An Inexpensive Vision-Based Approach for the Autonomous Detection, Localization, and Quantification of Pavement Defects

This project will develop and demonstrate the application of an imaging system based on inexpensive RGB-D sensors for automated detection and quantification of pavement defects, including cracks and potholes. Work in Stage 1 will focus on developing a hardware system for data acquisition of full lane width with traffic flow, and synchronization of the heterogeneous collected data. The proof-of-concept system with the RGB-D sensor has the limitation of capturing 30 frames per second which restricts the current prototype to collect data at low velocity (about 35 miles per hour). To address this issue, a number of these sensors in parallel will be used. Furthermore, the current proof-of-concept prototype inspects only about two feet width of the road in one pass rather than the full lane width (about 12 feet). To address this issue, a sensor array will be employed and data collected by the array sensors will be synchronized. The developed system will be mounted on a mobile unit and tests will be conducted to examine the nominal and practical speed under which the system is able to perform effectively with a goal to collect data accurately and reliably at a speed of at least 50 mile per hour. Work in Stage will focus on developing an interpretation system for extracting useful knowledge from the sensed data. Towards this end, defect detection and quantification modules will be developed to process the collected data. Data acquired from the depth image will be fused with data gained from the RGB image. Defect detection modules capable of detecting a variety of pavement defects will be developed along with algorithms to estimate the length, width, mean-depth, maximum-depth, area, volume, and orientation of the detected defects. To improve the accuracy of the acquired data, the sensors will be calibrated to accurately capture the depth information. The developed modules (such as defect detection and quantification modules) will be rigorously tested with field data. To evaluate the defect-detection module, several RGB and depth images of various types of defects as well as defect-free frames will be captured with the proposed sensing system. The collected RGB-D frames will be processed to classify frames as defective or defect-free. A human operator will manually label the frames as defective or defect-free to be used as ground truth. Several performance indices, including accuracy, precision, sensitivity and specificity, will be computed to quantitatively evaluate the capabilities as well as the limitations of the defect-detection module. To evaluate the defect-quantification module, several characteristics of the defective regions will be manually measured and used as the ground truth. The results from the quantification module will be compared to the ground truth to report errors and uncertainties of the estimated characteristics. The final report will provide all relevant data and guidelines and specifications for using the new pavement defect detection system as well as plans for its implementation and commercialization.


  • English


  • Status: Active
  • Funding: $124999.00
  • Contract Numbers:

    Project 20-30, IDEA

  • Sponsor Organizations:

    Federal Highway Administration

    1200 New Jersey Avenue, SE
    Washington, DC  United States  20590

    American Association of State Highway & Transportation Officials (AASHTO)

    444 North Capitol Street, NW, Suite 225
    Washington, DC  United States  20001

    National Cooperative Highway Research Program

    Transportation Research Board
    500 Fifth Street, NW
    Washington, DC  United States  20001
  • Project Managers:

    Jawed, Inam

  • Performing Organizations:

    University of Southern California, Los Angeles

    University Park Campus
    Los Angeles, CA  United States  90089
  • Principal Investigators:

    Becerik-Gerber, Burcin

  • Start Date: 20141210
  • Expected Completion Date: 0
  • Actual Completion Date: 0
  • Source Data: RiP Project 38322

Subject/Index Terms

Filing Info

  • Accession Number: 01547528
  • Record Type: Research project
  • Source Agency: Transportation Research Board
  • Contract Numbers: Project 20-30, IDEA
  • Files: TRB, RIP
  • Created Date: Dec 11 2014 1:00AM