Exploring in partial views: Prediction of 3D shapes from partial scans

Abstract

Recognition of shapes in the three dimensions is a challenging task when point clouds are at the input. Recognition from partial pieces is another challenging task in general. Recognizing objects from partial 3D data is a strong need in industrial automation, autonomous driving, cooperative robot systems including machine-machine and machine-human interactions and in surveillance. This papers investigates the needs and requirements of a system detecting partial point clouds of possible objects, and it proposes a new solution, including local scale estimation, key-point detection and local structure definition for building a 3D recognition or prediction framework.