FREE-FORM ANISOTROPY: A NEW METHOD FOR CRACK DETECTION ON PAVEMENT SURFACE IMAGES

Transcription

1 FREE-FORM ANISOTROPY: A NEW METHOD FOR CRACK DETECTION ON PAVEMENT SURFACE IMAGES Tien Sy Nguyen, Stéphane Begot, Forent Ducuty, Manue Avia To cite this version: Tien Sy Nguyen, Stéphane Begot, Forent Ducuty, Manue Avia. FREE-FORM ANISOTROPY: A NEW METHOD FOR CRACK DETECTION ON PAVEMENT SURFACE IMAGES. 18th IEEE Internationa Conference on Image Processing, Sep 2011, Bruxees, Begium. <ha > HAL Id: ha Submitted on 27 Sep 2011 HAL is a muti-discipinary open access archive for the deposit and dissemination of scientific research documents, whether they are pubished or not. The documents may come from teaching and research institutions in France or abroad, or from pubic or private research centers. L archive ouverte puridiscipinaire HAL, est destinée au dépôt et à a diffusion de documents scientifiques de niveau recherche, pubiés ou non, émanant des étabissements d enseignement et de recherche français ou étrangers, des aboratoires pubics ou privés.

2 FREE-FORM ANISOTROPY: A NEW METHOD FOR CRACK DETECTION ON PAVEMENT SURFACE IMAGES Tien Sy NGUYEN (1,2), Stéphane BEGOT (1), Forent DUCULTY (1), Manue AVILA (1) (1) PRISME Laboratory, University of Oreans, France, (2) Vectra road engineering, Emai: manue.avia@univ-oreans.fr ABSTRACT This paper presents a new measure which takes into accounts simutaneousy brightness and connectivity, in the segmentation step, for crack detection on road pavement images. Features which are cacuated aong every free-form paths provide detection of cracks with any form and any orientation. The method proposed does not need earning stage of free defect texture to perform defaut detection. Experimenta resuts were conducted on some sampes of different kinds of pavements. Resuts of the method are aso given on other kinds of images and can provide perspectives on other domains as road extraction on sateite images or segment bood vesses in retina images. Index Terms Image segmentation, crack detection, defect detection, texture anayses 1. INTRODUCTION From 1990, there is a growing interest in pavement defect detection using image processing techniques [1]. Crack detection on pavement surfaces is a difficut probem due to the noisy pavement surfaces. There are different kinds of texture that can be encountered on road pavements. Cracks can have any form; crack size can be as sma as 1 pixe in width and thinner than aggregate size. Figure 1 iustrates some road pavement image sampes. (a) (b) (c) Figure 1: pavement image sampes: (a) ongitudina crack, (b) transversa crack, (c) aigator crack. The paper is organized as foows: In section 2, a short review of defect detection methods is proposed. In section 3, we introduce a new approach based on Free-Form Anisotropy (FFA) for segmentation. First, we reca Conditiona Texture Anisotropy (CTA), which was introduced by F. Roi [2], and adapted for pavement crack detection [3]. Then, we expain FFA method which overcomes CTA imitations (orientations and inear form of crack). Section 4 summarizes experimenta resuts. Finay, we concude and propose others purposes for the method. 2. REVIEW OF DEFECT DETECTION METHODS Because of the road pavement image nature, crack detection methods, in iterature, were based on stabe characteristics of cracks. We can give the two foowing characteristics of cracks [1], [4]: - Brightness: crack pixes are darker than their neighbors. - Form: crack is continuous or coud be formed by various continuous segments. Its ength is greater than its width and than granuate size. Both of these characteristics can be noised with shadows, ane marking, etc. Usuay, crack pavement detection methods can be divided into four sequentia stages: pre-processing, segmentation, post-processing and cassification. According to [5], in most of existing methods, cassification step is trivia due to the easy task consisting in separating different crack types (ongitudina, transversa and aigator). Most of approaches, in iterature, use brightness characteristic of crack for segmentation foowed by a postprocessing step, which uses connectivity characteristic to connect crack segments and to eiminate noises. Threshoding is frequenty used to segment cracks, fixed threshod in [6-7] or fuzzy threshod in [8]. Some methods [9, 10] divide image into grid ces and then cassify each ce as crack or crack-free ce by comparing mean and standard deviation of the ce with their neighbors or by UINTA fitering [11]. Authors in [5] supposed that, by appying a 2D Continuous Waveet Transform (CWT), the differences between crack pixes and background pixes coud be raised up. After segmentation step, crack appears as discontinuous regions with noises. Post-processing step is needed to remove noise and to connect crack segments. In case of use of grid ces, crack as thin as 1 pixe cannot be detected considering ony statistica features of intensities. Use of waveet [5] is a good approach by considering muti-resoution aspect, but their resuts showed that CWT not ony rises up cracks but aso noises. Recent approaches [9, 11] provide very noisy resuts for the /11/$ IEEE 1093

