A Road Trafc Signal Recognition System Based on Template Matching Employing Tree Classier

Embed Size (px)

Citation preview

  • 7/30/2019 A Road Trafc Signal Recognition System Based on Template Matching Employing Tree Classier

    1/6

    A Road Traffic Signal Recognition System based on Template

    matching employing Tree classifier

    Varun S, Surendra Singh, Sanjeev Kunte R, Sudhaker Samuel R D and Bindu PhilipDepartment of Electronics & Communication

    S J College of Engineering

    Mysore - 570006, Karnataka, India

    Email: [email protected], [email protected]

    Abstract

    The traffic sign detection and recognition system is an essential module of the driver warning

    and assistance system. Smart vehicles are the order of the day. Such vehicles have the ability

    to warn drivers of pending situations, remind them of speed limits and even automatically take

    evasive action. Due to the visual nature of existing infrastructure, signs and line markings, image

    processing will play a large part in these systems. In this paper we present an automated Traffic

    sign recognition system allowing an invariance localization to changes in position, scale, rotation,

    weather conditions, partial occlusion, and the presence of other objects of the same color. The

    reliability demonstrated by the proposed method suggests that this system could be a part of an

    integrated driver warning and assistance system based on computer vision technology.

    1. Introduction

    Traffic signs provide regulatory, warning, or guidance information to drivers. They are designed

    to be easily recognized by human drivers mainly because their color and shapes differ from those

    present in natural environments. As such, effective and timely conveyance of information from

    traffic signs to drivers can definitely decrease the number of traffic accidents. The information

    conveyance depends in part on the legibility and recognition of the traffic sign content, which is

    a function of traffic sign and observer characteristics. In autonomous vehicles, they should thus

    provide the control system with information similar to that offered to human drivers. Traffic Sign

    Recognition (TSR) Systems are today being incorporated into the Autonomous Intelligent Vehicles

    of the future, which will take crucial decisions regarding speed, trajectory of motion, etc. Such

    vehicles will incorporate vision-based vehicle guidance systems that will perform road detection,

    obstacle detection and sign recognition.

    Some of the early works on traffic sign recognition system are: An automatic target recognition

    system developed by Olson and Huttenlocher in [1] uses a hierarchical search tree. Work by Lu et

    al. [2] is an example of neural network techniques for traffic sign recognition. Piccioli et al. [3]

    used black and white images. After the extraction of the edges, there is a shape analysis looking

    for circular and triangular contours. Besserer et al. [4] created a pyramidal structure from the

    original image. An edge detector is applied to every image of the pyramid. The edges of an image

    joined the edges of the upper image. By analyzing the generated contours, the signs are classified

    into triangular, circular, or rectangular signs. De La Escalera et al. in [5] have proposed system

    for European warning signs (equilateral triangles with one vertex upward) and circular prohibition

    International Conference on Computational Intelligence and Multimedia Applications 2007

    0-7695-3050-8/07 $25.00 2007 IEEE

    DOI 10.1109/ICCIMA.2007.190

    360

    International Conference on Computational Intelligence and Multimedia Applications 2007

    0-7695-3050-8/07 $25.00 2007 IEEE

    DOI 10.1109/ICCIMA.2007.190

    360

  • 7/30/2019 A Road Trafc Signal Recognition System Based on Template Matching Employing Tree Classier

    2/6

    signs. The system has different stages for color segmentation, corner detection, shape recognition,

    sign classification. The paper proposes a TSR system developed in two phases: Detection phase and

    Recognition phase. In the former, the color thresholding in the RGB color space is used to segment

    the image. The features of traffic signs are investigated and used to detect potential objects. The

    later phase will be trained to perform the classification and validation. The joint use of classification

    and validation modules can reduce the false recognition rates.

    The rest of the paper is organized as follows. Section 2 gives a brief overview of the proposed

    TSR system. Different modules used in TSR system are described in detail in section 3. Experi-

    mental results and conclusion are discussed in section 4 and 5.

    2. Overview of the TSR system

    After acquiring image of the object through a camera, the image is examined to determine

    whether a traffic sign is present in it or not in the first phase. In the next phase, the detected traffic

    sign is matched with three templates corresponding to circle, circle-bar and triangular respectively.

    This is based on the fact that all the mandatory and cautionary signs used in India are characterized

    by equilateral triangles or circular shapes with or without a cross. These three templates form our

    basic templates. The remaining traffic signs are derived from these templates. Hence they are re-

    ferred as derived templates. For example, all the speed signs are derived templates which belong to

    the circular basic template. Therefore the basic template category of the traffic sign is determined.

    In the last phase, the detected traffic sign is matched against the derived templates belonging to the

    same basic template class.

    Figure 1. Three-phase decision tree used for traffic sign recognition.

    3. Modules of TSR system

    The TSR system gives an accurate output to the user through the interaction of five modules

    namely, (a) Pictures module (b) Objects module (c) Database module (d) Classifier module (e) GUI

    module. The interaction between these modules is illustrated in Figure 2.

    3.1. Picture module

    This module deals with the following: (a) Image Capturing (b) Preprocessing the image (c)

    Segmentation of the image on the basis of color information (d) Retrieval of all the objects present

    in the current image.

    361361

  • 7/30/2019 A Road Trafc Signal Recognition System Based on Template Matching Employing Tree Classier

    3/6

    Figure 2. Interaction between the various modules of the TSR system.

