Video Traces: Create, Disseminate, Analyze

Arizona State University, Scottsdale AZ

Investigators

Abstract

Encoded video is expected to account for a large portion of the traffic in the Internet of the future and next generation wireless networks. The video carried over future networks is expected to be (1) coded using heterogeneous encoding schemes, (2) coded at a range of quality levels, and (3) to a large extent scalable encoded (e.g., into several layers) to enable scalable multimedia services over highly heterogeneous networks. To date the traffic characteristics of these heterogeneous and scalable encoded videos are largely unexplored (only a few initial observations based on a small number of short video sequences have been made). Even less is known about the implications of these traffic characteristics for the network transport of heterogeneous and scalable encoded video. Therefore, the researcher proposes a comprehensive large scale study of the traffic characteristics and the network transport of heterogeneous and scalable encoded video. The researcher will investigate 1. the traffic characteristics (i.e., statistical properties) in conjunction with the video quality and the content features, and 2. the network capacity, i.e., the number of video streams that a given network can support (multiplexing performance) subject to given quality of service (QoS) constraints both for (a) non scalable video encoded at a range of quality levels using diffierent encoding schemes, and (b) scalable encoded video. These traffic and networking studies are traditionally based on traces of real video traffic. The existing video trace libraries, however, provide only frame size traces of a limited number of single layer MPEG 1 encoded videos. Hence, the existing trace libraries are inadequate for this study (and also for other researchers studying scalable multimedia services over networks). Therefore, the researcher proposes to create an extensive publicly available trace library of heterogeneous and scalable encoded video as the basis for the proposed study. The researchers will (1) encode over 100 di.erent videos, spanning a wide range of video genres and lengths and including diverse texture and motion characteristics. (2) encode each video into a single layer (non scalable) at a range of quality levels using di.erent encoding schemes. (3) encode each video into several quality layers, using the temporal, spatial, SNR, and data partitioning scalability modes, as well as the object based scalability mode of MPEG 4. (4) broaden the notion of video traces by capturing the video traffic, the video content features, and the video quality in traces. (5) disseminate the created video trace library through publicly available web sites. The trace library will stimulate research on protocols and mechanisms for scalable multimedia services in heterogeneous wireline and wireless networks. To date, the research in this area has been hampered by the lack of representative traces of scalable encoded video. In fact, the lack of realistic traffic data for scalable video may have prevented the development of e.cient and reliable protocols for heterogeneous multimedia services. As part of the seed work for the proposed project, the researchers have created a preliminary publicly available library of traces of non scalable MPEG 4 and H.263en coded video (http://www.eas.asu.edu/trace) and studied the statistical properties of the traces in the library. This library is a first attempt at providing MPEG 4 and H.263 video traces. These traces are already widely used by networking researchers. The PIs complement each other in their competencies. Martin Reisslein brings to the project his expertise in developing and evaluating traffic management schemes for video traffic in high speed wireline networks and wireless networks. Sethuraman Panchanathan brings to the project his expertise in video compression, scalable video coding, video content analysis, and MPEG 4.