One of the challenging problems in the area of automatic interpretation of the pavement condition videos or images is detecting the presence of low- to medium-severity cracking when the mean crack width is less than 1/4 in. This problem is further complicated by the texture of the background pavement. The variations in light intensity between aggregates and bituminous material are similar to that of the actual cracking (noisy images). The traditional approach to solving this problem includes three techniques. The first is to increase the resolution of the video camera by using a higher resolution camera or alternatively by recording the pavement image on super VHS tapes. The second technique that is often used in noisy images is applying a low-pass filter to remove the pixels with rapid intensity variations between them (inevitably reducing the sharpness of edges). The third technique involves the use of a localised edge detector by which local edge points can be detected in small subareas. The shortcoming of the third technique is that it includes isolated noisy spots as edge points. Development of an innovative technique on edge detection is based on the Sobel operation coupled with Kittler's automatic thresholding and a sequence of postprocessing operations. Image examples of thin cracks on asphalt concrete and PCC surfaces are used to demonstrate the capability of this image analysis technique.
Abstract