Page 35 - 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. 35
ffic Sign Symbol Recognition with the D2 Shape
Function
Simon Mezgec and Peter Rogelj
Faculty of Mathematics, Natural Sciences and Information Technologies
University of Primorska
Koper, Slovenia
simon.mezgec@student.upr.si, peter.rogelj@upr.si
ABSTRACT Piccioli presented a robust method for traffic sign detec-
tion and recognition [10]. One year later, de la Escalera
This paper describes our application of a novel method in developed an algorithm that focuses on the recognition of
the field of traffic sign recognition - the D2 shape function. traffic signs - the algorithm uses neural networks for the
We first give an overview of the advances and research in classification of traffic signs [4]. De la Escalera built on
this field. We then describe the D2 shape function that was that work in his 2003 article where he used a genetic algo-
originally used to classify 3D models of various objects - rithm for the detection step, which allowed further invari-
because of its robustness we propose its use in the field of ance [3]. In 2000, Pacl´ık proposed his own algorithm for traf-
traffic sign recognition. We describe how the program that fic sign classification using the Laplace probability method
was developed using this method works and present the re- and an Expectation-Maximization algorithm to maximize
sults on a testing set of self-acquired speed limit traffic signs, the likelihood function [9]. In 2004, Fang developed a traf-
evaluate its performance and compare it to the performance fic sign detection and recognition system that is based on
of statistical moments. a computational model of human visual recognition pro-
cessing [5]. Two years later, Gao improved upon the hu-
General Terms man vision model recognition using separate models for ex-
tracting color and shape information, respectively [6]. In
Algorithms 2007, Maldonado-Basco´n described a traffic sign detection
and recognition system that is based on support vector ma-
Keywords chines which is invariant even to partial occlusions [7]. In the
same year, Cyganek also used support vector machines in his
Symbol recognition, traffic signs, D2 shape function system for traffic sign detection, whereas the recognition is
performed using neural networks [2]. Finally, in 2009, Baro´
1. INTRODUCTION presented a system that performs traffic sign detection with
the boosted detectors cascade and traffic sign recognition
While driving a vehicle, the driver has to be constantly aware with a forest of optimal tree structures that are embedded
of a number of factors, one of which are traffic signs. These in the ECOC (Error-Correcting Output Code) matrix [1],
signs can have specific properties, such as color and shape, whereas in the next year, Ruta added a tracking procedure
which allow their recognition with the use of Computer Vi- between the steps of traffic sign detection and recognition -
sion. So it is no surprise that this is one of the important this tracker predicts the position and scale of the sign can-
steps towards the automatization of driving vehicles. Once didate to reduce computation. That system uses Color Dis-
the traffic sign is detected, we would usually like to know tance Transform for the classification of traffic signs [11].
what kind of symbol (if any) the traffic sign contains. Recog-
nition is useful for many purposes: the driver could forget In this paper we present an alternative method for traf-
what the last speed limit he drove by was, an inexperienced fic sign recognition, based on the D2 shape function that
driver could see a traffic sign for the first time and he would promises high robustness required for such applications. The
like to know its meaning etc. It is therefore apparent that method is presented in the next Section and its practical ap-
traffic sign recognition has a real-world use. plication in Section 3.
There have been many different methods proposed in the 2. METHOD DESCRIPTION
field of traffic sign detection and recognition in the last two
decades. Let us mention some implementations that at- For the recognition of traffic sign symbols we use the D2
tracted the most attention from the community. In 1996, shape function - a method first described by Osada in his
2001 article [8] - it is therefore a fairly new Computer Vision
method. The function samples random pairs of points and
creates a histogram of distances between these point-pairs.
Figure 1 shows how the D2 shape function creates the dis-
tance histogram between pairs of points on a cup [12].
