Abstract: Traditional video super-resolution methods were found to be ineffective in addressing the geometric distortion problems caused by equirectangular projection when processing panoramic videos. Deficiencies were also observed in inter-frame alignment and feature fusion, which resulted in poor reconstruction quality. To further improve the super-resolution reconstruction quality of panoramic videos, a panoramic video super-resolution network combining spherical alignment and adaptive geometric correction, named 360GeoVSR, was proposed. Precise alignment and efficient fusion of inter-frame features were achieved through a Spherical Alignment Module (SAM) and a Geometric Fusion Block (GFB). In SAM, spatial transformation and deformable convolution were combined to address global and local geometric distortions. In GFB, feature alignment was dynamically corrected using an embedded Adaptive Geometric Correction submodule (AGC), and multi-frame information was fused to capture complex inter-frame relationships. The results of subjective and objective comparison experiments on the extended ODV360Extended panoramic video dataset showed that 360GeoVSR outperformed five representative super-resolution methods, including BasicVSR++ and VRT (Video Restoration Transformer), in both objective metrics and subjective visual effects, validating its effectiveness. 

Key words: panoramic video, super-resolution, geometric correction, spherical alignment, deep learning

摘要： 现有的传统视频超分辨率方法在处理全景视频时，难以有效解决等距矩形投影带来的几何畸变问题，且在帧间对齐和特征融合方面存在不足，导致重建效果不佳。为进一步提升全景视频的超分辨率重建质量，提出了一种联合球面对齐与自适应几何校正的全景视频超分辨率网络——360GeoVSR。该网络通过球面对齐模块(SAM)和几何融合块(GFB)实现帧间特征的精确对齐与高效融合。SAM结合空间变换和可变形卷积，处理全局与局部几何畸变；GFB通过嵌入的自适应几何校正子模块(AGC)动态校正特征对齐，并融合多帧信息以捕捉复杂的帧间关系。在扩展的ODV360Extended全景视频数据集上的主客观对比实验结果表明，360GeoVSR在客观指标和主观视觉效果上均优于BasicVSR++、VRT（Video Restoration Transformer）等5种代表性超分辨率方法，验证了它的有效性。

关键词: 全景视频, 超分辨率, 几何校正, 球面对齐, 深度学习