Multiscale Sliding Voxel for 3D Planes Detection and Segmentation

Abstract

3D planes detection is a core task of robot navigation and human crafted objects recognition. However, robots operate with restricted computing power and memory, hence; there is a need of highly efficient algorithms with better accuracy. In a previous work, the Sliding Voxel algorithm outperformed state of the art methods such as RANSAC and the Randomized Hough Transform in precision and computational efficiency. However, one of its limitations was the amount of planes it can detect in one execution. In order to solve this limitation, we propose the Multiscale Sliding Voxel for 3D planes detection. A method that executes the Sliding Voxel algorithm at multiple scales and provides an efficient planes consensus and segmentation.