Taxonomy of 3D Sensors

A Survey of State-of-the-Art Consumer 3D-Reconstruction Sensors and their Field

of Applications

Julius Sch

oning and Gunther Heidemann

Institute of Cognitive Science, University of Osnabr

uck, Osnabr

uck, Germany

Keywords:

3D Sensors, Time of Flight, Structured Light, Taxonomy, 3D-reconstruction.

Abstract:

Sensors used for 3D-reconstruction determine both the quality of the results and the nature of reconstruction

algorithms. The spectrum of such sensors ranges from expensive to low cost, from highly specialized to out-

of-the-shelf, and from stereo to mono sensors. The list of available sensors has been growing steadily and

is becoming difﬁcult to manage, even in the consumer sector. We provide a survey of existing consumer 3D

sensors and a taxonomy for their assessment. This taxonomy provides information about recent developments,

application domains and functional criteria. The focus of this survey is on low cost 3D sensors at an accessible

price. Prototypes developed in academia are also very interesting, but the price of such sensors can not easily

be estimated. We try to provide an unbiased basis for decision-making for speciﬁc 3D sensors.

In addition to the assessment of existing technologies, we provide a list of preferable features for 3D recon-

struction sensors. We close with a discussion of common problems in available sensor systems and discuss

common ﬁelds of application, as well as areas which could beneﬁt from the application of such sensors.

1 INTRODUCTION

The ﬁrst consumer RGB-D Camera named Kinect

was launched in November 2010 by Microsoft. Be-

fore, RGB-D cameras were only available for special-

ized industrial applications. Triggered by this ﬁrst low

cost consumer RGB-D device, it was believed that

a huge amount of consumer RGB-D cameras would

be available on the market in the upcoming years and

would provide a good alternative to, e.g., laser scan-

ners. Now, six years later, the number of available

RGB-D cameras is still limited. However, the avail-

able low cost RGB-D cameras are widely used in

research for reconstruction (Handa et al., 2014; Cui

et al., 2010), mapping (Henry et al., 2014; Huang

et al., 2011), forensics (Dupuis et al., 2014; Nguyen

et al., 2014), robotics (El-laithy et al., 2012; Yip et al.,

2014) and various other applications (Banerjee et al.,

2014; Gallo et al., 2011).

In this literature survey, we summarize and com-

pare existing low cost 3D sensors. “Low cost” means

a price below 5.000e. We try to verify the statement

by Henry et al. (2014) that RGB-D cameras provide

depth information only up to a limited distance of typ-

ically less than ﬁve meters. We introduce applications

and point out some drawbacks of existing cameras.

Therefore, a special focus on quality and nature of

3D-reconstruction algorithms and processes depend-

ing on these sensors is given. Finally, we discuss com-

mon problems in available sensors using a structured

light camera in different test setups.

Contrary to initial expectations, academic proto-

types like (Zollh

ofer et al., 2014) cannot be taken into

consideration in this taxonomy, because their total

costs cannot be calculated without taking into account

the work that has to be invested in order to set up

such a prototype. Although the hardware costs may

seem favorably small, total costs may well rise above

5.000e if manpower is taken into account. Therefore,

this survey discusses commercial, out-of-the-box sys-

tems only.

2 COMPARISON OF

CONSUMER 3D SENSORS

All cameras discussed here can be assigned to one of

two distinct groups according to their depth measure-

ment principle: structured light (SL) or time of ﬂight

(ToF). Cameras working with structured light emit a

light pattern onto the scene and calculate the depth

194

Schöning, J. and Heidemann, G.

Taxonomy of 3D Sensors - A Survey of State-of-the-Art Consumer 3D-Reconstruction Sensors and their Field of Applications.

DOI: 10.5220/0005784801920197

In Proceedings of the 11th Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2016) - Volume 3: VISAPP, pages 194-199

ISBN: 978-989-758-175-5

Table 1: Comparison of consumer 3D-Cameras; *of measured distance, - not speciﬁed.

