Realworld 3D Object Recognition Using a 3D Extension of the HOG Descriptor and a Depth Camera.

Cristian Vilar,Silvia Krug,Mattias O’Nils

doi:10.3390/s21030910

Abstract

3D object recognition is an generic task in robotics and autonomous vehicles. In this paper, we propose a 3D object recognition approach using a 3D extension of the histogram-of-gradients object descriptor with data captured with a depth camera. The presented method makes use of synthetic objects for training the object classifier, and classify real objects captured by the depth camera. The preprocessing methods include operations to achieve rotational invariance as well as to maximize the recognition accuracy while reducing the feature dimensionality at the same time. By studying different preprocessing options, we show challenges that need to be addressed when moving from synthetic to real data. The recognition performance was evaluated with a real dataset captured by a depth camera and the results show a maximum recognition accuracy of 81.5%.

Highlights

Since the popularization of consumer depth cameras, 3D data acquisition and surface description techniques enabled a significant number of new applications like object recognition [1], robot grasping [2], 3D reconstruction [3] or autonomous navigation [4]
Advanced object recognition approaches are typically based on convolutional neural networks (CNNs) to extract a hierarchy set of abstract features from each object to capture key information regarding the class to which it belongs
In order to evaluate the effect of the depth camera non-idealities and the different preprocessing and dataset options defined in Sections 3.1 and 3.2, we defined the sequence of experiments, Table 5

Summary

Introduction

Since the popularization of consumer depth cameras, 3D data acquisition and surface description techniques enabled a significant number of new applications like object recognition [1], robot grasping [2], 3D reconstruction [3] or autonomous navigation [4]. Depth cameras are not affected by illumination changes and are especially suitable for safety applications. Advanced object recognition approaches are typically based on convolutional neural networks (CNNs) to extract a hierarchy set of abstract features from each object to capture key information regarding the class to which it belongs. Despite their recognition performance, CNN approaches demand high computational and memory resources, making them difficult to implement when there are strong limitations in hardware [5].

Objectives

Discussion

Conclusion