Seed-Based Approach for Automated Crack Detection from Pavement Images

Abstract

An accurate and reliable pavement crack detection system plays an important role in evaluating pavement condition and providing needed information for decision making for pavement maintenance and rehabilitation. Among existing crack detection methods, the seed-based image segmentation method has proved to be fast and efficient for automated crack detection. However, its performance is not stable under varying conditions. This paper proposes an extended and optimized seed-based crack detection method after an extensive review of current practices. The proposed method included two main steps. In the first step, pavement images were preprocessed. Lane marking was masked to be a noncrack area and the nonuniform background of the images was corrected. In the second step, crack seeds were detected through grid cell analysis and then connected through a Euclidean minimum spanning tree construction. In addition, undesirable small objects, such as minimal branches and noises, were removed by a path length–based removal method. The proposed algorithm was evaluated by using 105 pavement images collected with the Texas Department of Transportation VCrack system. The experiment results showed that the proposed method could accurately and efficiently detect cracks in the images.