VIRTUAL REALITY SCENE RENDERING AND INTERACTION TECHNOLOGY BASED ON MACHINE LEARNING AND SEMI-SUPERVISED LEARNING

Xiguang Zhang, Yuxi Ma, and Longfei Zhang

Keywords

Semi-supervised learning, machine learning, virtual-reality–reality interaction, self-attention mechanism, scene rendering

Abstract

In the development of virtual reality application, the combination of scene rendering and interaction technology promotes the development of mixed reality field. The fineness of virtual reality scene and the naturalness of interaction directly affect the user experience. However, there are challenges in the efficiency and effect of traditional methods in scene rendering and interaction. To solve this problem, this study proposes the research of virtual reality scene rendering and interaction technology based on machine learning and semi-supervised learning. Firstly, a semi-supervised learning model is designed based on convolutional neural network (CNN) to interact with the virtual reality scene shock to predict and optimise the rendering effect. Secondly, in order to further improve the robustness of virtual-real interaction technology, CNN is used as the backbone of the network, and self-attention mechanism is introduced to further improve the robustness of the model and the ability to capture complex features. Finally, the model is improved by using a bidirectional long short-term memory network to better process sequence data and optimise the accuracy of virtual-real interaction. Experimental results show that the framework achieves an average accuracy of 86.6% on the dataset, which is 3.4% higher than other algorithms mentioned in the study. This shows that the framework has greater accuracy in recognising and processing mixed reality data, and is able to more accurately understand the user’s intentions and actions. An average peak signal-to-noise ratio of 32.2 means that the frame’s output images perform well in terms of detail and clarity. At the same time, the average similarity reached 96.3%, indicating that the generated image has a high similarity to the real scene, further enhancing the user’s sense of immersion and sense of reality. By introducing Transformer and BiLSTM, the framework effectively reduces background errors in virtual-real interactions. This helps reduce user confusion and discomfort in mixed reality environments and enhances the overall user experience. The semi- supervised learning framework has significant advantages in the field of mixed reality, which can effectively improve the output image quality and reduce the interaction error, and provides a strong support for the development of virtual reality technology.

Important Links:



Go Back