Optimizing Visual Appearance used for Refinement of Object Poses

Preben H.S. Holm, Anders G. Buch, and Henrik G. Petersen


Pose refinement, pose estimation


A general algorithm for determining good poses for refinement is presented. It consists of two parts: a coarse filtering stage and a fine filtering stage. The coarse filtering stage generates a random set of poses as well as a set of perturbated poses. For each random pose, we perform a refinement towards all its associated perturbated poses. The result of this process is used to select a subset of the random poses for use in the fine filtering. In the fine filtering we optimize the parameters used for a contour extraction such that the real image matches the simulated image better. Finally, we show experimental results for both simulation and a real robot setup. These experiments indicate that it is possible to make a rough estimate of the performance of the refinement algorithm in order to determine optimal refinement poses.

Important Links:

Go Back