The National Academies

NCHRP IDEA 20-30/IDEA 169 [Completed (IDEA)]

An Inexpensive Vision-Based Approach for the Autonomous Detection, Localization, and Quantification of Pavement Defects
[ NCHRP 20-30 (NCHRP-IDEA) ]

  Project Data
Funds: $124,999
Staff Responsibility: Dr. Inam Jawed
Research Agency: University of Southern California
Principal Investigator: Burcin Becerik-Gerber
Fiscal Year: 2013

The goal of this project was to develop and demonstrate the application of an imaging system based on inexpensive sensors for an automated detection and quantification of pavement defects, including cracks and potholes. The approach used off-the-shelf Microsoft Kinect sensors, costing less than $200 each, to collect color images and 3D point clouds of roadway surfaces. A compact-size pavement data collection system was built that could be easily installed on a car and collect data at highway speeds. The system used multiple Microsoft Kinect sensors to cover a lane width and was designed to reach scanning speed. It also included 3-axis accelerometers to record orientations of the system and GPS to obtain location and velocity. Several road tests were performed on local streets and freeways. The tests presented a few challenges. The main challenges were sunlight interference, motion blur, and rolling shutter distortion. A top-cover was designed to reduce the sunlight interference. A stroboscopic technique was used to solve the motion blur problem for Kinect’s color image acquisition and capture slow motion pictures.  Both the color camera and depth camera of a Kinect sensor consisted of rolling shutter CMOS image sensors that provided low-noise, low-power, fast data processing; however, this image acquisition method created distortions when shooting moving objects. A rectification algorithm was developed to correct distorted images  These improvements enabled the pavement data collection system to obtain good imaging results when moving at less than 30 mph (residential speed limit in most states). Furthermore, pavement crack detection using a hybrid algorithm based on anisotropic diffusion filtering and Eigen analysis of Hessian matrix showed a promising outcome in segmenting the cracks as compared with a modified bottom-hat morphological method. Discussions with FHWA, Fugro Consultants, and the California DOT have been initiated to evaluate and implement the developed technology in real-world applications.

The final report is available.

To create a link to this page, use this URL: http://apps.trb.org/cmsfeed/TRBNetProjectDisplay.asp?ProjectID=3529