Improvement of Crack-Detection Accuracy Using a Novel Crack Defragmentation Technique in Image-Based Road Assessment

Abstract

A common problem of crack-extraction algorithms is that extracted crack image components are usually fragmented in their crack paths. A novel crack-defragmentation technique, MorphLink-C, is proposed to connect crack fragments for road pavement. It consists of two subprocesses, including fragment grouping using the dilation transform and fragment connection using the thinning transform. The proposed fragment connection technique is self-adaptive for different crack types, without involving time-consuming computations of crack orientation, length, and intensity. The proposed MorphLink-C is evaluated using realistic flexible pavement images collected by the Florida Department of Transportation (FDOT). Statistical hypothesis tests are conducted to analyze false positive and negative errors in crack/no-crack classification using an artificial neural network (ANN) classifier associated with feature subset selection methods. The results show that MorphLink-C improves crack-detection accuracy and reduces classifier training time for all 63 combinations of crack feature subsets that were tested. The proposed method provides an effective way of computing averaged crack width that is an important measure in road rating applications.