6D Pose Estimation from Point Cloud Using an Improved Point Pair Features Method

The Point Pair Features (PPF) method has been shown to be effective for pose estimation under clutter and occlusion. Our improved method mainly includes: (1) an approach for solving normal orientations of closed geometries based on Odd-even Rule; (2) an efficient downsampling approach by dividing a voxel grid into equivalent angle cells; (3) a verification step based on fitting points. The method is evaluated on real datasets of clutter and shows better performance as well as efficiency for 6D pose estimation compared to original PPF method.