Camera Principle measuring

range [m]

Error Res. RGB

Res. Depth

FPS FoV

H×V

PL / SDK Price [e]

Structure

Sensor

SL 0,4 − 3, 5 1%* 640 × 480

640 × 480

30/60 58

◦

× 45

◦

C/C++ 305

Kinect 1

Gen.

SL 0,4 − 3, 5 < 4cm 640 × 480

640 × 480

15/30 57

◦

× 43

◦

C#/C++/VB/

JAVA etc.

160

Xtion PRO

Live

SL 0,8 − 3, 5 - 1280×1024

640 × 480

30/60 58

◦

× 45

◦

C#/C++/

JAVA

140

RealSense SL 0,2 − 1, 2 1%* 1920×1080

640 × 480

60 - C#/C++/

JAVA/

JavaScript

Senz3D ToF 0,2 − 1, 0 - 1080 × 720

320 × 240

30 74

◦

41,6

◦

C++/C 115

Argos 3D -

P100

ToF 0,1 − 3, 0 < 3 %* n/a

160 × 120

160 90

◦

67,5

◦

Matlab/

Labview

1.200

Kinect 2

Gen.

ToF 0,5 − 4, 5 - 1920×1080

512 × 424

15/30 70

◦

× 60

◦

C#/C++/C 160

Swiss

Ranger

4500

ToF 0,8 − 9, 0 < 4cm n/a

176 × 144

10/30 69

◦

× 55

◦

C++/C/

Matlab/

Halcon

3.930

CamBoard

pico

ToF 0,2 − 1, 0 < 6mm n/a

160 × 120

45 82

◦

× 66

◦

C++/C/

Matlab

585

information based on the deformation of the pattern.

In most cases, the emitted light pattern has a wave-

length in the infrared spectrum and is thus invisible

for the human eye. ToF cameras use a laser to emit

a pulse of light and calculate the distance based on

the time span until the pulse is seen by the detector

and the speed of light. Next to this main attribute, we

use the ﬁeld of view (FoV) characteristic, as shown in

Figure 1, for deﬁning the comparison of all sensors in

Table 1. As an indicator for the ﬁelds of application,

the SDKs and the supported programming languages

(PL) for accessing the sensor software APIs are men-

tioned. The most important non-technical attribute of

our comparison matrix is the price which allows us to

calculate a price-performance ratio.

2.1 Structured Light Sensors

According to its fact sheet (Occipital, Inc, 2015),

the Structure Sensor is designed to work at distances

ranging from 40 centimeters to three and a half me-

ters. The producer claims that the error in the z-

direction (the direction of depth) is less than one per-

cent of the actual distance. The camera is based on

the structured light principle and provides a depth im-

age with a resolution of 640×480 pixels at framerates

between 30 and 60 frames per second (fps). However,

its main drawback is its platform limitation since it

is designed to work exclusively with an Apple iPad.

This might potentially discourage Windows and Linux

users, as well as users who require a more computa-

tionally powerful setup.

The ﬁrst generation Kinect sensor (Microsoft,

2015b) is most prominent and well embraced by, e.g.,

the robotics community. It is built on the structured

light principle as well. It features a depth range from

40 centimeters to 3,5 meters, where the distance er-

ror under moderate constraints is below four centime-

ters (Khoshelham and Elberink, 2012). It offers the

highest RGB-D resolution with 1280 × 1024 pixels

at 15fps and a resolution of 640 × 480 at 30fps. To-

gether with its low price, Kinect is an interesting op-

tion for benchmarking 3D-targeted research. How-

ever, since its maximum reliable distance of 3,5 me-

ter might be too small to cover all requirements for

the envisioned ﬁeld of application, it seems ques-

tionable whether it could fulﬁll the practical needs

for applications which require reliability at a larger

scale – however tempting it may seem at ﬁrst.

resolution RGB-D sensor

vertical angle

horizantal angle

near plan

far plan

camera

depth measuring range

Figure 1: Field of view (FoV) characteristic incl. depth

measuring range and resolution of the RGB-D sensor.

