Simultaneous object recognition and position tracking for robotic applications

Abstract

Object recognition techniques are well known in the field of machine vision, and aim at the classification of certain observed rigid objects based on the information acquired by a specific sensor. These techniques can either be performed in the 2D image space by simply applying suitable image processing algorithms, or in the real world 3D space by performing surface reconstruction of the object's surface and comparing it with a database of known CAD data. This paper presents a combination of techniques which not only recognize and categorize the observed rigid objects but also compute their 3D pose (position and orientation) with respect an industrial robot's workspace frame. This task is very useful in the field of robotics as it allows an industrial robot to simultaneously recognize and follow the displacements of a specific rigid object, among other objects that may co-exist in the same environment.