Detection, Classification, Evaluation and Compression of Pavement Information

Transcription

1 Detection, Classification, Evaluation and Compression of Pavement Information S.Vishnu Kumar Maduguri Sudhir Md.Nazia Sultana ABSTRACT This paper presents crack detection and evolution algorithm describes the procedure of pavement image. The entire project is built upon wavelet transform and radon transform. The application of wavelets to images yields different frequency sub-bands. Also, this algorithm is based on the fact that crack pixels are darker than their surroundings. The crack detection algorithm decomposes a pavement image into four sub-bands one low frequency sub-band called approximation and three high frequency sub-bands called detail. Cracks appear clearly in the low frequency sub band at the first level through extended pseudo color matrix scaling. The algorithm of classification first revises and re-computes the horizontal, vertical and diagonal details at the first level through the energy conservation function, and then forms a new image through adding corresponding points in the four new sub-bands; finally it applies Radon transform to this new image. The Radon transform used to classification and evolution of pavement image. Keywords: Radon Transform, pavement distress, detection, evaluation I. INTRODUCTION Road maintenance to be performed before defects develop into more serious problems, such as potholes and popouts. The traditional manual pavement crack detection methods and approaches are too expensive, time consuming, dangerous and labor-intensive. The previous methods are 1) Visual inspection 2) A video based system. In the visual inspection method two operators travel at 20 km/h in which One as the driver, another to record the defect. In the visual inspection method we have to record the pavement up to 100 km/h The recorded video is then inspected off-line at speed of 20 km/h. these two method are Time consuming, costly and can be dangerous. In order to meet the the requirements of the rapid development of public transportation we have to detect the defect quickly. In order to provide better road network for people to use and Reduce maintenance cost we require the road maintenance. This paper presents a method for pavement crack detection, classification and evaluation using the Radon transform. The detection part of the algorithm is built upon the wavelet transform and the evaluation part is considered in the Radon transform domain. Since cracks have specific linear features in the space domain, the Radon transform can effectively be used on a binary image to classify and evaluate the cracks based on their possible patterns. The proposed algorithm assumes the fact that crack pixels are darker than their surroundings. The simulation results show that the proposed method is very effective and can provide information about the type and severity of the pavement cracks. II. THE PROPOSED ALGORITHM The proposed algorithm consists of three stages, namely (1) detection; (2) domain mapping; and (3) classification. A flowchart indicating the sequence is shown in Fig. 1. The three stages of the algorithm are explained in detail as follows. input image Crack Detection in Wavelet Domain Domain Mapping using Radon Transform Crack Classification/Evaluation Fig. 1: Flowchart of the Algorithm 2.1 Crack Detection In this stage, the true color pavement image is first transformed into a gray scale image. Following this step a discrete 2-D wavelet transform using db2 wavelet is applied to this gray scale image to yield four sub-bands namely, HH, HL, LH and LL. The notion behind this process is to first filter each row followed by a down sampling to obtain two N X M/2 images from an N x M image. Then, apply the filter columnwise and subsample the filter output to obtain four N/2 x M/2images. This will lead to a set of 4 sub-images known as LL, LH, HL, and HH subbands. The decomposition of an image into four subbands is represented in the Fig

2 M/2 M/2 N/2 LL LH N/2 HL HH Fig. 2: Wavelet Decomposition of an Image Finally, extended pseudo color matrix scaling is performed on the approximation matrix. A pseudo-color image is derived from a gray scale image by mapping each pixel value to a color according to a table or function. Pseudo-coloring can make some details more visible, by increasing the distance in color space between successive gray levels. Alternatively, depending on the table or function used, pseudo-coloring may increase the information contents of the original image. In this way, the cracks can easily be detected from the pavement image. alligator cracks. This can be seen in Figure 9. On the other hand, if there are two or three peaks at different angles, the cracks are the combined single cracks of longitudinal, transversal, or diagonal types. If there are four or more peaks, one needs to determine first whether they form the patterns. If they do, this indicates the existence of block or alligator cracks. In summary, the Radon transform is applied to the image after crack detection. The output of the Radon transform shows the value range of the Radon transform of the image, the projection angle and the projection position. Since peaks in the Radon domain have a relationship with the crack in the space domain, a graph of the Radon transform of the image can display some important information as follows: (1) Number of the peaks may indicate number of the cracks (2) Projection degree plus 90 degrees shows whether it is longitudinal, horizontal or diagonal; (3) Area of the peak determines the width of the crack (4) Value in the Radon domain means the rough length of the cracks (5) Projection position gives the rough locations of the cracks. 2.2 Domain Mapping Using Radon Transform Radon transform utilizes a set of projections at different angles in an image f(x, y). The resulting projection is the sum of the intensities of the pixels in each direction, i.e. a line integral. The result is a new image R(ρ,θ ). It can be expressed using the following mathematical formula: R(ρ,θ ) = f (x, y)δ (ρ xcosθ ysinθ dxdy Where δ (.) is the Dirac delta function, ρ and θ represent the distance to the origin of the coordinate system and the angle of the line, respectively. When referring to the pavement image, a crack will be projected into a peak or bin in the Radon transform and the projection direction will be perpendicular to the crack. The task in this stage is to build the relationship between the Radon domain and space domain. To better understand their relationship, different types of typical cracks are simulated by considering black lines on a gray background, as shown on the left side of the various figures. The right side of the figures shows their corresponding Radon transform. The angle of a crack is defined as the angle between the direction of the crack and the lateral direction of the pavement. From Figure 3, it can be seen that a peak at 90 in the Radon transform indicates a transversal crack on a pavement, and the x coordinate of the peak determines the position of the crack. Similarly, a peak at 0 in Figure 4 indicates a longitudinal crack, and a peak at 135 in Figure 5 or 45 in Figure 6 indicates a diagonal crack at 45 and 135, respectively. For block and alligator cracks, the number of the peaks increases to at least four. Block cracks form two groups of peaks located with a difference angle of about 90. A peak array is defined as having at least two peaks at a certain angle. For the block cracks shown in Figures 7 and 8, it is evident that the difference of their angles is 90. When the difference of the angles of two groups of peaks deviates from 90, the pattern forms Fig. 3: Transversal Crack and its Radon Transform Fig. 4: Longitudinal Crack and its Radon Transform Fig. 5: Diagonal Crack at 45 and its Radon Transform 808

