3D shape descriptor for objects recognition

Abstract

3D point cloud classification is an important task in applications for many areas such as robotics, urban planning and augmented reality. 3D sensors measure a high amount of points in the 3D scene objects' surface at a high collect rate, so robust techniques are needed to process all input data and also deal with some imprecision. A common solution for these tasks is the use of robust features extraction techniques to gather representative scene information at the lowest computational cost possible. This paper presents a new approach for object recognition in 3D scenes, using a novel 3D shape descriptor which is used as input for a supervised machine learning method. Proposed robust 3D feature is invariant to translation and scale and provides a very simplified object representation for pattern recognition input. Experiments were performed using an Artificial Neural Network to recognize six different object shapes, and obtained results showed that the proposed method is a promising approach for object recognition in 3D scenes.