X WORKSHOP DE AGENTES FÍSICOS, SEPTIEMBRE 2009, CÁCERES 1

Visual Goal Detection for the RoboCup Standard

Platform League
José M. Cañas, Domenec Puig, Eduardo Perdices and Tomás González

Abstract—This paper presents a new fast and robust goal platform league (SPL). Maybe the most appealing one is the
detection system for the Nao humanoid player at the RoboCup SPL as the hardware is exactly the same for all participants.
standard platform league. The proposed methodology is done The behavior quality and performance differences lie com-
totally based on Artificial Vision, without additional sensors.
First, the goals are detected by means of color based segmentation pletely in the software. In addition, the code of the robots
and geometrical image processing methods from the 2D images must be publicly described, so the knowledge sharing pushes
provided by the front camera mounted in the head of the Nao the overall quality. Until 2007 the hardware platform was the
robot. Then, once the goals have been recognized, the position Sony Aibo. Since 2008 the SPL hardware platform is the
of the robot with respect to the goal is obtained exploiting 3D Aldebaran Nao humanoid (Fig.1). Its main sensors are two
geometric properties. The proposed system is validated with real
images by emulating real RoboCup conditions. non-stereo cameras and with it the teams have been exposed
to the complexity of biped movement.
Index Terms—RoboCup and soccer robots, Artificial Vision
and Robotics, Nao humanoid, Goal Detection, Color Segmenta-
tion, 3D geometry.

I. I NTRODUCTION
1

T HE RoboCup is a scientific annual competition aimed

to foster Robotics and Artificial Intelligence research.
It offers soccer as a dynamic, competitive and cooperative
benchmark for testing the robotics technology and pushing
it forward. Its long term goal is to build a soccer team of
robots able to beat the human world champion team by 2050.
Maybe this is the equivalent to the long tearm milestone of Figure 1. Nao humanoid and Webots simulator
AI community with artificial chess players, and Deep Blue
defeated Gary Kasparov in 1997. The current state of the In SPL, the robot players must be completely autonomous.
robotics technology is far from such ambitious goal, but In order to build a soccer robot player many different habilities
progress has been made since the first RoboCup was celebrated must be programmed, both perceptive and motion or control
in 1997. oriented. For instance the goto ball behavior, the follow-
In the last years several public challenges and competitions ball behavior, the ball detection, the kicking, self-localization,
have arisen around robotics. For instance DARPA Grand standing up in case of fall, etc.
Challenge and Urban Challenge have contributed to foster This work is focused in the goal detection, based on the
the research in robotics, providing proofs of concept about camera images of the Nao. The goal detection helps the robot
the feasibility of autonomous robot on real transportation to decide whether to kick the ball towards the opponent’s goal
missions. or just turn to clear the ball out of its own goal. It can also
RoboCup has worldwide scope and in the last years has provide good information to self-localization inside the game
included new categories beyond soccer: Junior, Rescue and field.
Home. The last ones trying to reduce the gap between the
The rest of this paper is organized as follows. Section II
contest and real applications. Several new leagues have also
reviews the state of the art in artificial vision systems in the
appeared around the soccer category, depending on the robot
RoboCup. In section III several solutions to the same problem
size and shape: small size, middle size, humanoid and standard
of goal detection in the images are proposed. Section IV
José M. Cañas and Eduardo Perdices are with the Rey Juan Carlos proposes a technique for obtaining spatial information from
University. E-mail: jmplaza@gsyc.urjc.es the previously detected goals. Section V shows experiments
Domenec Puig and Tomás González are with the Rovira i Virgili
with proposed techniques. Finally, conclusions and further
University. E-mail: domenec.puig@urv.cat improvements in given in section VI.
This work has been partially funded by projects RoboCity2030 (ref.S-
0505/DPI/0176) of the Comunidad de Madrid, and by the Spanish Ministries II. V ISION BASED SYSTEMS IN THE ROBO C UP
of Education and Science under projects DPI2007-66556-C03-03 and
DPI2007-66556-C03-01. Over the last years, considerable effort has been devoted to
1 www.robocup.org the development of Artificial Vision systems for the RoboCup
2 X WORKSHOP DE AGENTES FÍSICOS, SEPTIEMBRE 2009, CÁCERES

