Finding perceptually optimal operating points of a real time interactive video-conferencing system

video-conferencing system;real time;perceptually optimal operating point;Skype;trade-offs;black box testing;testbed;H.264;network and conversational conditions;frame rate;frame size;mouth-to-ear delay\r(MED);piggybacking;video generation;uneven packet transmission rate;side-by-side;PlanetLab;objective metrics;subjective evaluation

Yeo, Wee Hong ; Wah ; Benjamin W.

关键词: video-conferencing system;    real time;    perceptually optimal operating point;    Skype;    trade-offs;    black box testing;    testbed;    H.264;    network and conversational conditions;    frame rate;    frame size;    mouth-to-ear delay\r(MED);    piggybacking;    video generation;    uneven packet transmission rate;    side-by-side;    PlanetLab;    objective metrics;    subjective evaluation;   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/24302/Yeo_WeeHong.pdf?sequence=1&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

This research aims to address issues faced by real time video-conferencing systems in locating a perceptually optimal operating point under various network and conversational conditions.In order to determine the perceptually optimal operating point of a video-conferencing system, we must first be able to conduct a fair assessment of the quality of the current operating point in the system and compare it with another operating point to determine if one is better than the other in terms of perceptual quality.However at this point in time, there does not exist one objective quality metric that can accurately and fully describe the perceptual quality of a real time video conversation.Hence there is a need for a controlled environment to allow tests to be conducted in and in which we can study different metrics and identify the best trade-offs between them.We begin by studying the components of a typical setup of a real time video-conferencing system and the impacts that various network and conversation conditions can have on the overall perceptual quality.We also look into different metrics available to measure those impacts.We then created a platform to perform black box testing on current video conferencing systems and observe how they handle the changes in operating conditions.The platform is then used to conduct a brief evaluation of the performance of Skype, a popular commercial video-conferencing system.However, we are not able to modify the system parameters of Skype.The main contribution of this thesis is the design of a new testbed that provides a controlled environment to allow tests to be conducted to determine the perceptual optimum operating point of a video conversation under specified network and conversation conditions.This testbed will allow us to modify certain parameters, such as frame rate and frame size, which were not previously possible.The testbed takes as input, two recorded videos of the two speakers of a face-to-face conversation and desired output video parameters, such as frame rate, frame size and delay.A video generation algorithm is designed as part of the testbed to handle modifications to frame rate and frame size of the videos as well as delays inserted into the recorded video conversation to simulate the effects of network delays.The most important issue addressed is the generation of new frames to fill up the gaps created due to a change in frame rate or delay inserted, unlike as in the case of voice, where a period of silence can simply be used to handle these situations.The testbed uses a packetization strategy designed on the basis of an uneven packet transmission rate (UPTR) and that handles the packetization of interleaved video and audio data; it also uses piggybacking to provide redundancy if required.Losses can be injected either randomly or based on packet traces collected via PlanetLab.The processed videos will then be pieced together side-by-side to give the viewpoint of a third-party observing the video conversation from the site of the first speaker.Hence the first speaker will be observed to have a faster reaction time without network delays than that of the second speaker who is simulated to be located at the remote end.The video of the second speaker will also reflect the degradations in perceptual quality induced by the network conditions, whereas the first speaker will be of perfect quality.Hence with the testbed, we are able to generate output videos for different operating points under the same network and conversational conditions and thus able to make comparisons between two operating points. With the testbed in place, we demonstrate how it can be used to evaluate the effects of various parameters on the overall perceptual quality.Lastly, we demonstrate the results of applying an existing efficient search algorithm used for estimating the perceptually optimal mouth-to-ear delay (MED) of a Voice-over-IP(VoIP) conversation to a Video Conversation.This is achieved by using the network simulator designed to conduct a series of subjective and objective tests to identify the perceptual optimum MED under specific network and conversational conditions.