A Simplified Computer Vision System for Road Surface Inspection and Maintenance

• Published in: IEEE transactions on intelligent transportation systems Vol. 17; no. 3; pp. 608 - 619
• Main Authors: Quintana, Marcos, Torres, Juan, Menendez, Jose Manuel
• Format: Journal Article
• Language: English
• Published: New York IEEE 01-03-2016; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Cameras; Classification; Computer vision; Cracks; Feature extraction; Flaw detection; Hough transforms; image processing; Lighting; pattern recognition; Road safety; Roads; Surface cracks; Texture; Vehicles; Vision systems; image processing; Road safety; computer vision; pattern recognition
• ISSN: 1524-9050; 1558-0016
• DOI: 10.1109/TITS.2015.2482222

This paper presents a computer vision system whose aim is to detect and classify cracks on road surfaces. Most of the previous works consisted of complex and expensive acquisition systems, whereas we have developed a simpler one composed by a single camera mounted on a light truck and no additional illumination. The system also includes tracking devices in order to geolocalize the captured images. The computer vision algorithm has three steps: hard shoulder detection, cell candidate proposal, and crack classification. First the region of interest (ROI) is delimited using the Hough transform (HT) to detect the hard shoulders. The cell candidate step is divided into two substeps: Hough transform features (HTF) and local binary pattern (LBP). Both of them split up the image into nonoverlapping small grid cells and also extract edge orientation and texture features, respectively. At the fusion stage, the detection is completed by mixing those techniques and obtaining the crack seeds. Afterward, their shape is improved using a new developed morphology operator. Finally, one classification based on the orientation of the detected lines has been applied following the Chain code. Massive experiments were performed on several stretches on a Spanish road showing very good performance.