soccer leagues. In this way, the increasing competitiveness and B. Related work on visual goal detection
evolution of the RoboCup leagues has conducted to vision sys- In general, the aforementioned research has been oriented
tems with high performance, which are addressing a variety of to solve visual tasks in the environment of the RoboCup.
typical problems [12], such as perception of natural landmarks However, some of those works have specifically proposed
without geometrical and color restrictions, obstacle avoidance, solutions to the problem of goal detection.
pose independent detection and recognition of teammates and In this regard, one of the earliest approaches was given by
opponents, among others. Cassinis and Rizzi [9] that performed a color segmentation
Several constraints in the RoboCup domain make difficult method using a region-growing algorithm. The goal posts are
the development of such vision systems. First, the robots then detected selecting the boundary pixels between the goal
always have limitated processing power. For instance, in the color and the white field walls. After that, image geometry is
Nao humanoid a single AMD Geode 500Mhz CPU performs used to distinguish between the left and the right goal post.
all the onboard computations and the Naoqi middleware con- The aforementioned work described in [8] has been also
sumes most of that capacity. Second, the robot cameras use to applied to the detection of the goals in the RoboCup leagues.
have poor quality. In the Aibos the camera was of 416x320 In this way, they detect goals by the size of the regions
pixels and the colors were not optimal. Third, the camera is obtained after applying the color based image segmentation
constantly in motion, not stable in height as the robot moves mentioned above. Moreover, [14] aims at recognizing the
through the field. vertical goal posts and the goal crossbar separately. Both
horizontal and vertical goal indications and confidence levels
are derived from the horizontal and vertical scanning of the
A. Background images, according to the amount of lines detected. Afterwards,
it is decided whether the previously obtained indications can
A number of initiatives for developing vision systems
be combined to offer a single goal indication and, finally,
conceived to give solutions to the aforementioned typical
different filters are used to reject unlikely goal indications.
problems have been carried out in recent years. In this line,
early work by Bandlow et al. [8] developed a fast and robust
III. G OAL DETECTION IN 2D
color image segmentation method yielding significant regions
in the context of the RoboCup. The edges among adjacent Two different approaches oriented to the detection of the
regions are used to localize objects like the ball or other robots goals that appear in the 2D images are described in the
on the play field. Besides, Jamzad et al. [10] presented several next two subsections. The first one puts the emphasis on the
novel initiatives on robot vision using the idea of searching on geometric relations that must be found between the different
a few jump points in a perspective view of robot. Thus, they parts that compose a goal, while the second is focused on edge
performed a fast method for reliable object shape estimation detection strategies and specifically in the recognition of pixels
without the necessity of previously segmenting the images. belonging to the four vertices of a goal: Pix1, Pix2, Pix3 and
On the other hand, the work by Hoffmann et al. [11] Pix4 as shown in Fig.2.
introduced an obstacle avoidance system that is able to detect
unknown obstacles and reliably avoid them while advancing A. Detection Based on Geometrical Relations
toward a target on the play field of known color. A radial The first proposed method is intended to be robust and fast
model is constructed from the detected obstacles giving the in order to overcome some of the usual drawbacks of the vision
robot a representation of its surroundings that integrates both systems in the RoboCup, such as the excessive dependency of
current and recent vision information. the illumination and the play field conditions, the difficulty
Further vision systems include visual detection of robots. in the detection of the goal posts depending on geometrical
In this sense, Kaufmann et al. [13] proposed a methodology aspects (rotations, scale,. . . ) of the images captured by the
that consists of two steps: first, the detection of possible robot robots, or the excessive computational cost of robust solutions
areas in an image is conducted and, then, a robot recognition based on classical Artificial Vision techniques. The proposed
task is performed with two combined multi-layer perceptrons. approach can be decomposed into different stages that are
Moreover, an interesting method presented by Loncomilla described in the next subsections.
and Ruiz-del-Solar in [12] describes an object recognition 1) Color calibration: The first stage of the proposed
system applied to robot detection, based on the wide-baseline method consists of a color calibration process. Thus, a set
matching between a pattern image and a test image where the of YUV images acquired from the front camera of the Nao
object is searched. The wide-baseline matching is implemented robot is segmented into regions representing one color class
using local interest points and invariant descriptors. each.
Furthermore, recent work proposed by Volioti and Fig.2 shows an example image captured by the Nao robot
Lagoudakis [14] presented a uniform approach for recognizing containing a blue goal.
the key objects in the RoboCup This method proceeds by The segmentation process is performed by using a k-means
identifying large colored areas through a finite state machine, clustering algorithm, but considering all the available centroids
clustering of colored areas through histograms, formation of as initial seeds. Thus, in fact, seven centroids are utilized,
a bounding boxes indicating possible presence of objects, and corresponding to the colors of the ball (orange), goals (yellow
customized filtering for removing unlikely classifications. and blue), field (green), robots (red and blue) and lines (white).
CAÑAS, PUIG, PERDICES, AND GONZÁLEZ: VISUAL GOAL DETECTION FOR THE NAO HUMANOID PLAYER AT THE ROBOCUP 3

