Kelvin C. P. Wang

Automated Pixel‐Level Pavement Crack Detection on 3D Asphalt Surfaces Using a Deep‐Learning Network

The CrackNet, an efficient architecture based on the Convolutional Neural Network (CNN), is proposed in this article for automated pavement crack detection on 3D asphalt surfaces with explicit objective of pixel-perfect accuracy. Unlike the commonly used CNN, CrackNet does not have any pooling layers which downsize the outputs of previous layers. CrackNet fundamentally ensures pixel-perfect accuracy using the newly developed technique of invariant image width and height through all layers. CrackNet consists of five layers and includes more than one million parameters that are trained in the learning process. The input data of the CrackNet are feature maps generated by the feature extractor using the proposed line filters with various orientations, widths, and lengths. The output of CrackNet is the set of predicted class scores for all pixels. The hidden layers of CrackNet are convolutional layers and fully connected layers. CrackNet is trained with 1,800 3D pavement images and is then demonstrated to be successful in detecting cracks under various conditions using another set of 200 3D pavement images. The experiment using the 200 testing 3D images showed that CrackNet can achieve high Precision (90.13%), Recall (87.63%) and F-measure (88.86%) simultaneously. Compared with recently developed crack detection methods based on traditional machine learning and imaging algorithms, the CrackNet significantly outperforms the traditional approaches in terms of F-measure. Using parallel computing techniques, CrackNet is programmed to be efficiently used in conjunction with the data collection software.

Matched Filtering Algorithm for Pavement Cracking Detection

There have been rapid developments in automated surveying of cracking pavements in recent years. Laser-imaging technology has made the acquisition of shadow-free images feasible. However, because of the complexity of pavement surfaces, the diverse characteristics of cracks, the presence of foreign objects, and varying identification protocols, the results of automated cracking recognition have had limited use. A matched filtering algorithm is introduced for detection of pavement cracking. Unlike traditional edge detection approaches that adopt first-or second-order derivatives of image signals, the matched filtering algorithm detects cracks by matching predesigned filters with crack features by shape, orientation, or intensity. Experiments were conducted to compare the results of five traditional edge detectors (Roberts, Prewitt, Sobel, Laplacian of Gaussian, and Canny) and the matched filtering algorithm. The matched filtering algorithm was shown to be a robust approach for detecting cracks and had better performance in noise removal and detection of cracking. With matched filters aligned at various orientations, the matched filtering algorithm showed its distinctive advantages for extracting a single crack in its actual form and recording the cracks orientation to be used for more accurate classification in the next step of automated processing.

Surface Drainage Evaluation for Rigid Pavements Using an Inertial Measurement Unit and 1-mm Three-Dimensional Texture Data

During high-intensity rainfall, hydroplaning is likely and can affect driving safety. Studies have indicated that the risk of hydroplaning increases with the increase in the water film depth that is dependent on surface texture properties, flow path slope, flow path length, rainfall intensity, and pavement surface type. However, little research work has been conducted to investigate pavement surface drainage at network levels because the existing data acquisition systems cannot continuously measure related data sets at high speeds. In the presented study, texture data were continuously collected at high speeds with the emerging 1-mm three-dimensional (3-D) PaveVision3D Ultra technology. The cross slope and longitudinal grade data were acquired with an inertial measurement unit system. Data from two rigid pavements constructed with dragged and grooved surface texture were used in this study. The analysis of variance test and the multifactor treatment statistical method were used to investigate the factors that influence the calculation of water film depth. Texture properties and flow path slope were determined to be more significant for surface drainage capacity than was flow path length. The widely used PAVDRN model was used to calculate hydroplaning speed, and the potential hydroplaning performance of the test sites was evaluated. The significance of the presented data is twofold. First, it integrates the real-time 1-mm 3-D surface data and inertial measurement unit system data into a hydroplaning speed prediction model. Second, this method can identify hazardous locations where there is hydroplaning so that pavement engineers may take remedial measures, such as constructing superior grooving texture or posting appropriate traffic speed signs, to decrease hydroplaning potential and minimize traffic accidents.

