An Invariance-Based Object Registration Architecture using Electronic Tags and XML

They combine some advantages over classic tags, one of which being that they require no processing effort for unambiguous remote detection. The present study considers the problem of 3D object registration from 2D images. This problem, challenging to the image processing community though easily carried out in human vision [6], has been tackled by using range images, multiple images, or active vision techniques such as structured light [3], [8]. 3D recognition from 2D image methods needs resort to certain assumptions about object classes, which limit their generality [6], [11]. Abstract-We present an architecture for knowledge-based object registration that relies on radio-frequency tagging to detect and identify objects. When a tagged object is detected, the network resources are searched for the object’s logic model, and when this model is retrieved various knowledge-based treatments can be performed. The three main components of our system of which we present an early implementation in this paper are the electronic tagging technology, the object logic representation in XML standard, and the object registration method relying on invariance matching. We believe that such architecture will allow new applications in computer vision. Object representation consists in storing a model of the object that best describes it in a given “space of information”. This representation is an important part of a system that implements tag detection, and it must follow some consistency rules, so as to allow network data exchange and automated processing. In order to make models available on the Internet and accommodate for automated parsing, we choose the popular XML standard as a description language for object models. Among object properties, the most critical part for our computer vision system is the description of its geometry.