Yichang Tsai

Critical Assessment of Pavement Distress Segmentation Methods

Image segmentation is the crucial step in automatic image distress detection and classification (e.g., types and severities) and has important applications for automatic crack sealing. Although many researchers have developed pavement distress detection and recognition algorithms, full automation has remained a challenge. This is the first paper that uses a scoring measure to quantitatively and objectively evaluate the performance of six different segmentation algorithms. Up-to-date research on pavement distress detection and segmentation is comprehensively reviewed to identify the research need. Six segmentation methods are then tested using a diverse set of actual pavement images taken on interstate highway I-75/I-85 near Atlanta and provided by the Georgia Department of Transportation with varying lighting conditions, shadows, and crack positions to differentiate their performance. The dynamic optimization-based method, which was previously used for segmenting low signal-to-noise ratio (SNR) digital radiography images, outperforms the other five methods based on our scoring measure. It is robust to image variations in our data set but the computation time required is high. By critically assessing the strengths and limitations of the existing algorithms, the paper provides valuable insight and guideline for future algorithm development that are important in automating image distress detection and classification.

Real-time Speed Limit Sign Recognition Based on Locally Adaptive Thresholding and Depth-First-Search

Road inventory for a highway includes a variety of traffic signs, such as stop signs and speed limit signs, pavement width, lane number, and others. This article describes a current research project being undertaken by the Georgia Department of Transportation to re-engineer and streamline the existing road inventory data collection process, in order to develop a real-time data collection system that is accurate, efficient, and safe. The authors present an algorithm for recognizing speed limit signs (SLS) from video imaging and extracting the numerical numbers of SLS to support real-time road inventory data collection operations. The algorithm consists of color segmentation based on locally adaptive thresholding extraction of regions of interest (ROI) using a depth-first-search algorithm, followed by speed limit sign detection and speed limit number extraction by means of optical character recognition and 2D correlation. The average processing time for an image of 1200 x 800 pixels is about 125 ms. Experimental results from 1,401 video images show 0 percent false positives out of 1,278 images containing no SLS, and 3 percent false negatives out of 123 images containing SLS. Signs that are tilted up to 15 degrees and surrounded by white and non-white backgrounds can be correctly detected independent of lighting conditions. Partly shadowed, partly blocked, or severely blurred SLSs (due to a fast-moving camera) would not be correctly detected.

Enhanced Roadway Geometry Data Collection Using an Effective Video Log Image-Processing Algorithm

With the advancement of information technology, video logging has become a feasible and common practice for state-level departments of transportation (DOTs) to visualize roadway conditions to support various transportation activities. A new and innovative image-processing algorithm and method using video log images to enhance the highway roadway geometry data collection is presented. The algorithm is based on Canny edge detection, slope-adaptive edge reduction, a novel recursive approximation of the vanishing point, localized color thresholding, and knowledge-based Hough transform. Experiments on real digital images indicate that the algorithm is feasible for automated road geometric data collection, especially in recognizing lane marks and shoulder edges. The algorithm was implemented with Visual C++ language and tested on a Pentium IV 3.06-GHz computer. It indicates that the average processing speed is about 280 ms for an image size of 1,300 x 1,050. The developed method can be used by state DOTs to sav...

Automatic Detection of Deficient Video Log Images Using a Histogram Equity Index and an Adaptive Gaussian Mixture Model

Video log images are often used by transportation agencies to manually or automatically extract roadway infrastructure information, including roadway geometry, signs, etc. Poor-quality images, especially those having illumination-related deficiencies caused by color corruption with a plain-like grayscale histogram, sun glare, or darkness problems, are unacceptable and need to be identified. Manually reviewing the tens of millions of video log images for quality control is labor intensive and time-consuming, so there is a need to develop automatic video log image quality control procedures. The contribution of this article is that it formulates a new problem of roadway video log image quality control and then proposes a reasonable solution to address this problem in the hope that it will motivate the development of new algorithms by other researchers. For the first time, an algorithm using a Histogram Equity Index and an adaptive Gaussian Mixture Model is proposed to address the video log image quality issue by automatically detecting illumination-related deficiencies. The Alberta Department of Transportation provided 15,489 video log images to test the proposed algorithm. Test results show that the developed algorithm can detect illumination-related video log image deficiencies with a false positive rate of 4%, 3%, and 12%; a false negative rate of 15%, 17%, and 19% for plain-like color corruption, dark, and sun glare conditions, respectively; computation time is 0.1 second/image. The proposed algorithm could potentially be used to improve video log image data quality control.

A Generalized Framework for Parallelizing Traffic Sign Inventory of Video Log Images Using Multicore Processors

Abstract:  Video log images could potentially be used for state transportation agencies to automatically inventory traffic sign assets. However, processing millions of video log images is prohibitively time-consuming. Taking advantage of the emerging chip multicore processor (CMP) technology, this article proposes a generalized framework for parallelizing traffic sign detection in a large number of high-resolution video log images. Based on an improved contour finding and workload identification strategy, task and data parallelism in traffic sign detection are fully developed at multiple levels. A generalized parallelization framework for dynamic workload scheduling using adaptive work-stealing of thread pool and dynamic circular lock-free double-ended queue is then proposed. Experimental results on 14,514 images provided by the Louisiana Department of Transportation show that the parallelized traffic sign detection algorithm has great potential to improve computation time with a parallel speedup of up to 18 times on multilevel parallel configurations and different CMP platforms while keeping the same accuracy as the serial version.

