深度相机与计算机视觉：研究主题与应用解析

需积分: 9 69 浏览量更新于2024-07-21 收藏 9.18MB PDF 举报

"《Consumer Depth Cameras for Computer Vision: Research Topics and Applications》是关于深度相机在计算机视觉应用领域的一本经典书籍，对深度相机的开发和理解具有重要参考价值。本书由多位知名专家编纂，包括来自ETH Zürich、Max Planck Inst. for Intelligent Systems、Intel Science and Technology Center以及Industrial Perception的学者，旨在深入探讨消费者级深度相机的研究主题和应用实践。" 正文: 计算机视觉领域的进步不断推动着新技术的发展，而深度相机作为这一领域的重要工具，其在近年来得到了广泛的关注。《Consumer Depth Cameras for Computer Vision: Research Topics and Applications》这本书详细介绍了深度相机的工作原理、技术挑战以及各种应用场景，对于理解并开发基于深度感知的系统至关重要。深度相机，也称为3D相机或结构光相机，通过测量物体与相机之间的距离来生成三维信息。这种技术在自动驾驶、机器人导航、人机交互、虚拟现实(VR)和增强现实(AR)等领域有着广泛的应用。书中涵盖了深度相机的关键技术，如时间飞行法(TOF)、结构光投射和立体匹配等，这些都是理解深度数据获取的基础。计算机视觉研究者和开发者可以通过本书了解到深度相机在真实世界中的应用，例如物体识别、场景理解、人体姿态估计和手势识别。这些应用都需要对深度图像进行处理，包括去噪、校正、配准和三维重建等。书中的实例和案例研究将有助于读者掌握这些技术的实际操作。此外，书中还讨论了深度相机的局限性和未来发展方向。比如，如何提高精度、降低功耗，以及在复杂环境下的鲁棒性问题。随着技术的进步，深度相机正变得越来越小型化、成本更低，并且在低光照条件下的性能也在不断提升。《Consumer Depth Cameras for Computer Vision: Research Topics and Applications》不仅提供了深度相机的理论知识，还为实际项目开发提供了宝贵的指导。无论是学术研究还是工业应用，这本书都是一份宝贵的参考资料，对于想要深入理解和利用深度相机技术的读者来说，它无疑是一本不可或缺的经典之作。

Part I

3D Registration and Reconstruction

Depth cameras are, ﬁrst and foremost, 3D measuring devices. A lot of early research

efforts using consumer depth cameras have been devoted to 3D registration and re-

construction: (1) to quantify the 3D measuring capabilities of Kinect and Prime-

Sense cameras; and (2) to register continuous streams of RGB-D (color+depth)

frames in real-time and to model objects and environments in 3D.

The chapter on 3D with Kinect from Smisek et al. presents an in-depth study

of Kinect cameras as 3D measuring devices. They develop a geometric for Kinect

as a multi-view system with IR, RGB and depth channels. They analyze optical

centers, distortions, and effective resolutions of the Kinect sensor. Accordingly, they

propose a calibration technique for Kinect that outperforms methods in open source

software. Depth data from Kinect are compared to a stereo system (using high-

quality SLR cameras) and a time-of-ﬂight system (SwissRanger SR-4000), where

Kinect performs similar to SLR stereo on a calibration pattern and much better than

the SwissRanger. Depth data from Kinect are combined with RGB-based Structure-

from-Motion to greatly improve the quality of 3D surface reconstruction.

Once we have a solid understanding of the data captured in a single Kinect frame,

the next step is to compute 3D registrations of multiple frames so that they can be

aligned in a global coordinate system. The RGB-D mapping work of Henry et al.

(RGB-D Mapping: Using Depth Cameras for Dense 3D Modeling of Indoor En-

vironments, IJRR, 2012) shows that established algorithms for 3D registration, in-

cluding both the Iterative Closest Point (ICP) and the Bundle Adjustment (BA) algo-

rithms, can adapt to and play an important role in the alignment of RGB-D data. One

interesting aspect of Kinect data is the frame rate (30 Hz): as shown in the Kinect-

Fusion work of Newcombe et al. (KinectFusion: Real-Time Dense Surface Mapping

and Tracking, ISMAR, 2011), 3D registration can run in real-time, utilizing all the

data coming in, when using modern hardware such as GPUs.

The two following chapters both address real-time 3D registration on GPUs: the

use of random ball cover (RBC) for correspondence in ICP matching, and the use of

semi-sparse features. The work of Bauer et al. adapts the RBC data structure, typi-

cally used for high dimensional data, to the Kinect case. This allows a direct nearest-

neighbor search in 6-dimensional RGB-D data, avoiding the approximations needed

Chapter 1

3D with Kinect

Jan Smisek, Michal Jancosek, and Tomas Pajdla

Abstract We analyze Kinect as a 3D measuring device, experimentally investigate

depth measurement resolution and error properties, and make a quantitative com-

parison of Kinect accuracy with stereo reconstruction from SLR cameras and a 3D-

TOF camera. We propose a Kinect geometrical model and its calibration procedure

providing an accurate calibration of Kinect 3D measurement and Kinect cameras.

We compare our Kinect calibration procedure with its alternatives available on In-

ternet, and integrate it into an SfM pipeline where 3D measurements from a moving

Kinect are transformed into a common coordinate system, by computing relative

poses from matches in its color camera.

1.1 Introduction

Kinect [4, 14, 22] has become an important 3D sensor. It has received a lot of at-

tention thanks to the rapid human pose recognition system developed on top of 3D

measurement [17]. Its low cost, reliability and speed of the measurement promise

to make Kinect the primary 3D measuring device in indoor robotics [25], 3D scene

reconstruction [7], and object recognition [12].

In this chapter we provide a geometrical analysis of Kinect, design its geomet-

rical model, propose a calibration procedure and demonstrate its performance. We

extend here our preliminary results presented in [18].

Approaches to modeling Kinect geometry, which have appeared recently, provide

a good basis for understanding the sensor. There exists the following most relevant

work: The authors of [2] combined OpenCV camera calibration [24] with Kinect

J. Smisek (

) · M. Jancosek · T. Pajdla

Center for Machine Perception, Dept. of Cybernetics, FEE, Czech Technical University

in Prague, Prague, Czech Republic

e-mail: smisek.jan@gmail.com

M. Jancosek

e-mail: jancom1@cmp.felk.cvut.cz

T. Pajdla

e-mail: pajdla@fel.cvut.cz

A. Fossati et al. (eds.), Consumer Depth Cameras for Computer Vision,

Advances in Computer Vision and Pattern Recognition, DOI 10.1007/978-1-4471-4640-7_1,

剩余219页未读，继续阅读

许岚

粉丝: 25
资源: 2

深度相机与计算机视觉：研究主题与应用解析

depth_research

Programming Computer Vision with Python 原版PDF by Solem

Time-of-Flight and Structured Light Depth Cameras

generalized object detection on fisheye cameras for autonomous driving: data

labview Vision Utilities

matlab图像处理界面合集

What's the differences between traditional and digital cameras?

cameralink fpga

Recovering Realistic Texture in Image Super-resolution by Deep Spatial Feature Transform

High precision 3D reconstruction based on binocular vision

最新资源