A concurrent, hierarchical approach to symbolic dynamic scene interpretation

摘要

Traditional methods for image scene interpretation and understanding are based mainly on such single threaded procedural paradigms as hypothesize-and-test or syntactic parsing. As a result, these systems are unable to carry out tasks that require concurrent hypotheses. In this paper we describe a hierarchical, network-of-frames system for symbolically interpreting images. The system is able to interpret a dynamic event based on object motion and interaction among objects. Such capability can be applied to many image applications such as biomedical images, traffic control and behaviour studies. The system has been implemented in an object-oriented environment in the logic programming language Parlog++ and propagates uncertainty through each frame using Baldwin's (1986) formulation. The system is illustrated with the legal interpretation of traffic intersection images