Normalization and alignment of 3D objects based on bilateral symmetry, planes

Abstract

Recent advancements in 3D scanning technologies have inspired the development of effective methods for matching and retrieving 3D objects. A common pre-processing stage of these retrieval methods is to normalize the position, size, and orientation of the objects based on PCA. It aligns an object's orientation based on PCA eigenvectors, and normalizes its size uniformly in all 3 spatial dimensions based on the variance of the object points. However, orientation alignment by PCA is not robust, and objects with similar shape can be misaligned. Uniform scaling of the objects is not ideal because it does not take into account the differences in the objects' 3D aspect ratios, resulting in misalignment that can exaggerate the shape difference between the objects. This paper presents a method for computing 3D objects' bilateral symmetry planes (BSPs) and BSP axes and extents, and a method for normalizing 3D objects based on BSP axes and extents. Compared to normalization methods based on PCA and minimum volume bounding box, our BSP-based method can normalize and align similar objects in the same category in a semantically more meaningful manner, such as aligning the objects' heads, bodies, legs, etc. © Springer-Verlag Berlin Heidelberg 2007.