Sistema de visão robótica para identificação e localização de objetos sob manipulação por robôs industriais em células de manufatura.




There are several techniques for applying robotic vision for identification and localization of industrial objects. These objects are typically found in manufacturing cells. The purpose is to catch and transport the pieces for automatically grouping. Robotic vision for industrial objects is an important and still difficult task, especially feature extraction of the object image. For that, this work proposes a model for recognition of objects different shapes, most of time symmetric, using the extraction of feature points (corners) for creating a database of invariants, for identification and localization. The step of object identification is developed in two different processes: on-line and offline. The off-line step has the objective of creating a database, where the information needed to identify the objects is recorded by using vectors of invariant projective. These vectors are defined from a list of corners extracted from an object image. The order of the corners listed is sorted in some way that its sequence shows the basic geometry of the model. The on-line process has the objective of identifying the objects previously recorded. The localization step was developed to assign positions and orientations of the object in a reference coordinate system. At this stage, it is possible to relate all the coordinate frames previously assigned to the object, like the image, camera or base manipulator frames The recognition of objects, through the identification of invariant projective and its localization in camera coordinates, allows the positioning of a manipulator to catch and to control it position and orientation.


projetivas invariantes vértices identificação e localização de objetos (robótica) visão computacional engenharia mecanica

Documentos Relacionados