The road pavement condition is a ected by various impacts such as trucks, deicing reagents, base erosion, etc. After some time on the road surface occur defects. Engineers are commonly used to collect pavement surface distress data, during periodic road surveys, but it takes a lot of time and manpower. In this paper, we present our automatic defects detection and classi cation on road pavement method. We suggest the novel approach to detect the di erent types of defects such as rupture of the road edge, potholes, subsidence depressions. Images of road pavement have been preprocessed to noise lter and smooth, then classi ed two class - defects/ non defects, next step to process with defects class. We propose three main steps in our approach. First step is to detect defect position (ROI). In the second step, defect is described by its features. The last step is to classify defect each using these di erent defect features such as Chain Code Histogram, Hu-Moments, size of defect region(width and length, area) and histogram of image. In our approach the following algorithms have been used: Markov Random Fields for image segmentation, Random Forests algorithm for data classi cation. Data collection on real roads, real-time processing and comparison with other algorithms, analyzes the advantages and disadvantages of each methods.
For eective management of the road networks, one needs accurate and up to date information about road pavement defects. Thousands of kilometers of road pavement need to be inspected each year. Earlier, road defects information was obtained manually by human inspectors. But such manual methods are very slow and uncomfortable for inspectors and road users. In the last years, several automated inspecting techniques were implemented. Many of these state-of-the-art technologies involve machine vision and machine learning method. The objective of this article is to contribute to this eld.
Defect detection problem becomes especially dicult for noisy surfaces.[
Road pavement defects exist in many forms such as: rupture of the road edge, cracks (grid cracking, large crack), potholes, subsidence depressions. Each form of road pavement has got certain features, which are not the same, help us to distinguish them. If we only consider the simple features such as: shape descriptors, region descriptors (length, width, area) the data is unclear and dicult to apply defects road pavement recognition. An image can be considered as a mosaic of dierent texture regions, and the image features associated with these regions can be used for recognition. The purpose of this paper is to study the use of combination of dierent types of features, in particular, textural. The article is organized as follows. First we provide brief overview of related work. Then we describe defect pavement detection method and improve quality of image segmentation by Markov random eld. Finally, we present data classication based on Random Forest algorithm and conclusions. 2
Several researchers have considered the use of such texture features for pattern
retrieval [
Defects Detection and Classication Method on Road Pavement Our goal is nd the most ecient method using combination of dierent kind of features (Histogram, CCH - Histogram chain code, Moments-hull, shape of features) and machine learning algorithms (MRF- Markov random elds, Random forest method). 3.1
Feature Extraction We propose to preprocess images before the feature extraction. First we apply noise ltering using Gaussian lter and convert to gray scale images. On the next step we perform image segmentation. We propose divide the image into separate regions. Then we separate pixel defects to detected a connected region. We use morphological method to detect pixels corresponding to defects and to remove small regions which are considered as noise. We consider the following defects. Block crack: Interconnected cracks forming a series of blocks approximately rectangular in shape, commonly distributed over the full pavement. Attributes of block crack defect are: Predominant crack width ( mm), predominant cell width (mm), area aected ( m2).
Longitudinal Cracks: Unconnected crack running longitudinally along the pavement. Attributes of Longitudinal Cracks defect are Crack width ( mm), Crack length (m), Crack spacing ( mm), Area aected ( m2).
Potholes: Irregularly shaped holes of various sizes in the pavement. Attributes of Potholes defect are depth of potholes ( mm) and area of pothole ( m2). From the analysis of the attributes of each defect, we selected the following features: Hu-moments: The most notable are Hu-Moments which can be used to describe, characterize, and quantify the shape of an object in an image. Hu-Moments are normally extracted from the shape of an object in an image. By describing the shape of an object, we are able to extract a shape feature vector (i.e. a list of numbers) to represent the shape of the object. We can then compare two feature vectors using a similarity metric or distance function to determine how ’similar’ the shapes are.
Chain code histogram : The chain code histogram (CCH) is meant to group
together objects that look similar to a human observer[
The Freeman chain code [
The calculation of the chain code histogram is fast and simple. The CCH is a discrete function: p(k) = nk=n; k = 0; 1; :::; K 1; where nk is the number of chain code values k in a chain code, and n is the number of links in a chain code. Beside we consider also size of defect region (width and length, area) and histogram of image. 3.2
Construction of Map of Defects To automatically label regions defect/non defect, a pattern recognition system operating over a simple feature space is proposed. The feature space is multidimensional, in this problem we build 4 dimensional, being constructed using regions local statistics, computed for normalized and saturated images. The rst features is the mean value of all pixel intensities in a region. The second is chain code histogram, third is Hu-moment used to describe, characterize, and quantify the shape of an object in an image. Fourth is size of defect region (width and length, area) and histogram of image.
The rst aim is to split the image database into two subsets: the training images set, used to train classiers with manually labeled samples(images regions) containing defect pixels. The testing image set, the remaining images are supposed to be automatically processed by program for defect pavement detection and defect pavement types classication.
