Abstract
This paper presents a descriptor for course alignment of point clouds using conformal geometric algebra. The method is based on selecting keypoints depending on shape factors to identify distinct features of the object represented by the point cloud, and a descriptor is then calculated for each keypoint by fitting two spheres that describe the local curvature. The method for estimating the point correspondences is to a larger extent based on geometric arguments than the method of Kleppe et al. (IEEE Trans Autom Sci Eng, 2017), which results in improved performance. The accuracy of the curvature-based descriptor is validated in experiments, and is shown to compare favorably to state-of-the-art methods in an experiment on course alignment of industrial parts to be assembled with robots.
Highlights
The 3D–3D registration problem [16] is well-established in computer vision, and is still an active field of research
Coarse registration is used to find the initial alignment, which is improved with a fine registration method
This is done by generating a local reference frame at each point using Principal component analysis (PCA), and selecting the points which have a unique geometry based on their shape factors
Summary
The 3D–3D registration problem [16] is well-established in computer vision, and is still an active field of research. A large number of methods have been proposed to solve the registration problem in 3D [4,24]. These methods can be classified as either coarse or fine registration methods. Both course and fine registration may have to be applied in order to find a globally optimal solution. Coarse registration is used to find the initial alignment, which is improved with a fine registration method
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have