A Novel Method for Extrinsic Calibration of Multiple RGB-D Cameras Using Descriptor-Based Patterns.

Hang Liu,Shaorong Xie,Jun Luo,Xiahua Liu,Hengyu Li,Yu Sun

doi:10.3390/s19020349

Abstract

This paper presents a novel method to estimate the relative poses between RGB-D cameras with minimal overlapping fields of view. This calibration problem is relevant to applications such as indoor 3D mapping and robot navigation that can benefit from a wider field of view using multiple RGB-D cameras. The proposed approach relies on descriptor-based patterns to provide well-matched 2D keypoints in the case of a minimal overlapping field of view between cameras. Integrating the matched 2D keypoints with corresponding depth values, a set of 3D matched keypoints are constructed to calibrate multiple RGB-D cameras. Experiments validated the accuracy and efficiency of the proposed calibration approach.

Highlights

In recent years, low-cost and efficient depth and color (RGB-D) devices such as the MicrosoftKinect, Intel RealSense, and Structure Sensor have attracted much attention because of their applications in indoor scene reconstruction and robot navigation
Estimate the Pose from 3D Point Sets. Based on these matched keypoints, we find their corresponding depth values from the depth maps generated by the depth camera in the RGB-D camera to construct 3D point sets to estimate poses
The Kinect v1 camera has an angular field of view (FoV) of 43◦ from the vertical

Summary

Introduction

Low-cost and efficient depth and color (RGB-D) devices such as the MicrosoftKinect, Intel RealSense, and Structure Sensor have attracted much attention because of their applications in indoor scene reconstruction and robot navigation. The depth cameras of these devices can provide a depth map with a VGA resolution (640 × 480) at video rate (e.g., 30 Hz) using efficient light-coding technologies that avoid the challenging task of dense 3D reconstruction from color images. With an RGB-D camera, the simultaneous localization and mapping (SLAM)-based approach is mainly used for fusing point cloud frames to reconstruct indoor scenes [3,4,5,6]. (1) The field of view (FoV) of depth cameras is limited; only a small part of the scene is represented in a single frame. (2) To track the poses of depth cameras to effectively fuse multiple point cloud frames, consecutive frames must be captured to have sufficient scene overlap. More than ninety percent of overlap is required, which further increases the number of frames for reconstruction

Methods

Results

Conclusion