Online smoothing of live video transmissions

In this paper, we consider the problem of online smoothing of live video transmissions, where low latency is critical. To address the low latency requirement, our approach is to wait for a small smoothing window of video frames to be generated, and smooth them prior to their transmission. We present two classes of techniques that adapt an optimal offline algorithm[11] for online use: jumping window algorithms, in which an algorithm executes on consecutive non-overlapping smoothing windows, and sliding window algorithms, in which an algorithm executes on overlapping smoothing windows. We then evaluate these techniques using MPEG-1 traces. We find that the videos smoothed by these techniques show dramatic reduction in their peak rate, coefficient of variation and effective bandwidth. Furthermore, these reductions can be obtained by using very small window sizes, from 1 s to 5 s, indicating that even live video applications that have extremely tight latency requirements can benefit tremendously from using online smoothing. It is also of interest to note that these reductions in peak rate, coefficient of variation and effective bandwidth are very close to those obtained by the optimal offline algorithm. This implies that most of the smoothing benefits that can possibly be seen by a live video transmission can be attained by online smoothing operating on a very small smoothing window, thus incurring only a very small additional latency.