Taxonomy of 3D Sensors - A Survey of State-of-the-Art Consumer 3D-Reconstruction Sensors and their Field of Applications

195

A sensor with speciﬁcations and capabilities

comparable to the Kinect is the Asus Xtion PRO

Live (ASUSTeK Computer Inc., 2015). Although the

producer does not state the measurement principle, it

is presumably also a structured light based method.

At distances between 80 centimeters and 3, 5 meters

it offers depth images at a resolution of 640×480 pix-

els. A ﬁxed scanning frequency of 30fps is stated by

the producer. However, with no declared error value,

a slightly higher price and marginally lower resolu-

tion it offers no distinctive advantage over the Kinect

sensor given the current application area.

In the ﬁrst quarter of 2015, Intels RealSense cam-

era was released. It targets the assessment of 3D

point clouds at very small distances with a structured

light depth sensor for distances between 20 centime-

ters and 1,2 meters (Intel Corporation, 2015a,b). Its

most salient advantages are its price, beating all other

considered cameras by a margin of more than 80e,

and the high framerate of 60fps.

2.2 Time of Flight Sensors

A different type of camera uses the ToF physical mea-

surement to assess the depth part of scenes.

The Creative Senz3D (Creative Technology Ltd.,

2015) is a ToF based depth camera with a targeted ap-

plication area in human-computer-interaction (HCI).

With an operating range between 20 centimeters and

one meter it offers the coverage of a person located

directly in front of a computer with the goal to recog-

nize hand and arm gestures to be fed to an interface

control task (cf. research area tangible user interfaces

(TUI)). The resolution of the 3D depth images is lim-

ited to 320 × 240 pixels only, while the regular 2D

webcam part of the sensor offers higher resolutions as

well. Again, the manufacturer states no error values,

which at 30fps and the lowest price of all considered

cameras makes it an end user toy device not suitable

for high quality application areas.

The Argos 3D-P100 (Bluetechnix Group GmbH,

2015) creates depth measurements in a similar set of

ranges as most of the cameras encountered before:

Depth is measured between half a meter and three

meters at an error rate below three percent. A reso-

lution of 160 × 120 pixels is obtained at a framerate

up to 160fps. Like other ToF cameras, the price of

1.200e is above the consumer-eletronics level.

The second generation of Microsoft Kinect does

not use structured light (Microsoft, 2015a) but relies

on the ToF principle. Compared to the ﬁrst genera-

tion, this results in a lower resolution of depth points

(512 x 424 instead of 640 x 480) but also a slightly ex-

tended range of admissible depth values (4, 5 instead

(a)

(b)

Figure 2: Indoor test scenario of an ofﬁce desk with some

objects; (a) shows the RGB channels and (b) shows the

depth channel of the Kinect camera, where depth values

above zero are represented as gray scale gradients.

of 3,5 meters). Its most striking advantage, how-

ever, seems to be the increased horizontal and vertical

viewing angle, giving the opportunity to obtain more

overlapping regions in consecutive depth images.

The Mesa Swiss Ranger (SR) 4500 ToF cam-

era (Heptagon Micro Optics, 2015) offers a quite dif-

ferent depth sensing ability than the other sensors

considered here. It can measure distances between

80 centimeters and nine meters with an error below

four centimeters. The depth resolution of 176 × 144

pixels can be obtained between 10 and 30fps. Unfor-

tunately, the price of nearly 4000 e marks the top end

of the sensors and cameras considered in this survey.

