The Problem:
Currently laser scanning technology has been used in the Engineering and Transportation industries to survey roads and monitor surface conditions. However, existing software that scans for road cracking requires a time-consuming pre-filtering of each data point to cut out false-identification of cracks. Also, current technology does not account for special cases such as sudden changes in elevation and man-made grooving. Additionally, existing practices involve manual labor, making them more time-consuming and subjective.
The Solution:
Researchers at the University of Alabama have developed an algorithm utilizing a deep-learning convolutional neural network and data fusion (both intensity and range images) in order to accurately and efficiently identify cracks in roads at pixel-level resolution. This unique algorithm does not require pre-filtering of each data set, which yields greater accuracy and reduces manual labor. The model detects cracks with approximately 99% accuracy, an improvement on current methods by at least 3%.
Benefits:
• Higher accuracy results • Less time and manual labor required • More cost-effective and time-efficient assessment than existing laser scanning systems • Prevents false positive crack identification • Offers important insights for maintenance practices