A Comparative Analysis of Viewing Prediction Techniques for 360° Video Streaming Applications

In this work, we implement multiple techniques for predicting users viewing directions while watching 360° videos. We utilize historical viewing traces to forecast future directions based on a real-life head tracking dataset. We compare the performance of linear regression (LR), artificial neural networks (ANN), long short-term memory (LSTM), and convolutional neural networks (CNN). We assess their efficiency in terms of viewing angles prediction errors. We also investigate tile viewing prediction in tile-based 360° video transmission scenarios. We built two classifiers based on ANN and LSTM to predict watched tiles and provide an evaluation of their performance in this article.