Individual ToF camera modules can be assembled

as, e.g., the PMD CamBoard pico (PMD Technolo-

gies GmbH, 2015). It offers depth measurements be-

tween 20 and 100 centimeters, but no error ﬁgures are

provided. At resolutions of 160 × 120 pixels offered

at 45fps, a three dimensional point cloud can be ob-

tained. Nevertheless, due to the limited low range of

measurement distances, application areas for this sen-

sor seem limited.

VISAPP 2016 - International Conference on Computer Vision Theory and Applications

196

(a) (b) (c) (d)

Figure 3: Two outdoor test scenarios, in the ﬁrst scenario a wheelbarrow with plants without direct sun light; (a) shows the

RGB channels and (b) shows the depth channel of the Kinect camera — in the second scenario a wheelbarrow with plants in

front of trees with direct sun light; (c) shows the RGB channels and (d) shows the depth channel of the Kinect camera, where

depth values above zero are represented as gray scale.

3 APPLICATION

An important question for typical research applica-

tions is whether the camera provides sufﬁcient func-

tionality and quality to perform sophisticated tasks

like 3D reconstruction or mapping. Therefore, RGB

and depth images in different scenarios will be ana-

lyzed. As a ﬁrst test scenario, an indoor scene is eval-

uated using a structured light camera—the Microsoft

Kinect of the ﬁrst generation. Figure 2(a) shows the

test scenario, a desk with ordinary objects like books,

pencils, and input devices. On the right hand, in Fig-

ure 2(b) the eleven bit depth image is shown, where

all areas marked green represent a depth value of zero,

and depth values greater than zero are represented as

grayscale gradient.

A 3D reconstruction application like, e.g., crime

scene investigation might not be able to reconstruct

this test scene because of missing depth information

for some regions in the image. Notable hot spots,

where the depth information does not correspond to

the real depth, frequently occur on very smooth sur-

faces. Two of these spots are the transparent carafe

in the center and the draw pad on the right-hand side.

For 3D reconstruction approaches the depth informa-

tion of transparent objects would be quite important

because RGB based algorithms such as structure from

motion cannot handle transparent objects well (Ihrke

et al., 2010). The missing depth information for the

following processes have to be handled by ﬁlling or

ﬁltering algorithms. One example is the voxel cloud

connectivity segmentation (Papon et al., 2013), which

uses the depth information next to RGB information

for the segmentation process. In case of missing depth

information, the voxel cloud connectivity segmenta-

tion performs hole ﬁlling using the SLIC algorithm.

Another approach to handle missing depth informa-

tion is the reconstruction of objects using inference

from their depth-shadows (Albrecht and Marsland,

2013). Similar to reﬂections on smooth surfaces, in

theory a further problem of cameras working with the

structured light principle is the total absorption of the

emitted (IR) light. However, in praxis we could not

observe such an effect in a scenario with materials like

ﬂeece or velvet.

In the preface to the book “Consumer Depth Cam-

eras for Computer Vision”, Jamie Shotton argues that

the depth camera technology is still not mature and

has a long way to go to reach the frame rates and reso-

lutions possible with traditional sensors, and does not

yet work satisfactory outdoors (Fossati et al., 2013).

To verify this statement about the outdoor capabili-

ties of structured light cameras, we tested with two

outdoor scenarios for 3D reconstruction. In the ﬁrst

outdoor scenario, Figure 3(a) and (b), the Kinect is

used at cloudy weather conditions without direct sun-

light. The depth information of Figure 3(b) is consis-

tent and does not show abnormalities. With sunlight,

as seen in Figures 3(c) and (d), the resulting depth in-

formation is greatly affected by the sunlight and of

very limited use.

Compared to the structured light cameras, time-

of-ﬂight cameras show acceptable results in outdoor

scenarios with or without sunlight. This has already

been evaluated in agricultural robotic scenarios for or-

charding and viticulture (Wunder et al., 2014). How-

ever, time-of-ﬂight cameras still have a depth distance

of nine meters only. This short depth distance limits

the maximal driving speed of, e.g., autonomous vehi-

cles because the inert drive train cannot stop instantly.