The line across these points is the intersection line between

the geometral plane and the image plane. In fact, the goal
posts will be searched above this line. Fig.4 (left) displays the
intersection between the geometral plane and the image plane
corresponding to the example image in Fig.2.
Furthermore, intersections among the grid and the top blue
or yellow pixels in the image are detected (taking into account
the inclination of the image). The line across those points
Figure 2. Example of original image from a RoboCup competition constitutes the intersection among the horizon plane and the
image plane. It is expected not to find useful information in the
images above this line. Fig.4 (write) displays the intersection
between the horizon plane and the image plane in the example
image in Fig.2. Note that, by definition, the geometral and the
horizon planes are parallel and delimit the region where the
goals are expected to be found.
3) Goal Posts Detection: The overall aim of this process
is to extract the goal posts and other interesting features that
could reinforce the detection of goals in the play field.
First of all, the color prototypes obtained as explained in
Section III.A1 are used to segment the blue and yellow goal
Figure 3. Color segmentation posts and crossbars. In order to do this, not all the image
pixels are analyzed, but a high resolution sampling grid is
utilized in order to detect blue or yellow lines in the image.
The range between the minimum and the maximum YUV Fig.5 depicts the detected lines corresponding to a blue goal
values in the regions obtained after that clustering stage are (long lines correspond to the posts and short blue lines to the
considered as the actual prototype values that characterize crossbar) corresponding to the example image in Fig.2.
each color class of interest. Fig.3 depicts the color image
segmentation produced by applying the range of color values
automatically obtained through the calibration to the example
image in Fig.2. The good segmentation results in Fig.3 indicate
that the prototype values for each color of interest have been
correctly determined during the calibration process.
2) Geometral and Horizon Planes Detection: The next step
consists of the estimation of the geometral and horizon planes
according to the robot head position. In order to do this, firstly,
the pitch and yaw angles that indicate the relative position of
Figure 5. Interest points and goal blobs
the robot head with respect to the play field are calculated. On
the one hand, the geometral plane is defined as the horizontal
In addition, a process to detect interest points is performed.
projection plane where the observer is located. On the other
The same grid mentioned before is utilized to detect crossings
hand, the horizon plane is parallel to the geometral plane and
between blue or yellow lines (belonging to the goal posts)
indicates the level above which there is no useful information.
and white lines in the play field (goal lines). Also, crossings
among green pixels (belonging to the play field) and white
lines that delimit the play field are identified. If those interest
points are detected close to the blue or yellow lines, previously
sampled, they reinforce the belief that those lines belong to the
goal posts. Red circles in Fig.5 (left) enclose interest points
identified in the original image shown in Fig.2.
4) Goal Recognition: Once a set of pixels distributed into
parallel lines corresponding to the goal posts and crossbar have
been identified according to the procedure described in the
Figure 4. Intersection of geometral and horizon plane with image plane previous section, the last step consists of a recognition process
that finally locates the gravity center of the goal.
Thus, the position matrix of the robot head is used for de- In order to perform such task, the aforementioned lines are
termining the horizontal inclination of the image with respect grouped into blobs. A blob is composed of neighbor lines with
to the play field. Then, a grid composed of series of paral- similar aspect ratio (an example is shown in Fig.5 (right)).
lel vertical lines perpendicular to the horizontal inclination Finally, the blobs identified in this way are grouped into a
previously mentioned is calculated. The intersection between perceptual unit that can be considered as a pre-attentive goal.
the grid and the green play field produces a set of points. Then, we apply an intelligent case reasoning strategy to bind
4 X WORKSHOP DE AGENTES FÍSICOS, SEPTIEMBRE 2009, CÁCERES