Quantitative Performance Evaluation Algorithms for Pavement Distress Segmentation

Algorithms for pavement distress image segmentation are crucial to developing an automatic pavement distress detection and classification system. Many algorithms for pavement distress segmentation have been developed in the past decade; however, the lack of good methods to evaluate their performance quantitatively hinders the focused development of better segmentation algorithms. In this paper, a novel method is developed to quantitatively evaluate the performance of different pavement distress segmentation algorithms. This method uses the buffered Hausdorff distance to estimate the deviation of the cracks in the automatically segmented image from the ground truth cracks. The proposed method captures the local effectiveness of segmentation methods around the crack region without compromising its robustness to isolated pixel deviations caused by noise. Besides real pavement images, synthetic images simulating extreme pavement distress conditions are used to evaluate the capability of the proposed method and show its merits. The proposed method outperforms four other possible quantification methods and demonstrates its superior capability in providing a better score separation to distinguish the performance of different segmentation algorithms.

Mobile Cross-Slope Measurement Method Using Lidar Technology

An effective cross slope facilitates drainage on highways and prevents hydroplaning. There is a need for transportation agencies to identify and measure road sections that have noneffective cross slopes so that timely corrective maintenance can be performed. However, the traditional manual methods used by transportation agencies to measure cross slopes with a digital level are time-consuming and labor intensive; these methods are not feasible for conducting a network-level cross-slope measurement. A proposed mobile cross-slope measurement method uses emerging mobile lidar technology that can accurately and effectively conduct network-level cross-slope measurement at highway speeds. The proposed mobile cross-slope measurement method uses emerging lidar technology (lidar calibration, data acquisition, region of interest extraction, and cross-slope computation). A sensitivity study was conducted to determine the key parameter (i.e., the region of interest interval) for the proposed method. The accuracy and the repeatability of the proposed method were critically validated through testing in a controlled environment. A case study demonstrated the capability of the proposed method. The results from the controlled test show that the proposed method can achieve desirable accuracy with an average measurement difference of 0.088 from the digital-level measurements and a desirable level of repeatability with a standard deviation of less than 0.038 in three runs. The results of the case study show that the proposed method can be operated at highway speed and is promising for the assessment of network-level cross-slope adequacy.

Dynamic Programming and Connected Component Analysis for an Enhanced Pavement Distress Segmentation Algorithm

Automatic pavement distress segmentation is essential for automatic classification and evaluation of pavement conditions during maintenance. Improving the speed and accuracy of the many algorithms for image-based pavement crack segmentation remains a challenge. Although a dynamic programming–based (DP-based) algorithm is more accurate than other crack segmentation methods, its practical use is limited by the required long computation time. A fast algorithm for pavement crack segmentation needs to be developed with the use of DP and multiscale characterization of fundamental crack elements on connected components in grid cells. The proposed algorithm integrated the accuracy of DP and the high speed of grid cell and connected component analyses. Region-based nonuniform background illumination was removed, and the pre-processed image was divided into grid cells. Multiscale characterization of fundamental crack elements was conducted to differentiate the significant crack elements from the noncrack components on the basis of the connected component in the cells on multiple scales. The crack regions of interest were then accurately estimated through the most significant crack elements and fed into the DP-based crack segmentation with the probabilistic scoring function. The proposed algorithm was tested on a diverse set of pavement images, provided by the Georgia Department of Transportation, taken from I-75/85 under varying lighting conditions near Atlanta. A buffered Hausdorff measure was used to quantitatively evaluate the accuracy of the proposed crack segmentation algorithm. Experimental results showed that the proposed algorithm ran three times faster than the original DP-based method while providing the same accuracy. The proposed algorithm is promising for the practical generation of pavement crack maps.

MULTIYEAR PAVEMENT-REHABILITATION PLANNING ENABLED BY GEOGRAPHIC INFORMATION SYSTEM: NETWORK ANALYSES LINKED TO PROJECTS

A pavement-rehabilitation planning system enabled by a geographical information system (GIS) is described; it can perform multiyear projectlinked network pavement-rehabilitation analyses subject to funding availability, minimum performance requirements, and other constraints. The system first uses information on the current and historical project-level pavement-condition evaluation stored in the central database to forecast project performance ratings and distresses. It then determines appropriate rehabilitation methods and costs and, finally, calculates life-cycle costeffectiveness ratios for all projects in the pavement network. With this information, the program performs analyses to determine multiyear minimum funding required to meet prescribed pavement-performance requirements and constraints and to determine optimum pavementrehabilitation plans subject to funding availability and other requirements, such as balancing funding distribution or future pavement performance among state congressional districts or Georgia Department of Transportation (GDOT) districts. The system uses dynamic segmentation to create GIS maps, links them with the central database and network analysis results, and thus allows users to make changes to the rehabilitation plans directly on the GIS maps and have the changes reflected automatically in the database. Several examples using the actual data on historical pavement condition evaluations from GDOT are presented to illustrate the capabilities of the system.

Pavement Crack Detection Using High-Resolution 3D Line Laser Imaging Technology

With the advancement of 3D sensor and information technology, a high-resolution, high-speed 3D line laser imaging system has become available for pavement surface condition data collection. This paper presents preliminary results of a research project sponsored by the U. S. Department of Transportation (DOT) Research and Innovation Technology Administration (RITA) and the Commercial Remote Sensing and Spatial Information (CRSS however, the data resolution limits the detection of hairline cracks to approximately 1mm. The findings are crucial for transportation agencies to use when determining their automated pavement survey policies. Recommendations for future research are discussed in the paper.