In 3D reconstruction, the limitation of the max-

imum depth also causes some drawbacks. With re-

spect to the maximum depth distances, as shown

in Figure 4, these low cost cameras can only scan

small objects like persons or cupboards in a station-

ary setup. Using the camera as a hand scanner as

in the Kinect fusion project (Newcombe et al., 2011;

Microsoft Research, 2015) or similar approaches (Lee

et al., 2014), the mentioned low cost 3D cameras yield

good results. But if we are going to reconstruct large

Taxonomy of 3D Sensors - A Survey of State-of-the-Art Consumer 3D-Reconstruction Sensors and their Field of Applications

197

0 1 2 3 4

5 6

7 8 9

[m]

120

160

300

500

1.200

4.000

[e]

RealSense

Senz3D

Xtion Pro Live

Kinect 1

Gen.

Kinect 2

Gen.

Structure Sensor

CamBoard pico

Argos 3D - P100

Swiss Ranger 4500

Figure 4: Comparison of consumer 3D-Cameras;

price [e] over depth measuring range [m].

monuments (the size of Colognes famous Cathedral),

the hand scanning device is clearly infeasible.

4 DISCUSSION AND OUTLOOK

Next to the price, an important reason for using a

camera in a wide application ﬁeld is the number

of supported programming languages. For exam-

ple, the Microsoft Kinect SDK of the ﬁrst generation

supports over four different programming languages.

As a consequence, SDKs like OpenCV, ROS, PCL

and OpenNI already offer implemented APIs for the

Kinect. The powerful interfaces have led to success in

diverse ﬁelds of application, which is remarkable for

a camera originally designed for video gaming.

Figure 4 conﬁrms the statement of Henry et al.

(2014) that RGB-D cameras provide depth informa-

tion only up to a limited distance of typically less than

ﬁve meters. The Mesa SR4500 with a depth of more

than ﬁve meter is a depth only camera, thus it is not

included in the statement of Henry et al. (2014). With

a maximum reliable depth distance of around ﬁve me-

ters, it appears questionable that requirements of out-

door applications can be fulﬁlled.

Structured Light Sensors are prone to very smooth

or transparent objects, which lead to blind spots in the

depth images. For 3D reconstruction in a controlled

setup, such objects can be modiﬁed to achieve depth

information. For example, DAVID Group (2015) of-

fers a 3D coating spray, that can be applied to such

objects for a reconstruction session and which is easy

to remove. In a setup where very smooth or trans-

parent objects such as windows can not be removed,

the overlaying algorithms have to handle them, exclu-

sively.

Summarizing, existing low cost cameras are a

solid basis for indoor applications. But due to limited

maximal depth and vulnerability by weather condi-

tions (sunlight, fog etc.), outdoor usage remains very

limited. Of course, such limitations can be cushioned

by overlaying algorithms or special setups, but there

are still challenging issues left. We noticed during this

review that in particular the Microsoft Kinect camera

is often regarded as a “professional” 3D camera, but

its original purpose—video gaming—should still be

kept in mind.

Depth information improves the 3D reconstruc-

tion process based on RGB data. But since depth

information may not be available for all objects, it

must be obtained from other sources. This is our mo-

tivation to develop an interactive approach (Sch

oning,

2015; Sch

oning and Heidemann, 2015), where the

user can conveniently ﬁll in missing depth informa-

tion to achieve better 3D reconstruction and more re-

liable models.

ACKNOWLEDGEMENTS

This work was funded by the German Research Foun-

dation (DFG) as part of the Scalable Visual Analytics

Priority Program (SPP 1335).

REFERENCES

Albrecht, S. and Marsland, S. (2013). Seeing the un-

seen: Simple reconstruction of transparent objects

from point cloud data. In 2nd Workshop on Robots

in Clutter.

