3D object recognition using scale-invariant features

Abstract

As 3D scanning technology has developed, it has become easier to acquire various 3D surface data

thus, there is a growing need for 3D data registration and recognition technology. In particular, techniques for finding the exact positions of 3D objects in a cluttered scene in which many parts of an object are occluded and multiple objects may be present is an important technology required by various fields such as industrial inspections, medical imaging, and games. Many existing studies have used local descriptors with local surface patches, and most of these use a fixed support radius so they cannot cope perfectly when the model and scene are at different scales. In this paper, we propose a new object recognition algorithm that exceeds the performance of existing studies. The process of 3D object recognition in a cluttered scene is largely composed of three steps: feature selection, feature description, and matching. In this study, we propose a perfectly scale-invariant feature selection algorithm by extending the 2D SIFT algorithm to a 3D mesh. The feature selection method proposed in this study can obtain highly repeatable feature points and support radii regardless of the scale. The selected features can effectively describe local information using the new shape descriptor proposed in this study. Unlike existing shape descriptors, it is possible to perform scale-invariant 3D object recognition and achieve a high recognition rate when combined with the feature point selection algorithm proposed in this study using the gradients of the scalar functions as defined on the 3D surface. We also reduced the searching space and lowered the false positive rate by suggesting a new RANSAC-based transformation hypothesis generation algorithm. Our 3D object recognition algorithm achieves recognition rates of 99.5% and 97.8% when tested on U3OR and CFVD datasets, respectively, which exceeds the results of previous studies.