Automatic estimation of the object pose for industrial robots

摘要

An industrial robot equipped with a camera system requires robust and accurate estimation of the pose of a target object. The pose estimation is achieved by a three-dimensional registration method. This paper presents this novel method for the registration of the rotated and translated part according to its camera images and three-dimensional model. The spatial registration parameters can be computed by finding which projection of the volume is the most similar to the given image. We propose a three-step procedure. First, we compute the Fourier transform (FT) of the volume projection by using FT projection theorem. Then, a cost function is designed according to the magnitude of projection FT and image FT. Last, an optimization algorithm called small world clonal selection algorithm (SWCSA) is applied to obtain spatial registration parameters. SWCSA combines the good local search ability in clonal selection algorithm and the random long-range search operator in small world, so the global optimization ability is enhanced. Experimental results demonstrate that the registration is accurate and robust.