Image Processing Techniques In Automated Pavement Crack Detection And Identification

Pavement surface crack detection plays a crucial role in roads maintenance, allowing accordingly a higher quality in transportation services. Since manual inspections through visual surveys by a road engineer are time consuming, expensive, subjective and unsafe, an automatic solution is proposed using digital image acquisition and digital image processing. This dissertation proposes an automatic system to identify cracks and sealed cracks in pavement surface images from a road pavement survey image database. The speed and efficiency of road surface analysis are improved while the road engineer’s effort and subjectivity of the achieved results are reduced. For this purpose, first a preprocessing algorithm is developed to improve the non-uniform background of collected pavement surface images and enhance the cracks and sealed cracks on the pavement surface. The pavement markings are subsequently removed using simple pixel value threshold on the pavement surface images. Second, since sealed cracks and cracks have the same characteristic, a sealed crack detection algorithm is developed to identify sealed cracks frame using developed heuristic thresholding method. Finally, the crack detection algorithm is developed based on a weighted neighborhood pixel. Seven different patterns are investigated and compared to find the best pattern. Consequently, a local threshold approach and shape filtering using eccentricity value parameter are applied to enhance the candidate cracks. Also, crack fragments are connected using dilation operator. The crack frames are identified from the acquired images using new thresholding method.The developed algorithm is evaluated through several cases studies, namely, 1-270,1-81, and 1-95 Interstate Highways in Maryland. The evaluation is based on well-known metrics such as recall, precision and accuracy, exploiting the availability of ground truth data that is manually provided for all the images. The obtained results compared with the ground truth images show high recall, precision and accuracy. In other words, the proposed algorithm can consistently identify the cracks and sealed cracks regardless of their types or noise background.