Modeling Users’ Behavior in 360-Degree Videos: A Framework for Synthetic Data Generation
Walid Abdallaoui, Ahmed Saadallah, Sidi‐Mohammed Senouci, Inès El-Korbi, Philippe Brunet · 2024
Over the past decade, there has been a surge in the popularity of 360° videos streaming. This immersive technology provides users with captivating experiences, driven in part by the growing accessibility of Head-Mounted Displays (HMDs). However, alongside their widespread adoption, 360° videos present several challenges that necessitate optimization models to ensure smooth streaming and a seamless viewing experience. Meeting these performance requires a significant volume of data for training and testing optimization models. This data must encompass various user behaviors to effectively address the complexities of 360° video streaming. Current datasets suffer from limitations in both the number of users and the duration of the videos. The traditional approach, which involves collecting sensor data from 360° videos viewing experiences, is not only time-consuming but poses privacy issues for users. This paper introduces a novel approach to address this data scarcity: the generation of synthetic 360° video viewing data. We aim to overcome the limitations of lack of user viewing data. We have modeled user’s fixation point as an autonomous agent in an environment represented with saliency map. This approach has the potential to revolutionize the development of 360° videos solutions, enabling researchers to create and test solutions with a wider range of user behaviors and video content. This approach can generate a significant amount of data for any 360° video, even without existing viewing data. The source code for the approach and demonstration videos are made available for the research community for tailored dataset generation.