New Compact 3-Dimensional Shape Descriptor for a Depth Camera in Indoor Environments.

Hyukdoo Choi,Euntai Kim

doi:10.3390/s17040876

Abstract

This study questions why existing local shape descriptors have high dimensionalities (up to hundreds) despite simplicity of local shapes. We derived an answer from a historical context and provided an alternative solution by proposing a new compact descriptor. Although existing descriptors can express complicated shapes and depth sensors have been improved, complex shapes are rarely observed in an ordinary environment and a depth sensor only captures a single side of a surface with noise. Therefore, we designed a new descriptor based on principal curvatures, which is compact but practically useful. For verification, the CoRBS dataset, the RGB-D Scenes dataset and the RGB-D Object dataset were used to compare the proposed descriptor with existing descriptors in terms of shape, instance, and category recognition rate. The proposed descriptor showed a comparable performance with existing descriptors despite its low dimensionality of 4.

Highlights

RGB-D sensors with affordable prices and decent performance have been available since 2010, and a new era in 3D computer vision and robotics has begun
The first level test was on shape recognition. It was a primitive performance for local shape descriptors to see how effectively descriptors distinguish between different local shapes
Since local shape descriptors have lower performances than local texture descriptors in general, shape descriptors have lower performances local texture descriptors in general, their Since large local dimensionalities motivated us to figure out the than origin of high dimensionalities and an their large dimensionalities motivated us to figure out the origin of high dimensionalities andtitle an alternative compact descriptor with comparable performance

Summary

Introduction

RGB-D sensors with affordable prices and decent performance have been available since 2010, and a new era in 3D computer vision and robotics has begun. There has been tremendous progress in research dealing with 3D data such as human pose and gesture recognition [1,2], point cloud registration [3,4], simultaneous localization and mapping (SLAM) [5], and object recognition [6]. In these studies, a vector that encodes distinctive property of local region, called a descriptor, plays an important role where descriptors are usually used to find correspondences [3,4,7,8] between two images or to encode a higher level descriptor for objects or scenes. After the monumental work of SIFT [9] and SURF [10], many researchers have competed for the best descriptor in terms of distinctiveness, processing time, and robustness to changes in transformation, noise, and illumination.

Objectives

Results

Conclusion