A Compact Planar-patch Descriptor based on Color

Eduardo Fernández-Moral, Javier González-Jiménez and Vicente Arévalo

Mapir Group, Universidad de Málaga, E.T.S. Ingeniería de Informática-Telecomunicación, 29071, Málaga, Spain

Keywords:

Planar-patches Matching, Color Descriptor, RGBD Sensors, Plane-based Mapping, Place Recognition, Loop

Closure.

Abstract:

The representation of the world upon planar patches has proven to be simple, robust and useful for a variety

of robotic tasks, including SLAM, autonomous navigation, or scene recognition. In this work we investigate

how to incorporate color information into such representation to improve the matching of planar patches while

maintaining the model compactness, which is essential for real-time applications. We propose a descriptor

based on the dominant color of the patch, which is deﬁned as the center of the biggest cluster in the patch

histogram. In the paper, different color spaces and methods for extracting the dominant color are analyzed.

We compare this descriptor with a recent proposal (saturated hue based histogram) and provide some con-

clusions on the trade-off between their descriptiveness and compactness. Finally, we present experimental

results showing how our color descriptor can be exploited to increase the efﬁciency of both: plane-based place

recognition and planar patch categorization.

1 INTRODUCTION

The combination of color and depth information pro-

vided by RGBD sensors (like Kinect) has demon-

strated to be useful for a wide range of problems as

3D modeling, object recognition or Simultaneous Lo-

calization and Mapping (SLAM), among others. One

of the main challenges when using RGBD sensors in

real-time robotics applications is how to effectively

process and represent the overwhelming ﬂow of data

they deliver. A suitable way of accomplishing that

is by extracting a structure of planar patches from

it. This strategy has already been applied in prob-

lems such as visual odometry (Martinez-Carranza and

Calway, 2012), augmented reality (Chekhlov et al.,

2007), SLAM (Weingarten and Siegwart, 2006) or

place recognition (Fernández-Moral et al., 2013).

Planar patches, or planes for short, can be efﬁ-

ciently computed from depth images (Poppinga et al.,

2008), (Holz and Behnke, 2013), and can be de-

scribed by very simple geometric features, like the

normal vector and the convex hull (see ﬁg. 1). The

question that we pose here is how to represent the

radiometric information of the planar patches while

maintaining the compactness of the representation.

The answer to this question depends clearly on the

application. In this work, the context of our research

is that of matching planar patches for real-time place

recognition, what involves extensive search for patch

a) b)

Figure 1: Plane-based representation. a) RGB image of the

scene. b) Point cloud representation with the segmented

planar patches superimposed.

correspondences. Thus, selecting a color descriptor

involves the non-trivial issue of maintaining a trade-

off between distinctiveness, compactness, and com-

putational cost.

This problem of ﬁnding a color descriptor for pla-

nar patches was posed recently in (Pathak et al., 2012)

in the context of registering 3D scans, where the au-

thors adopted a hue based histogram to improve the

efﬁciency of registration. In this paper, in contrast,

we explore the idea of ﬁnding a descriptor based on

the dominant color of the plane. This insight, though

intuitive, has not been sufﬁciently studied in the liter-

ature. This problem has interest for the research com-

munity for several reasons: ﬁrst, most planes in in-

door environments do have a dominant color; second,

the dominant color is more robust to the partial obser-

vation of planes than histograms; and ﬁnally, the lack

296

Fernández-Moral E., González-Jiménez J. and Arévalo V..

A Compact Planar-patch Descriptor based on Color.

DOI: 10.5220/0005015102960302

In Proceedings of the 11th International Conference on Informatics in Control, Automation and Robotics (ICINCO-2014), pages 296-302

ISBN: 978-989-758-040-6

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

of distintiveness is largely compensated by the high

efﬁciency required by online back-end processes for

classiﬁcation or place recognition which beneﬁt from

a more compact, fast to compare descriptor.

In order to ﬁnd such compact descriptor, we study

different color spaces and radiometric features, look-

ing for invariance to illumination, point of view and

partial occlusion (section 2). We conclude that a color

descriptor based on the patch dominant color in nor-

malized RGB space provides the best balance between

distinctiveness and compactness. In section 3, this de-

scriptor is compared with the hue based histogram re-

ported in (Pathak et al., 2012), which was previously

proposed for a similar problem, showing that though

the distinctiveness of both are simmilar, our alterna-

tive is advantageous in terms of computation time.

