Video service systems for networked video libraries.

摘要

Recent advances in digital storage technologies and digital compression techniques have enabled new applications in the area of digital video. Such emerging applications include video analysis tools such as shot cut detection, image-based scene analysis, object tracking, and story extraction. By providing a collection of video objects together with tools which operate on those objects, a digital video library (DVL) is formed. A digital video library provides a facility which permits scholars in a wide range of disciplines to execute research on moving image data by treating video as a first class data type.;While highly desirable, providing real-time, remote client access to a digital video library through the current Internet infrastructure is problematic. Consequently, in the second DVL context, in which remote access is provided to a small number of distant client populations, this thesis discusses the architecture of branch libraries. By maintaining copies of objects from the main repository closer to user population centers than the principal DVL, branch libraries provide remote clients with improved quality of service.;The final context discussed in this thesis considers large, distributed networks of digital video libraries. For such an environment, a novel resource location algorithm is presented called the Caching Tree Algorithm. This distributed algorithm is shown to possess a polynomial (the square of the number of network nodes) worst-case execution time, and simulation results presented for the Caching Tree Algorithm show that it performs well, reducing the operational cost of the network by more than 20% in most cases.;This work discusses digital video libraries in three contexts. In the context of tightly controlled local area networks, this thesis presents the first characterization of DVL server performance. This characterization, which includes video-on-demand applications as well as real-time and non-real-time non-linear applications, is based upon a novel model of the workload generated by digital video library user populations.