Probabilistic Pose Estimation from Multiple Hypotheses

Pose estimation assesses the 6D pose of one or many objects in a scene. Considerable attention has been dedicated to advancement algorithms capable identifying orientation multiple within scene cases where partial occlusion occurs. However, only few works focus on developing parallelizable hypotheses-based estimator that naturally handles object symmetries. These should also tackle some issues: meaningless perspectives, with uncertain local poses but single global correct pose, and poses. This paper proposes novel probabilistic algorithm for addresses these issues. combines information from cameras achieve unique prediction assembles information. The is tested over synthetic simulate It achieves rotation error below 1.5°, translation 1.5 pixels datasets used. Those results suggest can handle mentioned issues up certain accuracy. Additionally, method compared against state-of-the-art methodology LineMOD dataset. comparison shows our compete terms