Database Support for Knowledge-Based Image Evaluation

This contribution introduces a subproblem of image sequence evaluation, the extraction of so-called episodes, and describes the abstraction level of episodes, the system architecture of an episode extraction system EPEK, and discusses the novel requirements that database systems must observe for image evaluation applications.

Abstract

The evaluation of images has to cope with vast amounts of raw data usually acquired within short periods of time. This is especially true for situations where sequences of images must be analyzed in order to determine motions, more complex actions, or to interpret entire processes comprising various combinations of actions. Database systems will become an essential part of the support structure for image evaluation systems in order to facilitate the management of raw data as well as of intermediate and final results. This contribution introduces a subproblem of image sequence evaluation, the extraction of so-called episodes, i.e. actions of relatively high semantic complexity. The extraction is based on a three-part database consisting of an object section, a process section, and a rule section. ‘Ihe contribution describes the abstraction level of episodes, introduces the system architecture of an episode extraction system EPEK, develops a suitable data model for the database of a case study of traffic scenes, determines its implementation in terms of the relational data model, and finally discusses the novel requirements that database systems must observe for image evaluation applications.