We provide experimental results for place recogni-

tion with plane-based maps using the color descriptor

proposed in this work. We demonstrate that the efﬁ-

ciency of the previous geometry-based solution is sig-

niﬁcantly improved (around 6 times faster) by using

color information. In all the experiments, we compare

our results with with the hue-based histogram.

2 SELECTING A COMPACT

COLOR DESCRIPTOR FOR

PLANAR PATCHES

In this section we address the problem of ﬁnding the

simplest color descriptor for a planar patch focused on

the problem of patch matching. This descriptor must

be highly invariant to viewpoint, lighting conditions

and partial occlusion, and also, it must be efﬁciently

calculated. Note that the utility of this descriptor is

not to unequivocally identify planar patches, but to

prune the search space by adding a very compact ra-

diometric information to the geometric features of the

planar model.

In the context of matching planar patches, a com-

mon solution is that of maximizing the photoconsis-

tency between them (Argiles et al., 2011). The main

limitation of this strategy, which comes as a con-

sequence of the lack of compactness and invariance

of the descriptor, is that maximizing the photocon-

sistency is prohibitively expensive for many applica-

tions, especially when there is not a good initial es-

timation of the registration (e.g. loop closure detec-

tion). Closer to our work are those that describe the

patch radiometric information through its histogram

(Hafner et al., 1995), (Swain and Ballard, 1991). In

this line, (Pathak et al., 2012) posed recently the prob-

lem that we address in this paper: showing how color

information can be exploited to increase the efﬁciency

of 3D scan registration. A well illuminated scene is

assumed in that work, where the authors adopt a hue

based histogram with 2 extra bins to keep intensity

saturated values (black and white), and test different

measures for histogram distance. However, they do

not take into account the fact that many planar patches

have a single color, so that the histogram contains

redundant information. Also, this descriptor is not

robust to partial occlusion, which is rather common

when doing exploration and mapping.

In this paper, we propose to describe the patch

with its dominant color. A similar strategy is used

in video compression (Manjunath et al., 2001) to de-

ﬁne blobs having the same color. In this way the de-

scriptor storage and the computation of distances are

reduced to a minimum. This is important in a number

of problems where many match combinations have

to be checked in real-time. In order to select such

a descriptor we need to address some issues: ﬁrst,

the selection of the color space which offers the best

suitability to obtain an invariant and distinctive domi-

nant color (subsection 2.1); second, to deﬁne the way

this dominant color is extracted (subsection 2.2); and

third, to adapt the descriptor for cases where the dom-

inant color is not reliable enough (subsection 2.3).

2.1 Selection of the Color Space

In order to obtain a distinctive dominant color, the his-

tograms of the patches must be invariant to illumina-

tion conditions, shading and viewpoint. These charac-

teristics are highly dependent on the color space used

to represent the radiometric information, as we show

in the analysis below. Note also that the fact of se-

lecting the dominant color makes the descriptor in-

herently robust to partial occlusion when the physical

plane has a clearly deﬁned dominant color, which is

the most common situation. If this is not the case,

e.g. a textured plane with different colors, the domi-

nant color is not a good descriptor and it should not

be used for matching.

Different color spaces have been studied in the

context of object recognition in (Gevers and Smeul-

ders, 1999). This work concludes that normal-

ized RGB (rgb), saturation and hue (HS), and the

color models c

and l

are highly invariant to

changes in viewing direction and illumination (see ta-

ble 1 for the formulation of these color spaces). Be-

low, we analyze these color spaces for a dataset con-

taining 1000 observations of plane surfaces from dif-

ferent scenarios, spanning diverse viewing conditions

(changing viewpoint and illumination, partial occlu-

sion, etc.). Below we study some relevant properties

ACompactPlanar-patchDescriptorbasedonColor

297

Table 1: Formulation of several color spaces from the RGB

data.

