A GPU Based High-efficient and Accurate Optimal Pose Alignment Approach of 3D Objects

Abstract

In this paper we present a new method for alignment of 3D objects. This approach is based on the exhaustive optimization search in the 3D space using GPU based genetic algorithm. The descriptor of 3D object used as the objective function to be optimized is a newly developed pose-variant similarity measure, which is obtained directly from the voxelized model's geometry and could be entirely implemented on the GPU. In order to reduce the traditional optimal algorithms' large processing time, we exploit the GPU's highly parallel architecture and transport our approach from CPU to GPU. Experimental results show that the proposed method is superior to existing normalization techniques such as PCA and provides a high degree of precision to align 3D objects.