A multiscale feature extraction approach for 3D range images

摘要

This paper presents a multiscale feature extraction technique for 3D range images. Optimized and improved 3D Gaussian filtering and saliency map computation have been conceived jointly to the exploitation of connectivity relationships naturally induced by the 2D acquisition grid. The proposed algorithmic and implementation solutions guarantee good repeatability of detected features on different views and demonstrated superior computational performance compared to other known approaches.