Color space Formulation

r(R, G,B) =

R+G+B

rgb g(R,G,B) =

R+G+B

b(R,G,B) =

R+G+B

HS H(R,G,B) = arctan

 √

3(G−B)

(R−G)+(R−B)



S(R,G,B) = 1 −

min(R,G,B)

R+G+B

(R,G,B) = arctan



max(G,B)



(R,G,B) = arctan



max(R,B)



(R,G,B) = arctan



max(R,G)



(R,G,B) =

(R−G)

+(R−B)

+(G−B)

(R,G,B) =

(R−B)

(R−G)

+(R−B)

+(G−B)

(R,G,B) =

(G−B)

(R−G)

+(R−B)

+(G−B)

of such color spaces:

2.1.1 Histogram Invariance

To extract a dominant color descriptor invariant to

viewpoint (including the effects of partial occlusion

and shades), the histograms main peak must be sta-

ble along different views of the same plane. To mea-

sure the histogram stability in a given color space, we

check the similarity of all histograms corresponding

to the same plane by means of a chi-squared (χ

) dis-

tance measure (Pele and Werman, 2010). This mea-

sure is used to compute the histogram distances of all

pairs of views of the same plane. Then, the mean dis-

tance of all analyzed pairs is averaged for all tested

planes to obtain a global measure of the color space

stability, see table 2.

2.1.2 Histogram Dispersion

The histograms of planes with a well deﬁned domi-

nant color are generally unimodal and with little dis-

persion. However, such characteristic does not ap-

ply to all the planes in the environment, and also, it

varies depending on the color space. To accept that

a plane has a dominant color we make use of a sim-

ple heuristics which requires that at least 50% of the

patch pixels are contained in a bandwidth of ±5% of

the histogram range, centered at such dominant color.

Thus, we deﬁne the concentration rate C as the num-

ber of planes that fulﬁlls this condition in all color

components between the total of planes. We have

found that the above condition is fulﬁlled in 97.5%

for planes represented with rgb and 92.8% for planes

represented with c

, while the other color spaces

Table 2: Suitability of different color spaces to represent

planar patches according to: histogram stability, histogram

dispersion and computation time. The values shown corre-

spond to the average of 100 different planes, with 10 obser-

vations each. For all properties, less means better.

rgb c

Stability χ

0.10 0.11 0.13 0.14

Disp (1 −C) 0.03 0.07 0.74 0.77

C. time (µs) 10.7 104.9 23.0 11.3

present much lower rates. Table 2 shows the disper-

sion rate in this experiment, deﬁned as (1 −C).

2.1.3 Computation Time

Another important criterion to consider is the com-

putation time required to transform the original color

space to the target one. This is less critical because

this cost is small in comparison with the whole pro-

cess of segmenting the planes, whichever the chosen

color space is. The average of this time for this dataset

is also indicated in table 2.

Taking into account the criteria studied above, we

notice that rgb is the one with the best properties, and

therefore, it is the one adopted in the rest of this paper.

2.2 Computing the Dominant Color of a

Plane

There exist several ways to deﬁne the dominant color

for a planar patch. In this work we have tested the

mode of the histograms, and the centroid of the largest

cluster extracted with two variants of the mean shift

algorithm: with ﬁxed (FMS) and variable bandwidth

(VMS), respectively. Mean shift has been broadly

used for color segmentation (Comaniciu and Meer,

1997). Though it has limitations for real-time appli-

cations due to its computational cost, in our case the

cost of the mean shift is affordable since most his-

tograms present unimodal distributions and we only

extract one cluster, so that it converges in very few

iterations.

We compare the distinctiveness of the dominant

color obtained with the above techniques using a bi-

nary classiﬁer based on the color difference of two

patches, expressed as |c

−c

|. Thus, when this dif-

ference is larger than a threshold the patches are con-

sidered to belong to different physical planes. This

classiﬁer is tested, for a range of thresholds, with the

previous dataset in which we know beforehand which

observation corresponds to each plane (i.e. the ground

truth).

From this experiment we obtain the distinctive-

ness of this classiﬁer in terms of its sensitivity (ratio of

actual positives which are correctly identiﬁed) and the

ICINCO2014-11thInternationalConferenceonInformaticsinControl,AutomationandRobotics

298

0.4 0.5 0.6 0.7 0.8 0.9 1

0.6

0.7

0.8

0.9

Specificity

Sensitivity

ROC color classifier

VMS

FMS

Mode

Figure 2: ROC curves (sensitivity vs. speciﬁcity) of the

color constraints as binary classiﬁers.

speciﬁcity (ratio of negatives which are correctly re-

jected) for the different techniques to obtain the dom-

inant color. These results are depicted as ROC (Re-

ceiver Operating Characteristic) curves in ﬁg. 2. Ev-

ery point of each curve represents a different thresh-

old for the classiﬁer, thus, more restrictive thresholds

result in higher sensitivity and lower speciﬁcity. Note

that the nearer the curve is to the optimum point (1,1)

the better the classiﬁer is. From this test we conclude

that VMS provides the most distinctive dominant color

since both, sensitivity and speciﬁcity, are higher than

for the mode and FMS for any threshold.