ASUSTeK Computer Inc. (2015). Asus Xtion Spec-

iﬁcations. http://www.asus.com/Multimedia/

Xtion

PRO LIVE/ speciﬁcations/.

Banerjee, T., Enayati, M., Keller, J. M., Skubic, M.,

Popescu, M., and Rantz, M. (2014). Monitoring pa-

tients in hospital beds using unobtrusive depth sen-

sors. In Engineering in Medicine and Biology Society

(EMBC), pages 5904–5907.

Bluetechnix Group GmbH (2015). Argos 3D - P100 product

website. http://www.bluetechnix.com/en/products/

depthsensing/product/argos3d-p100/.

Creative Technology Ltd. (2015). Creative Senz3D web-

site. http://us.creative.com/p/web-cameras/creative-

senz3d.

Cui, Y., Schuon, S., Chan, D., Thrun, S., and Theobalt,

C. (2010). 3D shape scanning with a time-of-ﬂight

camera. In Computer Vision and Pattern Recognition

(CVPR), pages 1173–1180.

DAVID Group (2015). David - 3D coating spray.

http://www.david-3d.com/products/accessories/

coating-spray-500.

Dupuis, J., Paulus, S., Behmann, J., Plumer, L., and

Kuhlmann, H. (2014). A multi-resolution approach

VISAPP 2016 - International Conference on Computer Vision Theory and Applications

198

for an automated fusion of different low-cost 3D sen-

sors. Sensors, 14(4):7563–7579.

El-laithy, R., Huang, J., and Yeh, M. (2012). Study

on the use of microsoft kinect for robotics applica-

tions. In Position Location and Navigation Sympo-

sium (PLANS), pages 1280–1288.

Fossati, A., Gall, J., Grabner, H., Ren, X., and Konolige, K.,

editors (2013). Consumer Depth Cameras for Com-

puter Vision: Research Topics and Applications (Ad-

vances in Computer Vision and Pattern Recognition).

Springer.

Gallo, L., Placitelli, A., and Ciampi, M. (2011). Controller-

free exploration of medical image data: Experienc-

ing the Kinect. In Computer-Based Medical Systems

(CBMS).

Handa, A., Whelan, T., McDonald, J., and Davison, A. J.

(2014). A benchmark for RGB-D visual odometry, 3D

reconstruction and SLAM. International Conference

on Robotics and Automation.

Henry, P., Krainin, M., Herbst, E., Ren, X., and Fox, D.

(2014). RGB-D mapping: Using depth cameras for

dense 3D modeling of indoor environments. In Ex-

perimental Robotics, pages 477–491. Springer.

Heptagon Micro Optics (2015). SR4500 data sheet. http://

downloads.mesa-imaging.ch/dlm.php?fname=pdf/

SR4500 DataSheet.pdf/.

Huang, A. S., Bachrach, A., Henry, P., Krainin, M., Fox, D.,

and Roy, N. (2011). Visual odometry and mapping for

autonomous ﬂight using an RGB-D camera. In Inter-

national Symposium of Robotics Research (ISRR).

Ihrke, I., Kutulakos, K. N., Lensch, H., Magnor, M., and

Heidrich, W. (2010). Transparent and specular object

reconstruction. In Computer Graphics Forum, vol-

ume 29, pages 2400–2426. Wiley Online Library.

Intel Corporation (2015a). Intel RealSense product brief.

https://software.intel.com/sites/default/ﬁles/managed/

0f/b0/IntelRealSense-WindowsSDKGold PB 1114-

FINAL.pdf.

Intel Corporation (2015b). RealSense 3D from lab to real-

ity. http://iq-realsense.intel.com/from-lab-to-reality/.

Khoshelham, K. and Elberink, S. O. (2012). Accuracy and

resolution of Kinect depth data for indoor mapping ap-

plications. Sensors, 12(2):1437–1454.

