Análise baseada em IoU de viewports em vídeos imersivos: implicações do áudio para a modelagem de saliência visual

The development of visual attention models for immersive environments is essential for virtual reality-based systems. In this work, we analyze the convergence of viewports in 360º videos using the Intersection over Union (IoU) metric to quantify the overlap between regions viewed by different observers. Using eyetracking data from the PAVS10K database (67 videos), we evaluate the effect of audio on visual attention. The results indicate that audio increases viewport overlap by approximately 18%. To the best of our knowledge, this is the first study to employ IoU for this analysis in omnidirectional videos. These findings reinforce the relevance of integrating audio information into visual saliency modeling for immersive videos.