3 Fig. 6: Diagonal Crack at 135 and its Radon Transform Fig. 7: Block Crack and its Radon Transform Fig. 8: Rotating Block Crack and its Radon Transform Fig. 9: Alligator Crack and its Radon Transform From the Radon transform, it is also known that the larger the peak, the wider the crack and the area of the peak can be used to determine the width or the severity of a crack. Similarly, the position of the peak can be used to estimate the rough location of the crack. Since Radon transform integrates the crack along its dominant orientation, the peak value has a strong relationship with the length of a crack. 2.3 Classification and Evaluation The process of crack classification and evaluation involves three important parameters, the number of cracks, the position, and the maximum value of the window in the Radon transform. The position includes the angle _ and location x of the window. Peaks in the Radon domain are related to the cracks in the space domain. Generally, cracks in the real images will not be exactly a straight line as was shown in the above simulation. Therefore, the classification and evaluation algorithm is more involved and requires a more detailed explanation. First, a threshold is needed to segment the Radon transform image to separate the peaks from the non-peak regions. A Peak is defined as 75% of the maximum value of the Radon Transform. After finding all the peaks, a clustering algorithm is applied to merge all the peaks in the same window. The procedure for crack classification and evaluation is summarized as follows: 1) Let s define an array of peaks to be a set of peaks with the same orientation (same angles) in which the number of peaks being greater or equal to two. Two arrays with a difference of angles equal to 90º is considered a block type crack. Note that the x coordinate of the window represents the location of the crack and the value of the window indicates the possible length. Two arrays with a difference of angles not equal to 90º is classified as an alligator type of crack. 2) When a crack pattern is not a block or alligator type, it will then indicate the presence of a single or a combination of single cracks. For a single crack, if 0º< _ <10º or 145º< _ <180º, this indicates the presence of a longitudinal crack. 3) Similarly, if 80º< _ <100º, this indicates the presence of a transversal crack. 4) On the other hand, if 35º< _ <80º this will be an obtuse angle diagonal crack, 5) otherwise, it will indicate an acute angle diagonal crack. In each case, the x coordinate of the window (representing the location) and the value of the window (representing the length) of a crack is also recorded. 2.4 Image Compression In this section, the image compression method is presented. The notion behind compression is based on the concept that the regular signal component can be accurately approximated using a small number of approximation coefficients (at a suitably selected level) and some of the detail coefficients. The compression procedure is composed of three steps: 1. Decomposition; 2. Detail coefficient thresholding and 3. Reconstruction. In step 2, a threshold is selected for each level from 1 to 3, and hard thresholding is applied to the detail coefficients. The entire compression procedure is accomplished by two functions named ddencmp and wdencmp applying to the gray scale pavement image. Ddencmp function gives all necessary default values for compression and wdencmp performs compression process of image using the parameters generated by ddencmp. After that, parameters such as percentage of zeros in reconstruction image called the compression performance and percentage of energy recovery are automatically calculated. The detail of two functions is expressed as follows: [THR, SORH, KEEPAPP, CRIT] = ddencmp (IN1, IN2, X) This function returns default values for compression, using wavelets, of matrix X, here is two-dimensional image. THR is the threshold, SORH is for soft or hard thresholding, KEEPAPP 809