    In this module, the image is first preprocessed to remove the noise. The R, G and B values for

    each pixel are then determined. The sum of green and blue pixel components is then compared

    with 1.5 times the red pixel component for each pixel. Pixels corresponding to relatively higher red

    component values constitute the feature pixels and are an indication of the objects of interest. A

    binary segmented image is then created using the known coordinates of the feature pixels. Figures

    3(a) and Figure 3(b) show the original image and its segmented version respectively.

    Figure 3. Example of a binary segmented image (a)Original Image (b)Segmented

    Image.

    3.2. Objects module

    This module deals with the following: (a) Detection of the various objects present in the image

    (b) Determines whether the objects are valid traffic signs or not (c) Determines the basic template

    class to which the detected traffic sign belongs to.

    In this module, all the 8-pixel connected objects are considered and labeled with a unique tag.

    If the number of detected objects is greater than 50, then it is an indication to the presence of

    unnecessary objects. Such objects are removed through a combination of morphological opening

    and closing processes [6].

    For each of the detected objects, the following tasks are performed:

    The object is first enclosed in a virtual quadrilateral.

    This quadrilateral is then examined to determine whether it is an approximate square with

    40X40 or greater dimensions. These conditions are the preliminary requirements that need

    to be satisfied to enable efficient extraction of features.

    362362

  • 7/30/2019 A Road Trafc Signal Recognition System Based on Template Matching Employing Tree Classier

    4/6

  • 7/30/2019 A Road Trafc Signal Recognition System Based on Template Matching Employing Tree Classier

    5/6

    normalization, the computed values are binned into ten containers. The divergence of the detected

    Traffic Sign is computed with respect to the previously computed normalized city block distances

    corresponding to the basic template it has been categorized under. If the divergence is below a

    certain threshold, the detected traffic sign is considered valid. Table 1 illustrates the previously

    computed binned city block distances corresponding to the three basic templates, namely circle-bar

    (CB), circle(C) and triangle (T).Table 1. Normalized City Block Distances Corresponding to the Basic Templates

    The holes in the feature image are then filled thereby creating a temporary mask. The mask

    obtained is used along with the original feature image to modify the segment of the original image

    containing the traffic sign such that the traffic sign features are superimposed on a white back-

    ground. This is called the Data Image. This is accomplished as follows: 1. The pixels in the

    original image segment, corresponding to non-zero mask elements, are retained. The remaining

    pixel components are set to 1. 2. The coordinates of all the pixel components, corresponding to

    the non-zero pixels in the binary feature image, are determined and the corresponding pixels in the

    image segment are set to 1.

    The Data image is then inverted such that the features correspond to the non-zero pixels. If the

    number of non-zero pixel components is greater than a suitable threshold indicating presence of

    sufficient feature data, the final classification step is carried out. Forward matching and reverse

    matching is performed by correlating the inverted data image against the derived templates. Thederived template corresponding to the minimum obtained score is determined and a suitable media

    output is provided to the driver.

    3.4. Database module

    The template database is built which will be used to determine the information conveyed by the

    traffic signs. Upon construction, it takes the added functionality of inserting and retrieving records

    from the database as and when the TSR system dictates. The traffic sign templates, endorsed by the

    Delhi Traffic Police, are frequently used on national & state highways. These templates are used

    for compiling the template database.

    3.5. GUI module

    This module deals with

    Interaction with the various modules of the TSR system

    Accepts the input images from a suitable image capturing device such as a camera

    Provision of an audio output to the user providing the driver with information that has been

    extracted from traffic signs.

    364364

  • 7/30/2019 A Road Trafc Signal Recognition System Based on Template Matching Employing Tree Classier

    6/6

    4. Experimental results

    A test set was created by capturing about 2000 images of traffic signs consisting of triangle

    shaped, circular shaped with or without a cross. The traffic sign recognition algorithm gave excel-

    lent results and met the constraints of real-time situations. The recognition rates for different basicshapes obtained are illustrated in Figure 6. The algorithm could detect the signs rotated up to 200.

    Also the system was tested with signs which are partially occluded and got recognition rate about

    82%.

    5. Conclusion

    In this paper we propose a Traffic Sign Recognition system which is capable of recognizing

    the traffic signs at high accuracy. The system is invariant to rotation, partial occlusion of traffic

    signs, and the presence of other objects of the same color in the image. By further extending

    the proposed work it can be used in an autonomous vehicle guidance system which make crucial

    decisions regarding speed, trajectory of motion and has a myriad of applications.

    Figure 5. System Performance.

    References

    [1] C. Olson and D. Huttenlocher, Automatic target recognition by matching oriented edge pixels, IEEETransactions on Image Processing, Vol. 6, No. 1, 1997, 103113.

    [2] S. Lu and A. Szeto, Hierarchical Artificial Neural Networks for Edge Enhancement, Pattern Recog-

    nition, Vol. 26, No. 8, pp. 1993, 427435.

    [3] Piccioli, E. De Micheli, P. Parodi, M. Campani, A Robust Method for Road Sign Detection and

    Recognition, ECCV, 1994, 495500.

    [4] B. Bessere, S. Estable, B. Ulmer, and D. Reichardt, Shape classification for traffic sign recognition,

    In Proc. 1st IFAC Int. Workshop Intelligent Autonomous Vehicles, 1993, 487492.

    [5] Arturo de la Escalera, Luis E. Moreno and Miguel Angel Salichs, Road Traffic Sign Detection and

    Classification, IEEE Trans. on Industrial Electronics, Vol. 44, No. 6, 1997, 848859.

    [6] Anil K. Jain, Fundamentals of Digital Image Processing, Prentice Hall, 2001.

    365365