Page 38 - Krész, Miklós, and Andrej Brodnik (eds.). MATCOS-13. Proceedings of the 2013 Mini-Conference on Applied Theoretical Computer Science. Koper: University of Primorska Press, 2016.
P. 38
e, so the only difference is in the method that was used conclude that the D2 shape function seems well suited for
for each prototype. We can therefore compare the results of this field of Computer Vision.
the two traffic sign recognition programs.
7. REFERENCES
After testing the statistical moments prototype, it quickly
became obvious that it would not be a good fit for traffic [1] Xavier Baro´, Sergio Escalera, Jordi Vitria`, Oriol Pujol,
sign recognition - we tested the program several times, every and Petia Radeva. Traffic Sign Recognition Using
time with different maximum orders of statistical moments Evolutionary Adaboost Detection and Forest-ECOC
(from 5 all the way up to 40). The problem was that the Classification. IEEE Transactions on Intelligent
calculated values of statistical moments did not yield any Transportation Systems, 10(1):113–126, March 2009.
discernible patterns, which is the basis for object recogni-
tion. The prototype thus behaved unpredictably and often [2] Boguslaw Cyganek. Circular road signs recognition
recognized a false symbol. After using the same program with soft classifiers. Integrated Computer-Aided
framework to develop a prototype using the D2 shape func- Engineering, 14(4):323–343, 2007.
tion and getting promising results, we concluded that the
fault was not in the way we developed the program or in [3] Arturo de la Escalera, Jos´e M. Armingol, and Mario
the set of acquired images, but in the method used. We Mata. Traffic sign recognition and analysis for
therefore abandoned statistical moments and moved on to intelligent vehicles. Image and Vision Computing,
search for more suitable methods, until we found the D2 21(3):247–258, March 2003.
shape function.
[4] Arturo de la Escalera, Luis E. Moreno, Miguel A.
While statistical moments have a wide range of use, we found Salichs, and Jos´e M. Armingol. Road Traffic Sign
the D2 shape function to be much more suitable for traffic Detection and Classification. IEEE Transactions on
sign recognition. Industrial Electronics, 44(6):848–859, December 1997.
6. CONCLUSION [5] Chiung-Yao Fang, Chiou-Shann Fuh, Pei-Shan Yen,
Shen Cherng, and Sei-Wang Chen. Road traffic sign
Shape functions up until now were not used in traffic sign detection and classification. Computer Vision and
recognition - we presented a prototype using the D2 shape Image Understanding, 96(2):237–268, November 2004.
function that recognizes traffic signs. It should be noted
that while the testing set, presented in Section 4, is a col- [6] Xiaohong W. Gao, Lubov N. Podladchikova,
lection of traffic sign images that are as varied as possible, Dmitry G. Shaposhnikov, Keum-Shik Hong, and
therefore maximizing the robustness of the traffic sign recog- Natalia A. Shevtsova. Recognition of traffic signs
nition program, the set is still quite small - only 46 images. based on their colour and shape features extracted
The problem is that there is no publicly available deposi- using human vision models. Journal of Visual
tory or collection of Slovenian traffic sign images and the Communication and Image Representation,
set therefore had to be acquired by hand, which, due to the 17(4):675–685, August 2006.
fact that we want to capture as many different traffic signs
as possible, is very time-consuming. That is why we focused [7] Saturnino Maldonado-Basco´n, Sergio Lafuente-Arroyo,
only on speed limit traffic signs, but we could easily expand Pedro Gil-Jim´enez, and Hilario Go´mez-Moreno.
the scope to all traffic signs as the D2 shape function is not Road-Sign Detection and Recognition Based on
limited to any number of components. To add a new type Support Vector Machines. IEEE Transactions on
of traffic sign, all we would need to do is to add the corre- Intelligent Transportation Systems, 8(2):264–278, June
sponding traffic sign symbol to the learning set. 2007.
As described in Section 4, the final performance of the pro- [8] Robert Osada, Thomas Funkhouser, Bernard
gram is 91.3% accurately recognized speed limit traffic signs, Chazelle, and David P. Dobkin. Matching 3D models
which is an encouraging result and an initiative for further with shape distributions. SMI 2001 Proceedings of the
use of the D2 shape function in the field of traffic sign recog- International Conference on Shape Modeling and
nition and possibly other fields too. It is a robust procedure Applications, pages 154–166, May 2001.
which also works on very small and low quality images - the
smallest image used for testing had a resolution of 38 × 31 [9] Pavel Pacl´ık, Jana Novoviˇcova´, Pavel Pudil, and Petr
pixels and the program correctly recognized it. It is worth Somol. Road sign classification using Laplace kernel
noting that this method for recognition is much simpler than classifier. Pattern Recognition Letters,
traditional methods, in particular comparing to solutions, 21(13-14):1165–1173, December 2000.
mentioned in the Introduction, where methods like neural
networks and support vector machines were used. In terms [10] Giulia Piccioli, Enrico De Micheli, Pietro Parodi, and
of performance we can not directly compare it to these solu- Marco Campani. Robust method for road sign
tions since we do not have a testing set that would be large detection and recognition. Image and Vision
enough, but we did get a sense that it is a robust method. Computing, 14(3):209–223, April 1996.
As a final note, the D2 shape function performs much bet- [11] Andrzej Ruta, Yongmin Li, and Xiaohui Liu.
ter than statistical moments for traffic sign recognition, as Real-time traffic sign recognition from video by
is described in Section 5. Considering our results we can class-specific discriminative features. Pattern
Recognition, 43(1):416–430, January 2010.