Prep-ME: An Multi-Agency Effort to Prepare Data for DARWin-ME

The newly released DARWin-ME design system is a significant advancement in pavement design, but requires significantly more inputs from pavement designers. Many data sets need to be pre-processed before their use in the DARWin-ME procedure, such as Weigh-In-Motion (WIM) traffic data. This paper introduces the ongoing pooled-fund study TPF-5(242) aiming to assist state DOTs in the data preparation and improve the management and workflow of the DARWin-ME input data. A production software program called Prep-ME is developed with comprehensive database features to store and process climate, traffic and materials data. Specifically, this tool is capable of pre-processing, importing, checking the quality of raw traffic data, and generating the required traffic inputs required in DARWin-ME software by recognizing the differences in loading patterns or traffic groups. In addition, there are a number of other features in Prep-ME that may be useful to any highway agency, including (1) geo-referencing of design sites, weather stations, and water table observations; (2) populating materials inputs for DARWin-ME; and (3) preparing other DARWin-ME inputs based on the consensus of participating states. It is envisioned that through this pooled fund study, a possible nationwide platform for data preparation of DARWin-ME can be established with guidelines and supports provided to individual states for local calibration and implementation

Automated Joint Faulting Measurement Using 3D Pavement Texture Data at 1 mm Resolution

Faulting is an important performance indicator for jointed concrete pavements since it is highly associated with pavement ride quality. Currently many state DOTs have considered faulting measurements as a component of network-level pavement survey and used various instruments to collect faulting data such as fault meter, inertial high speed profiler, and newly developed laser based 3D technologies. The manual faulting measurement approach is time consuming, labor intensive, and presents a potential safety hazard to operators and traveling public. For the inertial high speed profiler, the joint faulting measured at different lateral locations may be different as a result of the irregularities of the joint characteristics. Moreover, past studies indicated existing algorithms for automated joint detection and identification need to be further developed. In this study a masking filter and the template match based algorithm are employed to locate the potential slab joints, and subsequently faulting at each joint is measured in accordance with the revised AASHTO R36-04 protocol. As a result, the slab joint quantity and locations identified using the proposed method is compared with that identified using the downward spike algorithm applied in ProVAL. Comparisons of the data indicate the template matched algorithm produces better identification results than the downward spike algorithm. The faulting validation would be made in the future research based on actual faulting measurements.

Reliability-Based Mechanistic-Empirical Pavement Design with Statistical Methods

“Reliability,” as defined in the Mechanistic–Empirical Pavement Design Guide (MEPDG), is an aggregated indicator defined as the probability that each of the key performance measures will be less than a selected critical level over the design period. Being such a complex system, the MEPDG—which is not a single closed-form design equation—cannot depend on classic reliability methods. Monte Carlo simulation is suitable for a robust reliability analysis but is impractical because of the extensive computation time required for any reasonable analysis with the MEPDG. The development of surrogate models for performance predictions becomes an option to represent the comprehensive modeling capability of the MEPDG efficiently. Improvements to the MEPDG reliability model based on several statistical methods are described. Five tasks are detailed. First, the MEPDG is calibrated to local conditions to reduce potential bias and variation from the national calibration. Then, risk analysis and screen analysis are conducted to determine the significant variables to include in surrogate models. Next, a comprehensive experimental design effectively plans MEPDG simulations. Then, response surface models of MEPDG are built through regression analysis. Finally, probabilistic design is achieved by Monte Carlo simulation. The result of these efforts is a state-specific MEPDG-based probabilistic pavement design tool kit named ReliME. This framework provides engineers more flexibility in data acquisition, design alternatives, optimization, and quality control. The frameworks successful application in Arkansas could easily be used in other states and incorporated into future mechanistic–empirical pavement design software.

