Team developing AI-enabled drones for pothole, crack detection
Marie Donlon | April 15, 2019
Source: HKUST Robotics InstituteA team of scientists from the Hong Kong University of Science and Technology (HKUST) Robotics Institute have detailed plans for AI-enabled drones that can inspect roads for potholes and other damage.
The team embedded AI into a quadcopter for the purpose of road inspection and used a stereo vision system — an AI algorithm capable of extracting 3D-depth data from images — to capture reference views. Comparing the differences between reference images and real-time images of the road generated a disparity map that was input into a mathematical function that made the damaged areas easier to see. The team mounted a stereo camera on a DJI Matrice 100 drone to capture the images of the road and then processed the images using a PC with a Nvidia Jetson TX2 graphics card.
During testing, the team generated three datasets that added up to 11,368 stereo image pairs composed of one original reference image and one target image both at 640 x 360 resolution. The images were then compared to both synthesized and real data sets of cracks, potholes and other road damage to improve accuracy.
The researchers concluded that the damaged road areas became “highly distinguishable” according to the disparity maps. “[T]his can provide new opportunities for UAV-based road damage inspection,” they wrote. “In the future, we plan to use the obtained disparity maps to estimate the flight trajectory of the [drone] and reconstruct the 3D maps using the state-of-the-art simultaneous localization and mapping (SLAM) algorithms.”
“The frequent detection of different types of road damage, e.g., cracks and potholes, is a critical task in road maintenance,” the researchers wrote, explaining that most inspections are still performed manually.
“[M]anual visual inspection [is] not only tedious, time-consuming, and costly, but also dangerous for the personnel. Furthermore, the detection results are always subjective and qualitative because decisions entirely depend on the experience of the personnel…With recent advances in airborne technology, unmanned aerial vehicles (UAVs) equipped with digital cameras provide new opportunities for road inspection.”
The research is detailed in the article “Real-Time Dense Stereo Embedded in A UAV for Road Inspection,” which appears in Arxiv.
Web-enabled cars with GPS receivers could self-report any significant pavement malady. No AI required, just some accelerometers on the suspension.
There are dozens of 'speed bumps' on the Maryland side of the Washington D.C. beltway between the Potomac River and Andrews Air Force Base. I wish the state would get out there and grind those damn things down.
See the article April brings showers -- and pothole repairs for a more comprehensive discussion of pothole science, if I may call it that.
Pothole-spotting sensors have already been deployed in some Ford vehicles. A bunch of cities still use old-fashioned eyeballs: they ask citizens to report potholes and manage this information in a much more sophisticated manner than in the past. by using data mining and analysis road maintenance departments can predict where to expect pothole problems and apply preventative maintenance.
A UK team has also been experimenting with drone use to spot *and* repair potholes, using an onboard 3D printer to produce the patch.