4 allows you to keep approximation coefficients (when it equals to 1), IN1 is 'cmp' for compression. IN2 is 'wv' for wavelet. [XC,CXC, LXC, PERF0, PERFL2] = wdencmp('gbl', X, 'wname',n,thr,sorh, KEEPAPP) Wdencmp is a two-dimensional compression-oriented function. It performs compression on an image. This function returns a compressed version XC of input image X (two-dimensional) obtained by wavelet coefficients thresholding using global positive threshold THR. Additional output arguments [CXC, LXC] are the wavelet decomposition structure of XC. PERF0 and PERFL2 are L2-norm recovery and compression score in percentage. PERFL2 = 100 * (vector-norm of CXC / vector-norm of C)2 if [C,L] denotes the wavelet decomposition structure of X. Wavelet decomposition is performed at level N. 'wname' is the wavelet name. SORH ('s' or 'h') is for soft or hard thresholding. If KEEPAPP = 1, approximation coefficients cannot be thresholded, otherwise they are subject to it. Fig. 10 (a, b, c, d) III. SIMULATION RESULTS The images used in this paper were downloaded from the web in true color. A crack is composed of sets of pixels that are darker than the pixels in its surrounding areas. To better test the algorithm, various pavement images containing, longitudinal, transversal, diagonal, block, and alligator cracks as well as other irregularly shaped cracks are considered in the simulation. When selecting a wavelet filter, it is desirable that the wavelet have the ability to rapidly detect the crack and be able to segment the crack from the input image. As a result, a number of wavelets have been tested on various pavement images. Daubechies wavelet with an order two had more satisfactory results, and was thus used in the paper. Figures 10, 11, 12 a, b, c, d represent an original pavement image, crack detected image, the Radon transform image, and the number of peaks windows for transversal, composite, and alligator type cracks, respectively. Various parameters of the cracks in the above Figures obtained from the simulation including the orientation, position, and types are summarized in Table (a) Grayscale Image (b) Detected Crack Fig. 11 (a, b, c, d) (a) Grayscale Image (b) Detected Crack (a) Grayscale Image (b) Detected Crack Fig. 12 (a, b, c, d) 810

5 input image Table 1: Classification of image (k) No. Of Windows Ѳ X max Type Figure Transversal Figure ,130,180 5,0,-5 40 Composite Figure 12 9 all all 92 Alligator AUTHOR S PROFILE Mr.S.VISHNU KUMAR B.Tech (ECE), M.Tech (Digital Electronics and Communication Systems). He was interested in Digital Image Processing and Electronics. IV. CONCLUSION In this paper, an algorithm that can be used for automatic crack detection, classification and evaluation from any pavement image is presented. This algorithm uses wavelet transform and pseudo coloring to detect the cracks and applies a Radon transform on the binary image to classify and evaluate the cracks. It has been shown that the Radon transform can be used to determine the possible type, location, area, length, and the width of the pavement distresses. Further, the proposed algorithm can be applied to nearly any image with no limitation to a specific resolution or a specific camera. REFERENCES [1] H.D. Cheng, M. Miyojim, Automatic pavement distress detection system, An International Journal of Information Sciences, 108(1), pp , [2] T. Yamaguchi, S. Nakamura, R. Saegusa, and S. Hashimoto, Image-based crack detection for real concrete surfaces," in IEEJ Transactions On Electrical And Electronic Engineering, vol. 3, pp , [3] J. Zhou, P.S. Huang, F. Chiang, Waveletbased pavement distress detection and evaluation," in Optical Engineering, Vol. 45(2), [4] J. Zhou, P.S. Huang, F. Chiang, Wavelet-based pavement distress classification", in Journal of the transportation research board, pp , [5] E.Teomete, V.R.Amin, H.Ceylan, and O.Smadi, Digital image processing for pavement distress analyses," in Proc. of the 2005 Mid-Continent Transportation Research Symposium, Ames, Iowa, [6] J. Bray, B.Verma, X. Li, and W.He, A neural network based technique for automatic classification of road cracks," in IEEE International Joint Conference on Neural Networks, [7] W. Xiao, X. Yan, and X. Zhang, Pavement distress image automatic classification based on density-based neural network," in Lecture Notes in Computer Science, Pattern Recognition, pp , [8] H.D.Cheng and C. Glazier, Automated real-time pavement crack detection and classification system," in Final Report for Highway IDEA Project 106, Transportation Research Board, Washington, D.C., [9] S. Mallat, A Compact Multiresolution Representation: The Wavelet Model, Proc. IEEE Computer Society Workshop on Computer Vision, pp.2-7, [10] Mr. MADUGURI SUDHIR B.Tech (ECE), M.Tech (DECS) Gold medalist is currently working as an Assistant Professor, in Department of Electronic, Communication & Engineering. He was received GOLD MEDALS in B.Tech (ECE) and M.Tech His Research interests on Electronic Circuits and Image Processing MD.NAZIA SULTANA B.Tech (ECE), M.Tech (DECS) Gold medalist. is currently working as an Assistant Professor, in Department of Electronic, Communication & Engineering. She was received GOLD MEDAL in B.Tech (ECE). Her Research interests on Electronic Circuits and Image Processing 811