Demonstration of Energy-Efficient Saliency-Guided Video Communication

Abstract

This paper demonstrates the use of a real-time saliency-guided High Efficiency Video Coding (HEVC) framework for video conferencing applications. Our framework reduces HEVC video communication bandwidth by extracting regions of interest (ROIs) from the input video with machine vision. The ROI information is used to supervise the HEVC encoder to prioritize ROIs in bit allocation and compress background more efficiently. The interval between compute-intensive ROI detections is increased by using motion information from the HEVC encoder to track ROIs between successive detections. Our interactive demonstration allows users to select among four ROI generation modes and directly observe their effects on visual quality, bitrate, computational complexity, and energy consumption. The demonstration setup is built on open-source Kvazaar HEVC encoder and uvgComm video conferencing components, and is publicly available under a permissive 3-Clause BSD license.