Sources: Dwight Day, 785-532-4660, firstname.lastname@example.org;
and Bala Natarajan, 785-532-4597, email@example.com
Photos available. Contact firstname.lastname@example.org or 785-532-6415.
News release prepared by: Erinn Barcomb-Peterson, 785-532-6415, email@example.com
Thursday, Oct. 29, 2009
K-STATE ENGINEERS DEVELOP TOOLS THAT HELP KANSAS DEPARTMENT OF TRANSPORTATION DETECT ROAD CRACKS, LEADING TO SMOOTHER, SAFER TRAVEL
MANHATTAN -- To help the Kansas Department of Transportation smooth cracks in the state's roadways, Kansas State University engineers had to pave the way for technology that can automatically identify the cracks in the first place.
Bala Natarajan and Dwight Day are K-State associate professors in the department of electrical and computer engineering. They just wrapped up a three-and-a-half-year project supported by the Kansas Department of Transportation to develop imaging technology that can be used to automatically identify and classify roadway cracks. They also refined how road crack data are presented to transportation officials.
Joining them in the research were Krithika Rajan, May 2008 graduate in electrical engineering, and Anirudh Radhakrishnan, master's student in electrical engineering. Rick Miller, assistant geotechnical engineer at the Kansas Department of Transportation, served as the program manager and played an active role in guiding the research team.
Like in Kansas, many state transportation departments have spent years collecting roadway data by traveling highways in vans equipped with sensors that measure bumps in the road and cameras that take images of the surface below. Day said a lot of people have worked on ways to better process these images using a computer capable of detecting defects.
"The human visual system has an amazing ability to take a whole image and focus in on the parts that it needs," Day said. "Humans are unbelievably adept at taking in the whole view and being able to go into the smaller view. The computer nearly always works just on that super-small piece, because that's all it can really fathom."
What the computers needed, he said, was a way to detect a crack on the small scale and convert the data into a single measurement of how bad the road is. This would allow transportation departments to make more informed decisions about which repairs should be given priority.
Natarajan said the K-State team came up with a novel algorithm that combines basic image processing techniques with sensor fusion principles. Given good lighting conditions in the images, the algorithm developed at K-State can detect 95 percent of all cracks.
"We go pixel by pixel in the image, trying to take a stab at whether a pixel is showing a crack or not," Natarajan said. "Usually cracks have some patterns, so we employ morphological operations to see if there is a sequence of cracked pixels. Then, we try to look at subimages and classify them as to whether they belong to a crack or not."
To determine this, Natarajan said the engineers looked at different features of that subimage and used sensor fusion techniques to clean up any false positives. And then, he said, they got a clear picture of where the crack was located.
Previously, Day said, the data were presented to transportation officials as a series of tables, so that the highways weren't mapped out but rather were listed by their number. So the engineers also worked on automated report generation, making it easier for transportation officials to visualize highway data and see exactly where the cracks are on the road. The engineers then went a step further and interfaced this system with Google Earth.
More recently, the engineers assessed standards that are being developed by the American Association of State Highway and Transportation Officials, known as AASHTO. They compared the performance of imaging technology with manual assessments done by trained raters. The results helped refine the association's standards.
Both professors said the project has changed their driving experiences.
"Every time I drive now, I'm looking for cracks," Natarajan said.
The research produced a paper, published in Proceedings of 2008 International Conference on Image Processing, Computer Vision and Pattern Recognition.