Point Cloud Object Recognition using 3D Convolutional Neural Networks

With the advent of RGB-D technology, there was remarkable progress in robotic tasks such as object recognition. Many approaches were developed to handle depth information, but they work mainly on 2.5D representations of the data. Moreover, the 3D-data handling approaches using Convolutional Neural Networks developed so far showed a gap between volumetric CNN and multi-view CNN. Therefore, the use of point clouds for object recognition has not been fully explored. In this work, we propose a Convolutional Neural Network model that extracts 3D features directly from RGB-D data, mixing volumetric and multi-view representations. The neural architecture is kept as simple as possible to assess the benefits of the 3D-data easily. We evaluate our approach with the publicly available Washington Dataset of real RGB-D data composed of 51 categories of household objects and obtained an improvement of around 10% in accuracy over the utilisation of 2D features. This result motivates further investigation when compared to some recently reported results tested on smaller datasets.

I. Introduction

Nowadays, there is great interest in understanding the way our brain processes 3D information for our everyday activities, amongst it, object recognition. The intricate system that underlies the process of visual recognition raises many questions about what type of processes are treated by different parts of the neural circuitry and how this complex processing is accomplished. Although the hierarchy between the areas that process the 3D information is not known precisely, successive theories in psychology [1], [2], [3], [4], [5] and neuroscience [6] have been tested, supported by advanced technologies of neural recording. In these works, it was discussed how the 3D information would be processed by the brain and if this processing would be directly related to object recognition tasks.