2.3 Dealing with Non-distinctive

Dominant Color

An important issue when describing patches with

their dominant color is dealing with those cases where

this description is not applicable. For example, we

can not use the dominant color as descriptor for planes

without a prevalent color (e.g. a checkerboard). To

take this situation into account we add a boolean to

our color descriptor to specify whether the distribu-

tion of the plane histogram in rgb has a low disper-

sion, as explained in the previous subsection.

Another case of interest is that of recognizing very

distinct intensity levels (e.g. black and white planes

have the same values r = g = b = 0.33). Despite this

problem depends on illumination, we reckon that a

minimum illumination is required to distinguish dif-

ferent colors, and such minimum can be sufﬁcient to

distinguish gray levels far apart when enough illumi-

nation is available. For this case, we propose to in-

clude in the descriptor the average intensity (Av.Int),

which is calculated as the average (R + G + B/3) of

the inliers supporting the dominant color given by the

previous mean shift segmentation. Thus, this part of

the descriptor can be used for well illuminated scenes

(what is common to many environments).

To sum up, the resulting descriptor contains 4 ele-

0.4 0.5 0.6 0.7 0.8 0.9 1

0.6

0.65

0.7

0.75

0.8

0.85

0.9

0.95

Specificity

Sensitivity

ROC color constraints

Normalized rgb

Dominant Brgb

Hue Histogram

Figure 3: ROC curves (sensitivity vs. speciﬁcity) of differ-

ent color descriptors: dominant rgb, dominant rgb including

intensity and hue histogram.

ments that are stored in a word of 4 bytes: 2 bytes for

normalized color r and g (note that b depends on these

two as r +g+b = 1), 1 byte for the plane intensity and

1 bit for to specify the existence of a dominant color.

3 DOMINANT COLOR

DESCRIPTOR vs. HUE

HISTOGRAM

In this section we evaluate the distinctiveness of the

proposed dominant color based descriptor and com-

pare it with the normalized, saturated hue histogram

proposed in (Pathak et al., 2012). To compare both

in the same conditions such evaluation is performed

in well illuminated scenes, and so, the intensity part

of our descriptor is also employed. Following the

work of (Pathak et al., 2012), we implement the paper

method with 74 bins, 72 bins for hue values and two

more bins for saturated black and white, and compare

the patch histograms h

with the Bhattacharyya dis-

tance (Bhattacharyya, 1946):

B(h

) =

1 −

∑

k=1

[k] ·h

[k] (1)

The sensitivity and speciﬁcity of a binary classiﬁer

based on the compared descriptors are evaluated using

different thresholds as we did in the previous section

(see ﬁg. 3). As expected, we observe that the pro-

posed descriptor is signiﬁcantly more distinctive than

the rgb dominant color, since the latter lacks informa-

tion about the reliability of the dominant color, and

also it cannot distinguish between different grayscale

levels.

By comparing our descriptor with the hue based

histogram we observe that their distinctiveness are

similar despite the richer information of the latter (see

ACompactPlanar-patchDescriptorbasedonColor

299

Figure 4: Plane based representation of a living room. The colored planes at the right have been extracted from the point

cloud at the left.

ﬁg. 3). The reason for this is that most planes have a

prevalent color in our test environments. The fact that

the sensitivity of the hue histogram is slightly lower is

explained because the histogram is less robust to par-

tial observation of planes. Contrarily, this descriptor

should perform better for textured surfaces and when

the patches present no occlusion, however, such cases

are rare in the home and ofﬁce environments we are

working in, where our dominant color descriptor is

more suitable.

Besides the distinctiveness of the descriptor, the

compactness and the computational cost of measur-

ing distances are two relevant issues in this work. The

color descriptor we propose is the most compact rep-

resentation for a number of applications, including

patch matching or visualization (Pathak et al., 2012).

Regarding the computation of distances between de-

scriptors, in our case it is reduced to a subtraction in

each element, while the computing distance between

histograms (e.g. with the Bhattacharyya distance) re-

quires more computation. This is an important advan-

tage when we require to perform this operation inten-

sively in real time as we will see in next section.

4 EXPERIMENTAL RESULTS

The experimental validation of the proposed color de-

scriptor is presented next for two different problems.

