Automated Measurements of Road Cracks Using Line-Scan Imaging

This paper presents a new line-scan imaging system for automated measurements of road crack. The system consists of off-the-shelf hardware for real-time image acquisition and the customized image-analysis software for crack detection. A line-scan camera with 2 k pixels, a GigE interface, and a line rate up to 36 kHz was used to scan 3.6-m wide pavements at highway speeds, and a laser line projector was used to cast a transverse beam that overlays the scanline of the camera to eliminate shadows of vehicles and roadside objects and to maintain consistent lighting conditions. In the crack detection algorithms, a pavement image was first divided into grids of 8×8 pixels, and each grid was classified either as a non-crack or crack grid (called seed) using the pixel information of the grid and the overall background. Then, seeds in the vicinity were connected based on geometrical and intensity constrains. The connected seeds served as a candidate for a crack, which were further verified by using the contrast to the pixels along its trace. The paper also reports the experimental results on a designated pavement that was manually rated by an expert, and scanned three-times by the system. The statistic analysis showed that the difference in crack length between the manual and automatic measurements was less than 10 %, and no significant difference among the multiple scans by the system.