that unit into a coherent goal. Fig.5 (right) illustrates the blobs exploit different geometric properties and use different image
that configure the goal that appears in Fig.2 after it has been primitives: line segments and points.
recognized by the proposed technique. The geometric center of
the goal is also indicated according to the recognition method. A. Line segments and thorus
Our first algorithm works with line segments. This algorithm
B. Detection based on color, edges and Hough transformation works in the absolute reference system and finds the absolute
We have also developed a second simple method to detect camera position computing some restrictions coming from the
goals in 2D images. It follows four steps in pipeline. First, a pixels where the goal appears in the image.
color filter in HSV color space selects goal pixels and maybe There are three line segments in the goal detected in the
some outliers. Second, an edge filter obtains the goal contour image: two goalposts and the crossbar. Taking into considera-
pixels. Third, a Hough transformation gets the goal segments. tion only one of the posts (for instance GP1 at Fig.2) the way
And fourth, some proximity conditions are checked on the in which it appears in the image imposes some restrictions
vertices of such segments, finding the goal vertices Pix1, Pix2, to the camera location. As we will explain later, a 3D thorus
Pix3 and Pix4. All the steps can be shown at Fig.6. contains all the camera locations from which that goalpost is
seen with that length in pixels (Fig.8). It also includes the two
corresponding goalpost vertices. A new 3D thorus is computed
considering the second goalpost (for instance GP2 at Fig.2),
and a third one considering the crossbar. The real camera
location belongs to the three thorus, so it can be computed
as the intersection of them.
Nevertheless the analytical solution to the intersection of
three 3D thorus is not simple. A numerical algorithm could
be used. Instead of that, we assume that the height of the
camera above the floor is known. The thorus coming from
the crossbar is not needed anymore and it is replaced by a
horizontal plane, at h meters above the ground. Then, the
intersection between three thorus becomes the intersection
between two parallel thorus and a plane. The thorus coming
from the left goalpost becomes a circle in that horizontal plane,
centered at the goalpost intersection with the plane. The thorus
coming from the right goalpost also becomes a circle. The
intersection of both circles gives the camera location. Usually,
Figure 6. Goal detection based on color, edges and Hough transformation
due to simmetry, two different solutions are valid. Only the
position inside the field is selected.
IV. G OAL DETECTION IN 3D To compute the thorus coming from one post, we take its
two vertices in the image. Using projective geometry and the
Once the goal has been properly detected in the image, intrisinc parameters of the camera, a 3D projection ray can be
spatial information can be obtained from the that goal using computed that traverses the focus of the camera and the top
geometric 3D computations. Let Pix1, Pix2, Pix3 and Pix4 vertex pixel. The same can be computed for the bottom vertex.
be the pixels of the goal vertices in the image, which are The angle α between these two rays in 3D is calculated using
calculated with the algorithms of section III. The position and the dot product.
orientation of the goal relative to the camera can be inferred,
that is, the 3D points P1, P2, P3 and P4 corresponding to the
goal vertices. Because the absolute positions of both goals are α
known (AP1,AP2,AP3,AP4) that information can be reversed α
to compute the camera position relative to the goal, and so,
the absolute location of the camera (and the robot) in the field. GOALPOST
α
In order to perform such 3D geometric computation the
robot camera must be calibrated. Its intrinsic parameters are α
required to deal with the projective transformation the camera
does over objects in 3D world when it obtains the image. The CIRCLE

