Registro de imagens aplicado à manipulação industrial robotizada

Abstract

Robot referencing in industrial object manipulation applications is typically based on pattern recognition techniques. However, this type of solution is stronghighly dependent on training, is computationally expensive, and is generally not susceptible to a theoretical performance analysis. In this work, the application of image registration algorithms is proposed to estimate the position and orientation of the objects to be manipulated. Therefore, we propose the use of a method based on Phase Correlation, in order to segment the object of interest and, simultaneously, estimate the parameters of interest. As one of the main contributions, considerations about the registration of images using Phase Correlation are presented, highlighting limitations that apparently went unnoticed by the scientific community in the last 25 years. Based on these considerations, under test conditions of interest to the application and assuming the mean absolute error as a figure of merit, the proposed method improves the performance of estimation of rotational movements by more than 1000% in relation to the way in which the Phase Correlation method is presented in the literature. [resumo fornecido pelo autor]