Efficient KNN searching of poses with N-dimensional axis-aligned trees

Abstract

With the rise of advanced technologies like robotics and computer vision, characterizing objects through poses (position and orientation) has become increasingly important. The efficient handeling of 3D data is often done using different kinds of tree type datastructures. One if these widely used data structure is the N-dimesnional axis-aligned tree, however, storing and processing poses using N-dimensional tree structures presents challenges. A limitation arises when using traditional tree structures to capture the continuity property of Euler angles and their associated metric space. This issue becomes evident when performing K-nearest neighbor(KNN) searches on poses, a task for which tree-type structures are typically highly efficient. This research aims to address these complications by exploring the development of specialized structures, such as an octree for 3-dimensional data, that can accommodate nearest neighbor searches on poses.