pinhole camera model has been used, with the focal distance,
Figure 7. Circle containing plausible camera positions
optical center and skew as its main parameters. In addition,
two different 3D coordinates are used: the absolute field based
reference system and the system tied to the robot itself, to its
camera. Let’s now consider one post at its absolute coordinates and a
We have developed two different algorithms to estimate vertical plane that contains it. Inside that plane only the points
the 3D location of the perceived goal in the image. They in a given circle see the post segment with an angle α. The
CAÑAS, PUIG, PERDICES, AND GONZÁLEZ: VISUAL GOAL DETECTION FOR THE NAO HUMANOID PLAYER AT THE ROBOCUP 5

thorus is generated rotating such circle around the axis of the

goalpost. Such thorus contains all the camera 3D locations
from which that post is seen with a angle α, regardless its
orientation. In other words, all the camera positions from
which that post is seen with such pixel length.

Figure 9. Projection rays for the four goal corners

of 10cm, starting close to the camera location and up to

the field size, as can be seen at Fig.10. The tuple with the
Figure 8. Thorus containing all plausible camera positions minimum cost is chosen as the right P1, P2 and P3 values.
P4 is directly computed from them. They are the relative 3D
position of the goal in the camera reference system.
B. Points and projection rays
The second algorithm works in the reference system tied
to the camera. It uses three goal vertex pixels Pix1, Pix2 and
Pix3. For Pix1, using the pinhole camera model, a projection
ray R1 can be drawn which traverses the camera focus and
contains all the 3D points which project into such Pix1. R2
and R3 rays are computed in a similar way, as seen in Fig.9.
The problem is to locate the P1, P2 and P3 points into their
corresponding projection rays.
Assuming that we know the position of P1 in R1 then only
a reduced set of points in R2 and R3 are compatible with
the real goal size. Because the distance between P1 and P2 is
known (D12), P2 must be in R2 and the sphere centered at P1
with D12 radius, named S2 (Fig.9). The general intersection
between R2 and S2 yields two candidate points: P2’ and P2”
(there can also be no interesection at all or only one single
point). Following the same development and the distance D13
between P1 and P3, two more candidate points are computed: Figure 10. Cost function for different λ values
P3’ and P3”.
Combining those points we have several candidate tu- Finally, the absolute 3D camera position can be com-
ples (P 1, P 20 , P 30 ), (P 1, P 200 , P 30 ), (P 1, P 20 , P 300 ) and puted from (P 1, P 2, P 3, P 4). Because the absolute posi-
(P 1, P 200 , P 300 ) All of them contain points located at the tions of the goal in the field reference system are known
projection rays and all of them hold the right distance between (AP 1, AP 2, AP 3, AP 4), we can find a rotation and trans-
P1 and the rest of points, but the distance between P2 and lation matrix RT that fits the transformation of P1 into AP1,
P3 may not be correct. Only the real solution provides good P2 into AP2, etc. We have used the algorithm in [1] for that.
distances between all of its points. A cost function can be The estimated translation represents the absolute position of
associated to choose the best solution tuple. We used the error the camera in the fied based reference system.
in distance between P2 and P3, compared to the good distance
D23.
V. E XPERIMENTS
In fact, the P1 position in R1 is not known, so a search
is performed for all the possible P1 values. The algorithm Several experiments have been carried out to validate our
starts placing P1 at λ distance from the camera location. All algorithms, both in simulation and with real images. For simu-
the candidate solution tuples are calculated and their costs lated images we have used Webots (Fig.1) and for real ones a
computed. For each λ the cost of its best tuple is stored. The benchmark of images collected from the Nao’s camera at the
search algorithm explores R1 increasing λ at regular intervals RoboCup2008, placing the robot at different field locations.
6 X WORKSHOP DE AGENTES FÍSICOS, SEPTIEMBRE 2009, CÁCERES