First we evaluate the improvement for recognizing

previous visited places based on a planar description

of the scene. Second, we analyze the advantages for

planar patch classiﬁcation using a random forest.

4.1 Improving Place Recognition

In this experiment, the proposed descriptor is

tested in a plane-based place recognition applica-

tion (Fernández-Moral et al., 2013) (available in

http://www.mrpt.org/pbmap). This application works

by matching sets of neighboring planar patches based

on their geometric attributes and their relations. Here,

we incorporate our dominant color based descriptor

to speed up the search process. Figure 4 shows an ex-

ample of such planar representation where each patch

is colored with its dominant color. We demonstrate

that the little expense of including the color descrip-

tor improves signiﬁcantly the matching of planes, and

so, the place recognition performance.

In this application the planar patches are ef-

ﬁciently extracted at frame rate from RGBD im-

ages using a region growing technique (Holz and

Behnke, 2013). Such planes are organized in a graph

which stores the connection between close-by planar

patches, so that a subgraph of connected nodes rep-

resents a local neighborhood of planes. Place recog-

nition is addressed as a problem of matching neigh-

borhoods of planes, thus, the poor distinctive infor-

mation of a single plane is compensated by the strong

relations between neighbor planes.

In order to match two sets of neighbor patches,

an interpretation tree is utilized to apply geomet-

ric restrictions in the form of unary and binary con-

straints (Grimson, 1990). These constraints make use

of thresholds that have been determined experimen-

tally from training carried out in diverse indoor sce-

narios: ofﬁce and home places. An important ad-

vantage of this approach is its ﬂexibility to recognize

places when the planes are partially observed or miss-

ICINCO2014-11thInternationalConferenceonInformaticsinControl,AutomationandRobotics

300

0.3 0.4 0.5 0.6 0.7 0.8 0.9

0.6

0.7

0.8

0.9

Specificity

Sensitivity

ROC unary constraints

Area

Elong

Color

Figure 5: Comparison of the different unary constraints by

their ROC curves (sensitivity vs. speciﬁcity).

ing. Yet, a planar-based scene representation has the

beneﬁt of being more robust to changes of viewpoint

and tolerates reasonably well changes in the scene.

This solution performs intensive computation to

search for previous places (loop closure) every time

the map is updated at 30 Hz. By adding the color

descriptor to this geometric description we can prune

branches of the interpretation tree to ﬁnd the solution

more efﬁciently. Thus, the proposed color descrip-

tor introduces a new unary constraint to avoid the

matching of planes with different color. To arrange

the order in which the new restriction will be eval-

uated, we compare its distinctiveness with the other

two unary restrictions: area and elongation of the pla-

nar patches. Similarly to the previous section, we es-

timate the ROC curves showing the balance between

sensitivity and speciﬁcity of the unary restrictions for

a range of thresholds, see ﬁg. 5. From this graph we

can see that the color is the most discriminative fea-

ture and, since all unary restrictions require similar

computation, we arrange their application order con-

sequently: color, area and elongation.

To illustrate the performance improvement, we

have carried out an experiment where we measure the

time required for searching a place with and with-

out the color descriptor. Figure 6 shows the aver-

age time of such search with respect to the number

of planes being evaluated. We observe that perform-

ing the search using the proposed color descriptor is

around 6 times faster. Such rate varies from 2 to 10

depending on the radiometric characteristics of the

planar surfaces of the environment. This presents

a signiﬁcant increase of efﬁciency over the previous

pure-geometric solution.

5 CONCLUSIONS

This paper presents a simple, highly compact color

4 5 6 7 8 9 10

x 10

Num planes to match

Num restrictions checked

Geometry

Geometry & Color

Figure 6: Performance of the place recognition process

using both: only geometry and color and geometry in

PbMaps.

descriptor for planar patches which can be used to im-

prove the performance of matching and classiﬁcation

algorithms based on planar models. In essence, we

propose to approximate the patch color by the most

representative color in it in normalized rgb space.