Development of simplified traffic loading for secondary road pavement design

How to recognise traffic loading clusters and estimate load spectra using historical weigh-in-motion (WIM) data is critical to pavement mechanistic-empirical (ME) design. Various clustering approaches have been proposed in recent years mainly for high-volume roads. These methods require site-specific information to determine the design location cluster. In most cases for secondary road pavements, such data are missing due to resource constraints. In this paper, a simplified approach is developed to generate traffic loading for secondary road pavement design. With WIM data in Arkansas, the K-means cluster algorithm is applied and simplified Truck Traffic Classification clusters are developed. This method only requires prior knowledge of the dominant truck distributions on the design location and will alleviate the work related to traffic load data preparation. A case study is provided to illustrate the applicability of using the simplified clusters to generate Level 2 traffic inputs for DARWin-ME.

Clustering Methods for Truck Traffic Characterization in Pavement ME Design

AbstractAxle loading spectrum inputs obtained from existing weigh-in-motion (WIM) stations are one of the key data elements required in the pavement mechanistic-empirical (ME) design. Because of limited number of WIM stations within a state agency, it is critical to implement clustering approaches to identifying similar traffic patterns and developing cluster average Level 2 inputs for a particular pavement design. Even though several states have applied clustering methods for this purpose, they rely solely on hierarchical-based method. Many other types of clustering techniques based on different induction principles are available but have not been tested. In this paper, four types of clustering methods, including agglomerative hierarchical, partitional K-means, model-based, and fuzzy c-means algorithms, are implemented to cluster traffic attributes for pavement ME design using data sets from 39 WIM sites in Michigan. Two case studies, one flexible pavement and one rigid pavement, are conducted. The impac...

Feasibility Study for Gray Theory Based Pavement Smoothness Prediction Models

In the proposed Mechanistic-Empirical Design Guide (MEPDG) (NCHRP 1-37A, 2001), the functional performance indicator is pavement smoothness as measured by the International Roughness Index (IRI). The MEPDG IRI prediction models were developed based on the general hypothesis that changes in smoothness result from various distress types that can be predicted by the MEPDG program. Using pavement distress data fro the Long Term Pavement Performance (LTPP) database, traditional regression analysis was used to statistically establish the MEPDG prediction equations. This paper attempts to use a new technique for pavement smoothness prediction. The gray system theory was derived in the 1980s for modeling uncertain systems with the characteristics of partially known information. A pavement performance prediction system can fit the domain of the gray system. The gray theory based prediction method is used in this paper to develop IRI prediction equations. With the data exported from the LTPP database, it is found that certain specific types of distresses significantly affect the accuracy of the predictions. After trial and error calculations, Gray Model based smoothness predictions are established using influencing factors similar to the ones used in MEPDG. Based on the comparisons of results from the two prediction methods with LTPP field data, it is shown that the Gray Model based method provides promising results and be useful for modeling pavement performance.

Lane Marking Detection and Reconstruction with Line-Scan Imaging Data

Lane marking detection and localization are crucial for autonomous driving and lane-based pavement surveys. Numerous studies have been done to detect and locate lane markings with the purpose of advanced driver assistance systems, in which image data are usually captured by vision-based cameras. However, a limited number of studies have been done to identify lane markings using high-resolution laser images for road condition evaluation. In this study, the laser images are acquired with a digital highway data vehicle (DHDV). Subsequently, a novel methodology is presented for the automated lane marking identification and reconstruction, and is implemented in four phases: (1) binarization of the laser images with a new threshold method (multi-box segmentation based threshold method); (2) determination of candidate lane markings with closing operations and a marching square algorithm; (3) identification of true lane marking by eliminating false positives (FPs) using a linear support vector machine method; and (4) reconstruction of the damaged and dash lane marking segments to form a continuous lane marking based on the geometry features such as adjacent lane marking location and lane width. Finally, a case study is given to validate effects of the novel methodology. The findings indicate the new strategy is robust in image binarization and lane marking localization. This study would be beneficial in road lane-based pavement condition evaluation such as lane-based rutting measurement and crack classification.