Reconhecimento de formas 2D usando uma técnica sequencial integrada e modelos poligonais.

Abstract

This thesis presents a new methodology for polygonal model based 2D shape recognition. This methodology enables a unified treatment to the model building and recognition problems, through the use of discriminant functions. An artificial vision system implementing the proposed methodology is also described. Among the relevant innovations presented by the system, one can mention the integration of the sequential procedures required and the use of discriminant functions with variable angular thresholds, for hypothesis grouping and validation. Five different hypothesis evaluation techniques have been tested, for five distinct test scenes. The resulting system is able to deal with 2D shapes, isolated or superposed (partially visible) and randomly positioned.