Object Recognition based on 3D and 2D Descriptors: a comparative study

by Isma Hadji

Introduction

   Object Recognition has long been a wide area of research in computer vision. Several 2D and 3D descriptors exist in order to understand images and cloud of points and  make sense of their content. In this work, we propose to compare some of the most well known 3D descriptors in the task of object recognition. Recently, there has been an increasing interest in this kind of descriptor due to the proliferation of 3D sensors. In addition to comparing 3D descriptors we present a comparison with some of the most widely used 2D descriptors for object recognition. We have used depth images and clouds from the RGB-D dataset for the 3D descriptors and their corresponding rgb images for the 2D descriptors. Also, in our tests with Local descriptors we have used the bag of words technique as a pre-processing step before using a classifier.

Our results prove the importance of having 3D information as it increases the classification performance on the RGB-D dataset.