3 segmentation step and it is hard to obtain good connection resuts. In the next part, we propose a new method which takes into account simutaneousy intensity and crack form features for segmentation step. 3. FREE-FORM ANISOTROPY 3.1. Conditiona Texture Anisotropy for crack detection Conditiona Texture Anisotropy (CTA) was first introduced by F. Roi [2]. The main idea was to find out a measure which produces sma vaues in one orientation (e.g. aong crack orientation) and higher vaues in other orientations. Let w 1 be the cass of defect-free pixes and w 2 be the cass of defaut pixes. The CTA of a pixe can be defined as: max{ p( x w1 )} min{ p( x w1 )} CTA( ) = (3.1) max p x w { ( )} where : x = (feature 1, feature 2,, feature n) is a set of n texture features computed aong the orientation. p(x w 1 ) is probabiity for the pixe to be a defect-free pixe aong the orientation. In [2], is usuay one of the 4 traditiona orientations (0,45,90,135 ). According to crack characteristics, p(x w 1 ) shoud take ow vaue on dominant orientation of crack. As we, p(x w 1 ) wi take high vaues for other orientations. We can deduce from equation (3.1) that CTA takes high vaue (cose to 1) on crack pixes and ow vaue for defect-free pixes (cose to 0). According to brightness characteristic of crack and Gaussian form histogram [5] of pavement image, mean and standard deviation of pixe intensity of oriented segment have been chosen as features to cacuate CTA. These oriented segments are composed of (2d +1) pixes. h( π, π ) = sup{min ( π, π )} (3.2) Figure 2 : oriented segments and sup-min function to evauate degree of coherence between two sources. To compute p(x w 1 ), we use possibiity theory [13] to evauate degree of coherence (3.2) between two sources (Figure 2). Source π i is composed by mean µ i and standard deviation σ i.. We dispose of 4 sources, one for each orientation. To avoid training stage for background characterization, we use this hypothesis: crack affect texture ony in one orientation (the crack orientation). Then to evauate background (or defect-free) source, we compute mean of the 3 sources which have higher vaues (Figure 3). A two eves threshod [3] is used to produce binary 1 images. The most important parameter of the method is the distance d, of oriented segments, used to compute features. Figure 3: Computation of CTA for 1 and 2 using the degree of coherence (right figures). In Figure 4, we see resuts of CTA, on an image, on which we produce synthetic defects. We create defects with 3 of the traditiona orientations and one different. These defects go from 1 to 4 pixes in width, with intensities chosen randomy with vaues near their neighbors. transversa profi d=4 d=16 origina image binary images Figure 4: CTA resuts for different distances d. In this Figure 4, we see that defects with traditiona orientations are correcty detected. When distance d is high enough, the background noises disappear, but we see aso that, for non traditiona orientation, ony arge defects are detected. So CTA is interesting for traditiona orientations and it provides efficient background suppression when the distance d is high enough. But for other orientations, CTA suppresses aso thin defects Free-Form Anisotropy To overcome the CTA imitations, we propose the Free- Form Anisotropy method (FFA) which cacuates, for each pixe, features aong every free-form path Definition We reach 4 minima paths according to 4 goba orientations as it is shown in figure 5. A minima path is defined as a path for which sum of pixe intensities is the smaest. Graph theory, for exampe Dikstra agorithm [14], provides 1094

4 soution to find efficienty these minima paths. transversa profi d=4 d=16 Figure 5 : Minima paths of pixe according to 4 orientations: (a) transversa, (b) ongitudina, (c) diagona 135, (d) diagona 45. Features are cacuated for each path and converted into sources and we can compute background source π as it was done for CTA. Then the FFA of each pixe can be formuated as: { h( π, π )} min{ h( π, π )} max{ h( π, π )} max FFA( ) = (3.3) with a goba orientation, and h π, π ) computed on ( the 4 minima paths (Figure 5) with pixe at the center of the path with (2d+1) ength. (a) (c) (d) Figure 6 : Computation of FFA for 2 pixes width d = 30. (a,b) crack pixe (c,d) defect-free pixe. As for CTA measure, the FFA measure is cose to 1 for crack pixes and cose to 0 for defect-free pixes. FFA computation is iustrated on figure 6. We can see that the free-form path foows the crack. This shows the abiity of FFA to expore with accuracy different crack forms. Without crack, minima paths produce sources with high correation eve (Figure 6 d). The same strategy, as for CTA, was used to test FFA. Figure 7 shows the origina image with synthetic cracks, FFA and binary images for different distances d, and profies of ine extracted on images. If distance d is high enough, a cracks are detected in any orientation and with the minima width (1 pixe). (b) origina image binary images Figure 7 : FFA resuts for different distances d Comparative resuts 4. RESULTS Figure 8 : Anisotropy vs. Subirat s 2D CWT. Inspected image (a), Subirat s CWT (b), CTA (c), FFA (d). Figure 8 (a) shows a ongitudina crack on rea image of road pavement. In this exampe, we compare CTA and FFA with 2D CWT method [9]. Both CTA resuts and FFA resuts (Figure 8 c and d) contain ess of noise. This is the abiity of the segmentation step to take into account intensity, form and connectivity of the defaut. FFA is better than CTA for connecting crack segments Texture variation We use different kinds of pavement images with different properties. Tabe 1 gives some of theses textures attributes for each kind of texture. We see granuate size, contrast. The two ast parameters are Haraick attributes extracted from co-occurrence matrix [15]. Image Granuate size (mm)min/max Contrast Correation Entropy 1 0/ ,001 8,7 2 0/10 537,3 0,0005 9,38 3 0/13 612,4 0,0032 9,4 4 0/18 876,72 0, ,9 Tabe 1 : Texture attributes of seected images. On these images (Figure 9), we generate synthetic cracks with the same method as in 3.1. Cracks have no segment forms and intensities are chosen randomy. cracks image1 image2 image3 image4 Figure 9 : Synthetic cracks on different kinds of pavement images. 1095

