Real-Time Monitoring of Video Quality in a DASH-based Digital Video Broadcasting using Deep Learning

Digital video processing and transmission can introduce numerous distortions while capturing signals from broadcasting stations. These distortions become a nightmare for multimedia companies, especially terrestrial broadcasting companies that have fully adopted the online video streaming service. While terrestrial broadcasting benefits from online streaming through over-the-top (OTT) channels, there is a potential setback to reducing the video quality due to preprocessing of signals. Video quality assessment (VQA) algorithms have been developed for analyzing the quality of videos in a database, but little attention has been paid to implementing such algorithms in a real-time situation. This paper develops a novel real-time VQA framework by integrating a deep learning technology into the broadcasting pipeline. Previous studies used objective metrics augmented with subjective values to validate techniques. However, this approach is not appropriate for real-time video evaluation. Our proposed framework uses objective metrics (devoid of subjective scores like mean opinion scores) but rather introduced a new metric to validate the framework. The whole framework is validated using compressed/uncompressed signals and varying devices to show the signal differences. Results show that the framework is a step toward feasible incorporation of a VQA tool in a digital terrestrial television model. Using 100 epochs for our simulated video stream, the restricted Boltzmann machine yields a root mean square and mean absolute of 3.6903 and 2.3861 respectively.