Defect pavement detection, where image region are labeled as containing defect pixels or not, and defect pavement type classication, where Block crack, Longitudinal Cracks, Potholes labels are assigned to each detected defect pavement. For defect pavement detection, an initial setup is required where operator selects images used to determine an optimum set of detection parameters accounting for pixel-by-pixel gray scale variation as related to defect pavement contrast, brightness, and surface conditions. During this setup phase, the program provides visual feedback of the detection results in the form of defect maps traced over the underlying images of control pavements.
These defect maps provide instant feedback on the eciency of the parameters. Through an iterative process, the optimal detection parameters are selected for each control pavement. Once the settings are selected, our program is programmed to automatically process the pavement images to detect defects pavement. For each defect, the length, width, and orientation are computed and saved. An example is a digital defect map as shown in Fig.1a demonstrates defect map corresponding to images shown on Fig.1b. To improve quality of image
These segments are called sites and have a predened orientation of 0, 45, 90 or 135 degrees. The separation between both cases is done with parameter k 2 (0; 1). Our goal will be to segment an image by constructing a graph such that the minimal cut of this graph will cut all the edges connecting the pixels of dierent objects with each other.
2.Set success := 0; 3.for each pair of labels D; N L do Find f^ = argminE(f ) among f within one D if E(f^) < E(f ) then set f := f^; success := 1; N swap of f ; end end 4.if success = 1 then
goto 2; end 5.Return f
We applied ecient graph based method to nd the optimal D(Defect)-N(Not
defect) as shown in Fig.3 swap or D - expansion given a labeling f . We use graph
cuts to eciently nd f^[
Defect on Road Pavement Classication
This section describes the classication based on unsupervised learning method
approach(Fig.5): Random Forest[
Load model classification of
machine learning Classification based on RandomForest algorithm Return type of defect road pavement End
Begin
Load road pavement image
database Preprocessing image Features extraction
Create features vector Fig. 5. Defect of pavement classi cation ow-chart. yi is assigned, where Y is the class set, yi 2 Y . The training set is: T = (x1; y1) ::: (xn; yn) : xi 2 R2; yi 2 fc1; c2:::; cng Where n is the number of points of the pattern vector x.
The Random Forest classier was built using the package Random Forest
4.516 for the R statistical environment [
Imbalanced data follows the idea of cost sensitive learning make random
forest more suitable for learning. Class weights are an essential tuning parameter
to achieve desired performance. In the tree induction procedure, class weights
are used to weight the Gini criterion for nding splits. In the terminal nodes
of each tree, class weights are again taken into consideration. We introduce the
concept[
True nagative rate = T NT+NF P , True Positive rate = T PT+PF N , Precision = T PT+PF P The classier was trained on pavements road dataset using Chain code histogram - CCH, Hu moments, size of defect for each variant. We also used method Boosting(GBTs) to classify this dataset and to compare results from two classication methods. The main dierence between these two algorithms is the order in which each component tree is trained.
The classier was built using the parameters and depth = (2; 5).
In table 2 shows the eect of increasing the number of trees in the ensemble. For both, increasing trees require more time to learn but also provide better results in terms of Mean Squared Error (MSE) is calculated as follows: ntree = (50; 100) and mtry = 2
n M SE = 1 X (f (xi) n i=1 yi)2
Where n is the number of test examples, f (xi) the classier’s probabilistic output on xi and yi are actual labels.
Random Forests are fast to train, but they often require deep trees. Random Forests do not overt as easily, but algorithm’s test error plateaus. Our experiments show that more trees are always better with diminishing returns. Deeper trees are almost always better subject to requiring more trees for similar performance. The above two points are directly a result of the bias-variance trade o. Deeper trees reduces the bias; more trees reduces the variance. There are several ways to control how deep our trees are (limit the maximum depth, limit the number of nodes, limit the number of objects required to split, stop splitting if the split does not suciently improve the t, ...). Most of the time, it is recommended to prune (limit the depth of) the trees if we are dealing with noisy data. Finally, we can use our fully developed trees to compute performance of shorter trees as these are a subset of the fully developed ones. 4
In this article we suggested the novel approach for road pavements defects automatic detection and classication. A simple boosting method is used to train the classier and the two sets (one for each road) make it possible to achieve results which demonstrates the robustness of the implemented method and algorithm for pavement crack detection based on Markov Random Fields. This method is based on the construction of an irregular lattice derived from the original image. The lattice is composed only by straight line segments. Firstly a local linear detection and an irregular lattice construction is done in order to highlight linear features locally.
We also propose to use to Graph cut method, which improve quality of image segmentation. From this we can detection part of pavement defect - non defect. The classication algorithm - Random Forest was able to correctly classify all the images contained in the two rst sets. In the test set simulating the real environment the achieved classication results were 95,5% which are very good. The authors are grateful to the attention and guidance of Prof. Dr. D. N. Sidorov. Authors are thankful to Center for Telecommunications and Multimedia, INESC TEC, Portugal for providing the dataset. ŁŁ Ł ,