Lee, S.-O., Lim, H., Kim, H.-G., and Ahn, S. C. (2014).

RGB-D fusion: Real-time robust tracking and dense

mapping with RGB-D data fusion. In Intelligent

Robots and Systems (IROS), pages 2749–2754.

Microsoft (2015a). Kinect 2 for Windows tech-

nical datasheet. http://www.microsoft.com/en-us/

kinectforwindows/meetkinect/features.aspx.

Microsoft (2015b). Kinect for Windows technical

datasheet. https://readytogo.microsoft.com/en-us/

layouts/RTG/AssetViewer.aspx?AssetUrl=https%3A

%2F%2Freadytogo.microsoft.com%2Fen-us

%2FAsset%2FPages%2F08%20K4W%20Kinect

%20for%20Windows Technical%20Datasheet.aspx.

Microsoft Research (2015). 3D surface reconstruction.

http://research.microsoft.com/en-us/projects/

surfacerecon/.

Newcombe, R. A., Izadi, S., Hilliges, O., Molyneaux, D.,

Kim, D., Davison, A. J., Kohi, P., Shotton, J., Hodges,

S., and Fitzgibbon, A. (2011). KinectFusion: Real-

time dense surface mapping and tracking. In Interna-

tional Symposium on Mixed and Augmented Reality

(ISMAR), pages 127–136.

Nguyen, T. V., Feng, J., and Yan, S. (2014). Seeing hu-

man weight from a single RGB-D image. Journal of

Computer Science and Technology, 29(5):777–784.

Occipital, Inc (2015). Structure Sensor & SDK fact

sheet. http://io.structure.assets.s3.amazonaws.com/

Structure%20Sensor%20Press%20Kit.zip.

Papon, J., Abramov, A., Schoeler, M., and W

org

otter, F.

(2013). Voxel cloud connectivity segmentation - su-

pervoxels for point clouds. In Computer Vision and

Pattern Recognition (CVPR), pages 2027–2034.

PMD Technologies GmbH (2015). Reference design brief

CamBoard pico. http://www.pmdtec.com/html/pdf/

PMD RD Brief CB pico 71.19k V0103.pdf.

Sch

oning, J. (2015). Interactive 3D reconstruction: New op-

portunities for getting CAD-ready models. In Imperial

College Computing Student Workshop (ICCSW), vol-

ume 49 of OpenAccess Series in Informatics (OASIcs),

pages 54–61. Schloss Dagstuhl–Leibniz-Zentrum fuer

Informatik.

Sch

oning, J. and Heidemann, G. (2015). Interactive 3D

modeling - a survey-based perspective on interac-

tive 3D reconstruction. In International Conference

on Pattern Recognition Applications and Methods

(ICPRAM), volume 2, pages 289–294. SCITEPRESS.

Wunder, E., Linz, A., Ruckelshausen, A., and Trab-

hardt, A. (2014). Evaluation of 3D-sensorsystems

for service robotics in orcharding and viticulture.

In VDI-Conference ”Agricultural Engineering” VDI-

Berichte Nr. 2226, pages 83–88. VDI-Verlag GmbH

usseldorf.

Yip, H. M., Ho, K. K., Chu, M., and Lai, K. (2014). De-

velopment of an omnidirectional mobile robot using a

RGB-D sensor for indoor navigation. In Cyber Tech-

nology in Automation, Control, and Intelligent Sys-

tems (CYBER), pages 162–167.

Zollh

ofer, M., Theobalt, C., Stamminger, M., Nießner, M.,

Izadi, S., Rehmann, C., Zach, C., Fisher, M., Wu, C.,

Fitzgibbon, A., and et al. (2014). Real-time non-rigid

reconstruction using an RGB-D camera. ACM Trans-

actions on Graphics, 33(4):1–12.

Taxonomy of 3D Sensors - A Survey of State-of-the-Art Consumer 3D-Reconstruction Sensors and their Field of Applications

199