The idea behind the function is that with a large enough
m a t c o s -1 3 Proceedings of the 2013 Mini-Conference on Applied Theoretical Computer Science 35
Koper, Slovenia, 10-11 October
Function
Simon Mezgec and Peter Rogelj
Faculty of Mathematics, Natural Sciences and Information Technologies
University of Primorska
Koper, Slovenia
simon.mezgec@student.upr.si, peter.rogelj@upr.si
ABSTRACT Piccioli presented a robust method for traffic sign detec-
tion and recognition [10]. One year later, de la Escalera
This paper describes our application of a novel method in developed an algorithm that focuses on the recognition of
the field of traffic sign recognition - the D2 shape function. traffic signs - the algorithm uses neural networks for the
We first give an overview of the advances and research in classification of traffic signs [4]. De la Escalera built on
this field. We then describe the D2 shape function that was that work in his 2003 article where he used a genetic algo-
originally used to classify 3D models of various objects - rithm for the detection step, which allowed further invari-
because of its robustness we propose its use in the field of ance [3]. In 2000, Pacl´ık proposed his own algorithm for traf-
traffic sign recognition. We describe how the program that fic sign classification using the Laplace probability method
was developed using this method works and present the re- and an Expectation-Maximization algorithm to maximize
sults on a testing set of self-acquired speed limit traffic signs, the likelihood function [9]. In 2004, Fang developed a traf-
evaluate its performance and compare it to the performance fic sign detection and recognition system that is based on
of statistical moments. a computational model of human visual recognition pro-
cessing [5]. Two years later, Gao improved upon the hu-
General Terms man vision model recognition using separate models for ex-
tracting color and shape information, respectively [6]. In
Algorithms 2007, Maldonado-Basco´n described a traffic sign detection
and recognition system that is based on support vector ma-
Keywords chines which is invariant even to partial occlusions [7]. In the
same year, Cyganek also used support vector machines in his
Symbol recognition, traffic signs, D2 shape function system for traffic sign detection, whereas the recognition is
performed using neural networks [2]. Finally, in 2009, Baro´
1. INTRODUCTION presented a system that performs traffic sign detection with
the boosted detectors cascade and traffic sign recognition
While driving a vehicle, the driver has to be constantly aware with a forest of optimal tree structures that are embedded
of a number of factors, one of which are traffic signs. These in the ECOC (Error-Correcting Output Code) matrix [1],
signs can have specific properties, such as color and shape, whereas in the next year, Ruta added a tracking procedure
which allow their recognition with the use of Computer Vi- between the steps of traffic sign detection and recognition -
sion. So it is no surprise that this is one of the important this tracker predicts the position and scale of the sign can-
steps towards the automatization of driving vehicles. Once didate to reduce computation. That system uses Color Dis-
the traffic sign is detected, we would usually like to know tance Transform for the classification of traffic signs [11].
what kind of symbol (if any) the traffic sign contains. Recog-
nition is useful for many purposes: the driver could forget In this paper we present an alternative method for traf-
what the last speed limit he drove by was, an inexperienced fic sign recognition, based on the D2 shape function that
driver could see a traffic sign for the first time and he would promises high robustness required for such applications. The
like to know its meaning etc. It is therefore apparent that method is presented in the next Section and its practical ap-
traffic sign recognition has a real-world use. plication in Section 3.
There have been many different methods proposed in the 2. METHOD DESCRIPTION
field of traffic sign detection and recognition in the last two
decades. Let us mention some implementations that at- For the recognition of traffic sign symbols we use the D2
tracted the most attention from the community. In 1996, shape function - a method first described by Osada in his
2001 article [8] - it is therefore a fairly new Computer Vision
method. The function samples random pairs of points and
creates a histogram of distances between these point-pairs.
Figure 1 shows how the D2 shape function creates the dis-
tance histogram between pairs of points on a cup [12].
The idea behind the function is that with a large enough
m a t c o s -1 3 Proceedings of the 2013 Mini-Conference on Applied Theoretical Computer Science 35
Koper, Slovenia, 10-11 October
