DOWNLOAD PDF. Report this file. Description. Download Fabio Marchesi-La Fisica Dell'Anima Free in pdf format. Sponsored Ads. Account Login. Coralligenous: Insight for a New Geomorphological Definition. Fabio Marchese. Valentina Bracchi. Alessandra Savini. cesare corselli. Daniela Basso. Monnalisa . Omni-directional vision with a multi-part mirror Fabio M. Marchese, Domenico G. Sorrenti Dipartimento di Informatica, Sistemistica e Comunicazione Università.
|Language:||English, Spanish, French|
|Genre:||Science & Research|
|Distribution:||Free* [*Register to download]|
Téléchargement Gratuit Fabio Marchesi Video livre ou tout simplement lire en ligne Fabio Marchesi Video livre en format pdf et epub. Let's not pretend that things will change if we keep doing the same things. A crisis can be a real blessing to any person, to any nation. For all crises bring. Con Fabio Marchesi e Marco Poggianella. Con il supporto di altri Entusiasmologi certificati. (indicativamente uno/una ogni 12 iscritti oltre ai 25 (max 60)).
If a robot knows its posture, on the image window B in Fig. The maximum value of red win- the goal with the ball in between. It can also know dow C in Fig.
The red color its team-mate postures and prepare a pass, or evaluate is the easiest to track and the one with least inter- the game state from the team locations . Due to its motion, the ball can be seen any- gorithm based on the isometric part of the multi-part where on the image, and so window C can be lo- mirror of the catadioptric vision system described in cated.
Section 2. The robot should not activate the middle-size league robot, with respect to a given co- kicking device when the ball is not ready to be ordinate system, from the observation of natural land- kicked, to save energy and avoid hurting its op- marks of the soccer field, such as the field lines and ponents. Therefore, the kicking device is activated goals, as well as its correlation, in the Hough transform only when the ball red cross represented by let- space, with a geometric field model.
Even though the ter C in Fig. This also means that the robot rently used, the wall replacement by the correspond- will kick the ball only when the ball is touching the ing field lines would not change the algorithm. The robot. Method description objects in the correct image windows.
Therefore, for easy calibration of the catadioptric system, the Even though the self-localization algorithm was de- camera lens must be placed inside the square given signed motivated by its application to robotic soccer, by letter F in Fig.
Use the relevant feature from the previous step of those environments. An important requirement is to proceed. The idea is to correlate a number of ac- are the following: tual straight lines, found in the image, sharing the same descriptive parameter e.
Correlate 4. Pick up the most relevant pair of straight lines i. Geometric field model 5 and 6, are plotted in Fig. Excluded lines were chosen because they image of tan 0. All the in steps of 1 pixel, corresponding to an actual field distances between lines are known from RoboCup resolution of 6. Changes in the dimensions are parameterized referred above results from the absence of information in a table.
The model reference frame is located at on which field lines lead to the most relevant pair.
This the bottom left of the model image. Orientation determination 4. Position determination Steps 1—6 of the algorithm described in Section 4. This is done together with the dis- solved later. The max- imum of the four correlation maxima occurs for the array pair representing the best match between image and actual field lines.
A companion array pair exists for each best pair. The two pairs uniquely identify two approximately orthogonal field lines, by checking the array positions where the maximum occurred vertical field lines are numbered 1,.
The intersection of the two lines is a reference point, whose coordinates are known in the world reference frame, from the field model. The explanation above is summarized in the fol- where the subscripts i, m and f stand for the image, lowing table the best and companion pairs positions field model and actual field reference frames, and the can be exchanged superscripts ref and r stand for the reference point and the robot, respectively.
First, the algorithm checks whether the robot criterion , followed by a translation that expresses the position is not outside the field. The second test con- center of the image i. The world reference frame is located in one of the goals blue or yellow in the actual field the middle of the soccer field, with the x-axis point- and applying to each of those points of the coordinates g g ing towards the blue goal and the y-axis is such that a xf , yf the inverse transform of 7.
