An Improved 6D Pose Estimation Method Based on Point Pair Feature

3D-Object detection and pose estimation of free-form rigid objects is a major part in manipulation tasks. Recently, the feature-based method Hs-PPF (point pair feature method proposed by Hinterstoisser) has shown a promising result with high recall rate and strong robustness against sensor noise and clutter. On this line, we propose here a new voting scheme with a series of improvements including scene points indexes disruption, scene points partition and supporting points removal strategies, which significantly reduces the computational time. The experiments on public available dataset demonstrate that our approach, with reasonable configuration, is much faster than Hs-PPF without significant sacrifice of recognition performance.