Scene inspection in complex heterogeneous environments by means of knowledge-maps

Automatic ambient inspection in real environments to detect changes due to anomalies, intruders, or other undesirable causes, is a very important and difficult task in many practical applications. In general, real world scenes are complex and comprise heterogeneous objects and entities which require different detection strategies to be reliably detected and controlled. Typically, the inspection task consists in analyzing a predetermined number of critical locations to verify the presence of some entities, their position in the scene and their status. In this paper a system suitable for the inspection of complex real environments is proposed. The inspection is performed by means of a single colour TV camera, i.e., through monocular vision. In order to maximize the overall performance and the scene inspection capabilities an object oriented approach, based on knowledge maps, is adopted. According to it the processing tools and the detection strategies are allowed to change from one zone to another of the same scene. © 1993 IOS Press. All rights reserved.