Improving Perceived QoS of Delay-sensitive Video Against A Weak Last-mile: A Practical Approach

We present a novel approach to improve the perceived QoS for video-feeds in delay-sensitive interactive applications under practical last-mile impairments. Conventional approaches consume additional communication and computation resources in the effort to regain the PSNR of the frames by trying to recover the lost information. Such approaches undermine the strict latency requirements leading to eventual frame-drop and reduction in overall system QoE. Instead, we aim to create an approximation of the corrupted frame by regaining the structural information and minimizing the effect of the lost packet(s). The objective is to minimize the possibility of frame drop under weakened last-mile connection so that the remote user can have a better perceived QoS at the application level and can have better probability of reacting to remote scene despite lossy condition. We prove the efficiency of the system under real-life situations with impairments in real field deployment. The benchmarking is performed in terms of full referential video quality metrics against established state-of-the art like WebRTC and RTP.