Though this idea may look simplistic and naive, we

notice that it has not been employed before, probably

due to the difﬁculty of ﬁnding an invariant descriptor

through such little information, and only tuned his-

tograms have been employed in similar problems. To

address the problem of ﬁnding an invariant color at-

tribute we have evaluated different color spaces and

different dominant color selection strategies, and have

included some extra information about its reliability

and its saturation. The proposed descriptor contains

the dominant rgb color of the plane, extracted us-

ing a mean-shift algorithm, the average intensity of

this dominant color and a boolean to indicate whether

such dominant color is representative enough. We

have veriﬁed in common ofﬁce and home environ-

ments that this descriptor is as distinctive as the hue

based histogram proposed previously, while it is more

compact and faster to compare. We demonstrate that

the performance of a place recognition application

based on matching planar patches is signiﬁcantly im-

proved (about 6 times speed-up) by using the compact

color descriptor proposed here. This descriptor has

also been tested to improve planar patch classiﬁcation

by adding such information to a compact geometric

descriptor, showing that the recognition rate improves

at the expense of a small increase in computation.

REFERENCES

Argiles, A., Civera, J., and Montesano, L. (2011). Dense

multi-planar scene estimation from a sparse set of im-

ages. In Intelligent Robots and Systems (IROS), 2011

IEEE/RSJ International Conference on, pages 4448–

4454.

ACompactPlanar-patchDescriptorbasedonColor

301

Bhattacharyya, A. (1946). On a measure of divergence be-

tween two multinomial populations. Sankhy

a: The In-

dian Journal of Statistics (1933-1960), 7(4):401–406.

Chekhlov, D., Gee, A., Calway, A., and Mayol-Cuevas,

W. (2007). Ninja on a plane: Automatic discovery

of physical planes for augmented reality using visual

slam. In Proceedings of the 2007 6th IEEE and ACM

International Symposium on Mixed and Augmented

Reality, pages 1–4. IEEE Computer Society.

Comaniciu, D. and Meer, P. (1997). Robust analysis of fea-

ture spaces: color image segmentation. In Computer

Vision and Pattern Recognition, 1997. Proceedings.,

1997 IEEE Computer Society Conference on, pages

750 –755.

Fernández-Moral, E., Mayol-Cuevas, W., Arévalo, V., and

González-Jiménez, J. (2013). Fast place recognition

with plane-based maps. In Robotics and Automation

(ICRA), in 2013 IEEE International Conference on.

IEEE.

Gevers, T. and Smeulders, W. (1999). Color based object

recognition. Pattern recognition, 32(3):453–464.

Grimson, W. E. L. (1990). Object Recognition by Computer

- The role of Geometric Constraints. MIT Press, Cam-

bridge, MA.

Hafner, J., Sawhney, H., Equitz, W., Flickner, M., and

Niblack, W. (1995). Efﬁcient color histogram in-

dexing for quadratic form distance functions. Pat-

tern Analysis and Machine Intelligence, IEEE Trans-

actions on, 17(7):729 –736.

Holz, D. and Behnke, S. (2013). Fast range image seg-

mentation and smoothing using approximate surface

reconstruction and region growing. In Intelligent Au-

tonomous Systems 12, pages 61–73. Springer.

Manjunath, B., Ohm, J.-R., Vasudevan, V., and Yamada, A.

(2001). Color and texture descriptors. Circuits and

Systems for Video Technology, IEEE Transactions on,

11(6):703 –715.

Martinez-Carranza, J. and Calway, A. (2012). Efﬁcient vi-

sual odometry using a structure-driven temporal map.

In Robotics and Automation (ICRA), 2012 IEEE Inter-

national Conference on, pages 5210–5215.

Pathak, K., Vaskevicius, N., Bungiu, F., and Birk, A.

(2012). Utilizing color information in 3d scan-

registration using planar-patches matching. In Multi-

sensor Fusion and Integration for Intelligent Systems

(MFI), 2012 IEEE Conference on, pages 371–376.

Pele, O. and Werman, M. (2010). The quadratic-chi

histogram distance family. Computer Vision–ECCV

2010, pages 749–762.

Poppinga, J., Vaskevicius, N., Birk, A., and Pathak, K.

(2008). Fast plane detection and polygonalization in

noisy 3d range images. In Intelligent Robots and Sys-

tems, 2008. IROS 2008. IEEE/RSJ International Con-

ference on, pages 3378–3383. IEEE.

Swain, M. J. and Ballard, D. H. (1991). Color indexing.

International journal of computer vision, 7(1):11–32.

Weingarten, J. and Siegwart, R. (2006). 3d slam using pla-

nar segments. In Intelligent Robots and Systems, 2006

IEEE/RSJ International Conference on, pages 3062–

3067.

ICINCO2014-11thInternationalConferenceonInformaticsinControl,AutomationandRobotics

302