[12] Walter Wohlkinger and Markus Vincze. Analysis and
Evaluation of Shape Functions for Object Class
Recognition. Proceedings of the 17th Computer Vision
Winter Workshop, February 2012.
m a t c o s -1 3 Proceedings of the 2013 Mini-Conference on Applied Theoretical Computer Science 38
Koper, Slovenia, 10-11 October
for each prototype. We can therefore compare the results of this field of Computer Vision.
the two traffic sign recognition programs.
7. REFERENCES
After testing the statistical moments prototype, it quickly
became obvious that it would not be a good fit for traffic [1] Xavier Baro´, Sergio Escalera, Jordi Vitria`, Oriol Pujol,
sign recognition - we tested the program several times, every and Petia Radeva. Traffic Sign Recognition Using
time with different maximum orders of statistical moments Evolutionary Adaboost Detection and Forest-ECOC
(from 5 all the way up to 40). The problem was that the Classification. IEEE Transactions on Intelligent
calculated values of statistical moments did not yield any Transportation Systems, 10(1):113–126, March 2009.
discernible patterns, which is the basis for object recogni-
tion. The prototype thus behaved unpredictably and often [2] Boguslaw Cyganek. Circular road signs recognition
recognized a false symbol. After using the same program with soft classifiers. Integrated Computer-Aided
framework to develop a prototype using the D2 shape func- Engineering, 14(4):323–343, 2007.
tion and getting promising results, we concluded that the
fault was not in the way we developed the program or in [3] Arturo de la Escalera, Jos´e M. Armingol, and Mario
the set of acquired images, but in the method used. We Mata. Traffic sign recognition and analysis for
therefore abandoned statistical moments and moved on to intelligent vehicles. Image and Vision Computing,
search for more suitable methods, until we found the D2 21(3):247–258, March 2003.
shape function.
[4] Arturo de la Escalera, Luis E. Moreno, Miguel A.
While statistical moments have a wide range of use, we found Salichs, and Jos´e M. Armingol. Road Traffic Sign
the D2 shape function to be much more suitable for traffic Detection and Classification. IEEE Transactions on
sign recognition. Industrial Electronics, 44(6):848–859, December 1997.
6. CONCLUSION [5] Chiung-Yao Fang, Chiou-Shann Fuh, Pei-Shan Yen,
Shen Cherng, and Sei-Wang Chen. Road traffic sign
Shape functions up until now were not used in traffic sign detection and classification. Computer Vision and
recognition - we presented a prototype using the D2 shape Image Understanding, 96(2):237–268, November 2004.
function that recognizes traffic signs. It should be noted
that while the testing set, presented in Section 4, is a col- [6] Xiaohong W. Gao, Lubov N. Podladchikova,
lection of traffic sign images that are as varied as possible, Dmitry G. Shaposhnikov, Keum-Shik Hong, and
therefore maximizing the robustness of the traffic sign recog- Natalia A. Shevtsova. Recognition of traffic signs
nition program, the set is still quite small - only 46 images. based on their colour and shape features extracted
The problem is that there is no publicly available deposi- using human vision models. Journal of Visual
tory or collection of Slovenian traffic sign images and the Communication and Image Representation,
set therefore had to be acquired by hand, which, due to the 17(4):675–685, August 2006.
fact that we want to capture as many different traffic signs
as possible, is very time-consuming. That is why we focused [7] Saturnino Maldonado-Basco´n, Sergio Lafuente-Arroyo,
only on speed limit traffic signs, but we could easily expand Pedro Gil-Jim´enez, and Hilario Go´mez-Moreno.
the scope to all traffic signs as the D2 shape function is not Road-Sign Detection and Recognition Based on
limited to any number of components. To add a new type Support Vector Machines. IEEE Transactions on
of traffic sign, all we would need to do is to add the corre- Intelligent Transportation Systems, 8(2):264–278, June
sponding traffic sign symbol to the learning set. 2007.
As described in Section 4, the final performance of the pro- [8] Robert Osada, Thomas Funkhouser, Bernard
gram is 91.3% accurately recognized speed limit traffic signs, Chazelle, and David P. Dobkin. Matching 3D models
which is an encouraging result and an initiative for further with shape distributions. SMI 2001 Proceedings of the
use of the D2 shape function in the field of traffic sign recog- International Conference on Shape Modeling and
nition and possibly other fields too. It is a robust procedure Applications, pages 154–166, May 2001.
which also works on very small and low quality images - the
smallest image used for testing had a resolution of 38 × 31 [9] Pavel Pacl´ık, Jana Novoviˇcova´, Pavel Pudil, and Petr
pixels and the program correctly recognized it. It is worth Somol. Road sign classification using Laplace kernel
noting that this method for recognition is much simpler than classifier. Pattern Recognition Letters,
traditional methods, in particular comparing to solutions, 21(13-14):1165–1173, December 2000.
mentioned in the Introduction, where methods like neural
networks and support vector machines were used. In terms [10] Giulia Piccioli, Enrico De Micheli, Pietro Parodi, and
of performance we can not directly compare it to these solu- Marco Campani. Robust method for road sign
tions since we do not have a testing set that would be large detection and recognition. Image and Vision
enough, but we did get a sense that it is a robust method. Computing, 14(3):209–223, April 1996.
As a final note, the D2 shape function performs much bet- [11] Andrzej Ruta, Yongmin Li, and Xiaohui Liu.
ter than statistical moments for traffic sign recognition, as Real-time traffic sign recognition from video by
is described in Section 5. Considering our results we can class-specific discriminative features. Pattern
Recognition, 43(1):416–430, January 2010.
[12] Walter Wohlkinger and Markus Vincze. Analysis and
Evaluation of Shape Functions for Object Class
Recognition. Proceedings of the 17th Computer Vision
Winter Workshop, February 2012.
m a t c o s -1 3 Proceedings of the 2013 Mini-Conference on Applied Theoretical Computer Science 38
Koper, Slovenia, 10-11 October