5 In Figure 10, we show FFA resuts for different distances d. In each case, a cracks are fuy detected. If distance is high enough, there is no noise detection. The method is abe to perform background suppression for different kinds of textures. This demonstrates the robustness of the method. characteristics as pavement cracks. Resuts show that FFA can aso be usefu on these kinds of images with fine structures. image1 image2 image3 image4 Figure 10: FFA resuts on different kinds of pavement images Resuts on rea defects on pavement images These tests were performed on rea images (Figure 11) captured dynamicay [3-4]. We obtained 93.6% detection rate and 13.7% fase aarm with FFA. With CTA method, we obtained ony 73.8% detection rate and more than 27.6% fase aarm. Computation time for these high resoution images (2048x2048 pixes) is about 20 seconds (De Precision PWS670, Xeon 3.6 GHz, RAM 4Go). Figure 11 : FFA resuts. (a) Longitudina crack, (b) transversa crack, (c) aigator crack, (d) defect-free. In this figure, a defects (figure 11 a, b, c) are detected. We see detais of aigator crack (c). Without defect (figure 11 d), nothing is detected. and we see the efficiency of the method which provides resuts with very ow noise. 5. CONCLUSION AND PERSPECTIVES d=4 d=32 In this paper, we have introduced a new method for crack detection on road pavement images. By considering a characteristics of crack and by unrestricting crack orientations and forms, the method provides good resuts on crack segmentation. Cracks which are as sma as 1 mm (1 pixe in width) coud be detected with any form and orientations. Fine structures on other kinds of images ike medica images (Figure 12 a) or sateite images (Figure 12 c) have simiar Figure 12 : FFA appied on other kind of fine structure extraction, bood vesses in retina images (a, b), road extraction on sateite images (c, d). Characteristics of these fine structures in other random texture surfaces are simiar to cracks on pavement surfaces. Good resuts obtained on some images of this kind suggest good perspectives for using FFA in other domains as: ceramic damages detection, road network extraction in sateite images and bood vesses segmentation in retina images. 11. REFERENCES [1] Schmidt B., Automated Pavement Cracking Assessment Equipment - State of the Art. 2003, Word Road Association (PIARC). [2] Roi F., Measure of texture anisotropy for crack detection on textured surfaces, in Eectronics Letters pp [3] Nguyen T.S., M. Avia, and S. Begot, Automatic defect detection on road pavement using anisotropy measure, in Proceedings EUSIPCO [4] Nguyen T.S., et a., Detection of defects in road surface by a vision system, in Proc. 14th IEEE Mediterranean Eectrotechnica Conference MELECON pp [5] Chambon S., J. Dumouin, and P. Subirats, Introduction of a waveet transform based on 2D matched fiter in a Markov Random Fied for fine structure extraction: Appication on road crack detection, in SPIE Conference on Image Processing: Machine Vision Appications II, San Jose, United-States [6] Naamothu S. and K.C.P. Wang, Experimenting with Recognition Acceerator for Pavement Distress Identification, in Transportation Research Record pp [7] Chua K.M. and L. Xu, Simpe Procedure for Identifying Pavement Distresses from Video Images, in J. Transp. Engrg pp [8] Cheng,.D., et a., Nove Approach to Pavement Cracking Detection Based on Fuzzy Set Theory, in Journa of Computing in Civi Engineering pp [9] Oiveira H., Correia P.L., Identifying and retrieving distress images from road pavement surveys, in Proc. 15th IEEE Internationa Conference on Image Processing ICIP pp [10] Huang Y. and B. Xu, Automatic inspection of pavement cracking distress, in Journa of Eectronic Imaging [11] Oiveira H.; JJC Caeiro; Correia, P.L.; Improved Road Crack Detection Based on One-cass Parzen Density Estimation and Entropy Reduction, Proc IEEE Internationa Conf. on Image Processing ICIP 2010, pp [12] Howe, R. and G.C.a. Gerardo, An assessment of the feasibiity of deveoping and impementing an automated pavement distress survey system incorporating digita image processing. 1997, The Virginia Transportation Research Counci. [13] Dubois D. and H. Prade, On the use of aggregation operations in information fusion processes. Fuzzy Sets and Systems , [14] Thomas, H.C., et a., Introduction to Agorithms. 2001: McGraw-Hi Higher Education. [15] Theodoridis S. and K. Koutroumbas, Pattern Recognition, Fourth Edition. 2008: Academic Press. pp