Construction of VR Video Quality Evaluation Model Based on 3D-CNN

Currently, virtual reality (VR) panoramic video content occupies a very important position in the content of virtual reality platforms. The level of video quality directly affects the experience of platform users, and there is increasing research on methods for evaluating VR video quality. Therefore, this study establishes a subjective evaluation library for VR video data and uses viewport slicing method to segment VR videos, expanding the sample size. Finally, a classification prediction network structure was constructed using a three-dimensional convolutional neural network (3D-CNN) to achieve objective evaluation of VR videos. However, during the research process, it was found that the increase in its convolutional dimension inevitably leads to a significant increase in the parameter count of the entire neural network, resulting in a surge in algorithm time complexity. In response to this defect, research and design dual 3D convolutional layers and improve 3D-CNN based on residual networks. Based on this research, a virtual reality video quality evaluation model based on improved 3D-CNN was constructed. Through experimental analysis, it can be concluded that the average overall accuracy value of the constructed model is 95.27%, the average accuracy value is 95.94%, and the average Kappa coefficient value is 96.18%. Being able to accurately and effectively evaluate the quality of virtual reality videos and promote the development of the virtual reality field.

@article{Zhao2023,
  title     = {Construction of VR Video Quality Evaluation Model Based on 3D-CNN},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {14},
  number    = {8},
  year      = {2023},
  publisher = {The Science and Information Organization},
  author    = {Hongxia Zhao and Li Huang},
  doi       = {10.14569/IJACSA.2023.0140892},
  url       = {https://doi.org/10.14569/IJACSA.2023.0140892}
}