Should they center of the robot front. This transformation can be estimated must be used for the robot position. The method was applied to a set of P. Position error histogram. The images were taken 9 cm in y. The results from the 90 well known e. In Fig. The lines represented are the possible lines of justed Gaussian function. The rectangle on the plot the field. Note that, in this test, Fig. Test image results. Yagi, S.
Kawato, S. Tsuji, Real-time omni-directional harder the posture determination process, because the image sensor copis for vision-guided navigation, IEEE other wall is not seen, and a relevant parallel line can- Transactions on Robotics and Automation 10 1 11— Mouaddib, C.
Conclusions  J. Suzuki, T. Katoh, M.
Asada, An application of vision-based This paper has shown the potential of omni- learning for a real robot in robocup — Learning of goal directional catadioptric systems for comprehensive keeping behavior for a mobile robot with omni-directional solutions for mobile robots moving within structured vision and embedded servoing, in: M.
Asada, H. Kitano Eds. Moreover, this  D. Nardi, G. Clemente, E. Bonarini, P. Aliverti, M. Lucioni, An omni-directional sensor for fast tracking for mobile robots, IEEE Transactions Further steps towards a more refined usage of the on Instrumentation and Measurements 49 3 — information provided by omni-directional vision sys- Bonarini, The body, the mind or the eye, first?
Veloso, E. Pagello, H. Borenstein, H. Everett, L.
Feng, Where am I? Enderle, M. Ritter, D. Fox, S. Sablatnog, G. Kraetzschmar, relevant objects e. Palm, Vision-based localization in RoboCup environments, in: P. Stone, T. Balch, G. Kraetzschmar Eds. Fox, W. The requirements the device must fulfill result from its use as the main perception system for the autonomous mobile robots used in F RoboCup competitions. The more relevant requirements which have been pursued are 1 range sensing in a quite wide region centered around the robot, with good accuracy; 2 sensing around the robot in a given vertical sector, in order to recognize team-mates and adversaries all robots have a colored marker above a given height ; 3 range sensing in a region very close around the robot, with the highest accuracy, to locate and kick the ball.
Such requirements have been fulfilled by the design of a mirror built up of three different parts. Each part is devoted to the fulfillment of one requirement.
This approach resulted to be similar to a previous work by Hicks and Bajcsy, although independently developed by the authors. Introduction This work has been accomplished in the framework of the Italian participation to RoboCup . For a more detailed introduction to RoboCup see . A robot capable to compete in a F RoboCup match F is the so called "middle-size" league, i. For the above-mentioned aims, an omni- directional vision sensor  seems appropriate Fig.
An omni-directional vision sensor should satisfy the following constraints: 1. Conic mirror Pin-hole Sensor plane Fig. Conic mirror Fig. Conic-spherical camera plus mirror configuration mirror In literature different mirror geometries have been proposed and even in RoboCup some teams already used mirrors other than the original conic one Fig. For instance in , a conic mirror with a "spherical vertex" Fig. Such mirrors introduce large distortions on image distances of objects at the playground level.
It should be noted that such distortion grows with the distance of the object from the observer Fig. On one hand, it is quite obvious that the "nominal" value of the estimate can be easily corrected exploiting the known profile of the mirror; on the other hand, the accuracy of the measure is corrupted without the possibility to compensate for such degradation. Therefore, one of the objectives of this work was to develop an optical compensation of the above-described distortion, working directly on the mirror profile in such a way that the absolute localization error remains constant with the object distance.
Both the idea of an optical compensation and the analytical set up for the determination of the compensating mirror profile turned out to be indistinguishable from a previous work, by Hicks and Bajcsy .
Such work does not detail some implementation aspects, which are presented here instead. Moreover, it should be mentioned that the approach here taken, which aims at the defintion of a complete solution to a set of real requirements, involves more than the optical compensation just described, which provides a solution to one requirement only. Such global approach implies the integration of the different proposed solutions to each requirement.
Due to the fact that each solution is the design of a part of a mirror, the overall outcome of this work is one single mirror, built up of different parts. Conic mirror: a the image dimension of a given object vs. Isometric mirror: a the image dimension of a given object vs. This allows for a better accuracy in locating objects, with respect to what can be attained with commonly used mirrors compare Fig. With a conventional e. It should now be clear that the effect of the distortion due to conventional mirrors represent a degradation.
