Abstract
[Japanese | Thesis | Researches in Minoh Lab | Minoh Lab]


A Video Streaming Method for Lectures that Adapts to the Situations in the Class and The Network Available Bandwidth


It is popular to stream video to distant places with computer network. Most of computer networks are called ``best effort service network'', that deliver data as poss ible, but they don't provide full reliability. It's is difficult to know the available bandwid th of a best effort service network previously, so if we stream data with the quantity over tha n the available bandwidth, that causes disposal of the data packets or jitter of the packet arr ival time. And it leads to disorder of the picture or short pause of the audio.

To avoid such problems, a method was proposed that divides video data into some layers and give priority to the each layer. In this method, if the available bandwidth gets narrow, they don't send the lower priority data. For example, we suppose that we divide the video data by frequen cy band and give high priority to the layer containing low frequency band, and give low priorit y to the layer containing high frequency band. Then, if the available bandwidth is narrow, coar se pictures are streamed, and if it is wide fine pictures are streamed.

As the method that divides the video data into layers, we give high priority to the layer that contains generally statistical important information, such as low frequency band of the picture . But, for example, if we want to stream a lecture video to the distant place, audio informatio n is important on the situation in which the lecturer speaks with no remarkable movement, and c haracter information is important on the situation in which he writes on a blackboard with no s peech. So in this supposition, importance of the information is different by the situations. Th erefore, if we set the purpose of the video streaming, we should apply a method that adapts to the purpose, so the method that sets importance of the information evenly regardless of the pur pose is insufficient.

Therefore, in this paper, picking up lecture video streaming, we propose the method that adapt s to the lecture situation and the available bandwidth. We classified the lecture situations in to five situations, such as "explanation of the teaching materials", "writing on a blackboard", ""speech with movement", "speech only", and "no speech and movement". And we set the three lec ture information such as character information, movement information, and audio information ada pting to the each situations. Audio quality is related to audio information, picture quality is related to character information, and frame rate is related to movement information. Each lect ure information needs a certain level of such related parameters. Therefore, in this method, wh en we stream video, we keep such levels of the parameters related to important information.

Because these parameters affect the quantity of the generated video data, in this method, as w e give priority to the parameters related to more important information, and check the paramete rs related to less important information, we can adapt the quantity of the generated video data to the available bandwidth.

But, depending on the lecture situation and available bandwidth, keeping the parameter related to important information make the quantity of the generated video data exceed the available ba ndwidth. Then, in this case, we propose that we buffer the video data that exceeded the bandwid th, and cost much time to stream the data. And the receiver side waits to play the video for th e whole data needed to play. This make the delay increases, and this method dare to tolerate th at. But this method doesn't allow infinite increasing of the delay. It discards generated video data on the situation that has no important information. It makes it possible to compensate th e delay.

To apply this method to the actual lecture, we made system that recognize the lecture situatio ns with lecture information acquired using result of our laboratory's previous work. The implem ented system did same judge of situation as human did at 82.0\% to simulated lecture. Therefore , this system did proper judge of situation at almost all time of the simulated lecture. Accord ing to recognized situation, we simulated this method assuming available bandwidth is 3Mbps, an d then delay of playing valid.


Go back to Thesis Page