The first set of results presented in this section correspond

to the 2D goal detection strategy presented in Section III-
A. In particular, Fig.11, Fig.12 and Fig.13 display three
examples corresponding to real RoboCup images and the
results produced by the different steps of the proposed method.
Thus, right images in the first row of each figure show the
intersection between the geometral plane and the image plane.
The left image in the second row of each figure displays those
pixels that are part of the goal according to the sampling grid
utilized and, also, the interest points detected. Finally, right
images depict the recognized goal and its gravity center for
each example image. As it can be appreciated, the proposed
strategy is able to recognize goals even in situations involving
certain difficulties, such as when only a small part of a goal
appears in the image, or if the play field is not visible in the
image, or when the goal is seen from a side in the play field.
Figure 13. Goal detected with method described at Section III-A

detection is overlapped as red squares at the goal vertices

(Pix1,Pix2,Pix3,Pix4). In the left side of the figure the field
lines and goals are drawn, and the estimated camera position
is also displayed as a 3D arrow. No accuracy has been
quantitatively measured yet, but the results seem promising
and qualitatively correct.

Figure 11. Goal detected with method described at Section III-A

Figure 14. 3D position from the goal detected in the image

The Nao camera has been calibrated as an ideal pinhole

camera, with its optical center at the middle of the image and
its focal distance inferred from the hardware specifications.
That’s why no further accuracy test has been carried out yet.
All the results presented in this paper have been obtained
Figure 12. Goal detected with method described at Section III-A on a 3 GHz Pentium-IV machine processing offline the real
images. The time consumption corresponding to both 2D and
A second set of results produced by the technique ex- 3D proposed techniques are shown in table I. In particular,
plained in section IV are shown in Fig.14. The input im- times for each of the processing steps to detect the goal in
ages are displayed at the right side. The output of the 2D 3D are shown. The 3D algorithm is fast. The projection rays
CAÑAS, PUIG, PERDICES, AND GONZÁLEZ: VISUAL GOAL DETECTION FOR THE NAO HUMANOID PLAYER AT THE ROBOCUP 7

method is slower than the thorus method, maybe because it is We are working on performing more experiments onboard
a search algorithm. For edge filter and Hough transformation the robot, with its limited computer. We intend to optimize the
we have used OpenCV library. implementation to reach real time performance.
The proposed 3D algorithms assume the complete goal
2D detection based on Geometrical Relations 9ms. appears in the image, but this is not the general case. The
2D detection based on color, edges and Hough transformation 13’2ms. second future line is to expand the 3D geometry algebra to
Thorus based 3D detection 1ms use the field lines and incompletely perceived goals as source
Projective rays 3D detection 2ms.
Table I of information. For instance the corners and field lines convey
T IME CONSUMPTION useful self-localization information too.