This is true not only under the point of view of the localization accuracy, but also under the point of view of the detection of relevant features e.
The smaller the ideal image size, the higher the probability of a detection failure of the feature. A failure can happen when the image formation will not take place under ideal conditions. Non-ideal conditions are due to shadows, non- uniform lightening, electronic noise in the hardware, etc. Mirror design In this section the procedures followed to design the proposed mirror are introduced.
To completely satisfy the requirements, the mirror design has been split in three parts: isometric, constant curvature and planar. Sketch for inferring the differential equation generating the isometric part of the mirror Isometric mirror part The first requirement implies the design of a mirror capable to compensate the distortion, introduced by the linear profile of a conic mirror, by means of a non- constant curvature of the profile.
This is the idea that turned out to be the same as the one proposed in . This design problem has been modeled by the following differential equation 1 , which can be inferred by applying the laws of the Linear Optics Fig. This equation has been written in a reference frame XY centered in the Pin-hole of the camera.
Refer to  to find a different formulation obtained under similar conditions. Although we independently obtained this formulation, the approach is quite similar to the one taken in  and the results are identical. Differently from , we developed a "geometrically" based integration of equation 1. Our approach bases on a local first order approximation of the profile: at each point the profile has been approximate by its tangent.
The higher the considered number of points on the profile, the better the approximation. Referring to Fig. Hence the choice is to have the circular crown at the same height of the last point of the constant curvature part of the mirror point B in Fig. The radial dimension point C is set as follows: The area observed by this part of the mirror at the playground level zone C is observed by the isometric part of the mirror too.
The image produced by the circular crown, on the other hand, is much larger Fig. This configuration produces a discontinuity in the image, which is not a problem.
The resulting mirror The mirror profile resulting from the design described before is shown in Fig. The mirror Fig. Its outer part allows the observation of objects from 0. The magenta cylinder is the marker. By comparing the third part of the mirror with the first, one can see that the third part allows an easier detection and localization of the ball. In Fig. The effect of the optical compensation lasts up to 6 m, but, thanks to the continuity with the second part, it is still possible to detect the yellow goal and the other robot marker, although they are distorted by the constant curvature part of the mirror.
The distorsion is due to the fact that the body of the objects, i. However, their distances from the observer, measured at the contact point with the playground, can still be recovered with the limited error provided by the isometric design. Profile of the overall mirror Fig.
The proposed mirror Marker Marker Fig. Image taken when the robot is in the Fig. Image taken when the robot is in center of the playground front of the blue goal Conclusions The paper stems from the definition of some requirements for the sensor system of a robot for F RoboCup competitions. The proposed solution is based on the design of a mirror of an omni-directional system, differently from approaches trying to compensate in software the shortcomings of conventional mirror design.
The proposed design resulted in a three-part mirror, each part being devoted to fulfill one of the requisites. The first part aims at an un-warped image of the playground; this development resulted nearly identical to a precedent work . Details on the integration of the differential equation governing this part of the design are presented. The third part allows to precisely localize the ball when very close to the robot. At the moment of writing the mirror is under construction and we hope to be able to install the mirror on the Italian National Representative robots in RoboCup F championships.
Acknowledgements This work exists thanks to the contribution of Prof. His contribution has been fundamental for our effective introduction to the field of RoboCup competitions. Without his experience the setting up of the requirements would have not been possible.
Moreover, we are grateful for the very fruitful conversations, which took place during the development of the work. References Y. Yagi, S. Kawato, S. Bonarini, P. Aliverti, M. Lucioni, "An omni-directional sensor for fast tracking for mobile robots", Proc. Hicks, R. Nardi, G. Clemente, E.
Asada ed. Kitano, M. Asada, Y. Kuniyoshi, I. Noda, E. Osawa, H. Matsubara, "RoboCup: Suzuki, T. Katoh, M.
Asada, "An application of vision-based learning for a real robot in RoboCup learning of goal keeping behavior for a mobile robot with omni-directional vision and embedded servoing", in Robocup