose recognition technology to improve the accuracy and efficiency of motion recognition and correction in art education. The MoveNet model is integrated with a fully convolutional network (FCN) architecture for pose recognition, and spatial coordinate system transformation techniques are applied to process recognition results from different viewpoints. Additionally, a fully connected neural network is utilized for motion correction. Experimental results demonstrate that MoveNet exhibits notable advantages in various art education scenarios, consistently achieving the lowest mean per joint position error (MPJPE) at 9.76 across iterations, outperforming other algorithms. MoveNet's error range falls between 0.039 and 0.047, significantly lower than competitors, while maintaining a peak recognition speed of approximately 60 frames per second (FPS) across all iterations. In dance instruction, the average joint recognition error is merely 1.5 cm, and the average response time in martial arts training reaches 0.1 seconds, both earning a usability score of 5. Furthermore, the system achieves a 98% finger joint recognition rate in music conducting scenarios and an average frame rate of 55 FPS during drama performance training.