R EFERENCES
The algorithm performs well both with the ideal images
[1] A. Lorusso, D.W.Eggert and R.B.Fisher, A Comparison of Four Algo-
coming from the Webots simulator and the real images from rithms for Estimating 3-D Rigid Transformations, Proceedings of the
the Nao at RoboCup2008. In the case of the 2D goal detection 1995 British conference on Machine vision, Vol. 1, pp 237-246, Itech
at III-B the color filter must be properly tuned for each Education and Publishing 1995.
[2] Thomas Rofer et al., B-Human, Team Report and code release 2008
scenario.
[3] S. Lenser and M. Veloso, Visual sonar: fast obstacle avoidance using
monocular vision in Proceedings of the IEEE/RSJ Int. Conf. on Intelligent
VI. C ONCLUSION Robots and Systems, 2003. (IROS 2003), pp 886-891, 2003.
[4] M. Jüngel, A Vision System for RoboCup, diploma Thesis, Institut für
Vision is the most important sensor of the autonomous Informatik, Humboldt-Universität zu Berlin, 2004.
robots competing at the RoboCup. Goal detection is one of [5] J.C. Zagal, J. Ruiz-del-Solar, P. Guerrero and R. Palma, Evolving Visual
the main perceptive habilities required for such autonomous Object Recognition for Legged Robots in RoboCup 2003: Robot Soccer
World Cup VII, LNCS 3020, pp 181-191, 2004.
soccer player. For instance, if the humanoid had the opponent’s [6] D. Herrero and H. Martı́nez, Embedded Behavioral Control of Four-
goal just ahead it should kick the ball towards it. If the goal legged Robots in Robotic Soccer, Pedro Lima editor, pp 203-227, 2007.
in front of the robot was its own goal, then it should turn [7] RoboCup Standard Platform League (Nao) Rule Book.
[8] T. Bandlow et al., Fast Image Segmentation, Object Recognition and
or clear the ball away. In this paper two different algorithms Localization in a RoboCup Scenario in M. Veloso, E. Pagello, and H.
have been proposed to detect the goal in the images coming Kitano (Eds.): RoboCup-99, LNAI, vol. 1856, pp. 174–185, 2000.
from the robot’s camera. One of them based on geometral and [9] R. Cassinis and A. Rizzi, Design Issues for a Robocup Goalkeeper, in
M. Veloso, E. Pagello, and H. Kitano (Eds.): RoboCup-99, LNAI, vol.
horizon planes. The second one using a HSV color filter, an 1856, pp. 254–262, 2000.
edge filter and Hough transformation to detect the post and [10] M. Jamzad, E.C. Esfahani, S.B. Sadjad, Object Detection in Changing
crossbar lines. Environment of Middle Size RoboCup and Some Applications, Proc. IEEE
Int. Symp. On Intelligent Control, Vancouver, Canada, pp. 807-810, 2002.
In addition, two new methods have been described which [11] J. Hoffmann, M. Jungel, and M. Lotzsch, A Vision Based System for
estimate the 3D position of the camera and the goal from the Goal-Directed Obstacle Avoidance used in the RC’03 Obstacle Avoidance
goal perceived inside the image. The first one uses the line Challenge, 8th International Workshop on RoboCup 2004 (Robot World
Cup Soccer Games and Conferences), LNAI, vol. 3276, pp. 418-425,
length of the posts and intersects two thorus to compute the 2005.
absolute 3D camera position in the field. The second one uses [12] P. Loncomilla and J. Ruiz-del-Solar, Robust Object Recognition using
the projection lines from the vertice pixels and searches in the Wide Baseline Matching for RoboCup Applications, in RoboCup 2007:
Robot Soccer World Cup XI, LNCS, vol. 5001, pp. 441-448, 2008.
space of possible 3D locations. This 3D perception is useful [13] U. Kaufmann et al., Visual Robot Detection in RoboCup Using Neural
to the self-localization of the robot into the field. Networks, in D. Nardi et al. (Eds.): RoboCup 2004, LNAI, vol. 3276, pp.
All the algorithms have been implemented as a proof of 262–273, 2005.
[14] S. Volioti and M.G. Lagoudakis, Histogram-Based Visual Object Recog-
concept. Preliminary experiments have been carried out that nition for the 2007 Four-Legged RoboCup League, in Artificial Intelli-
validate them and the results seem promising as shown in gence: Theories, Models and Applications, LNCS, vol. 5138, pp. 313-326,
Section V. 2008.