Graduation Year


Document Type




Degree Granting Department

Civil and Environmental Engineering

Major Professor

Manjriker Gunaratne

Co-Major Professor

Sudeep Sarkar


Digital image based automated pavement crack detection and classification technology has seen vast improvements in the recent years. In spite of these developments, although pavement crack lengths and widths can be evaluated using state-of-the-art software with a reasonable accuracy, no reported evidence is found in extending this technology to evaluate crack depths. Hence a preliminary study was carried out to model the digital image formation of cracked concrete pavements based on the Bidirectional Reflection Distribution Function. It was revealed that a definitive theoretical relationship exists among the crack widths and depths and the maximum pixel intensity contrasts in the images of cracks. The above relationships fortified by appropriate calibration were verified using actual crack data not used in the calibration that can be useful in predicting crack depths. Secondly, a number of innovative techniques in computer vision such as image characterization using quantification of optical texture properties of images and a number of widely used optical texture related techniques for characterization of digital images which have not been exploited adequately in pavement evaluation, were introduced highlighting their useful applications in pavement evaluation. One such application, the automated and accurate detection of correspondences in progressive images of the same pavement captured during different times, would be essential for close monitoring of cracks or wear at the project-level. Two reliable methods for determining correspondences among pavement images illustrated in this work are; (1) texture masking and minimum texture distance method applicable to locations with no significant distress, and (2) homogeneous coordinates based geometrical matching and the maximum texture distance to detect the locations of distress and be applied to detect exact locations of crack propagation and excessive pavement wear. Thirdly, the BRDF based pavement image formation model revealed that quantifiable changes in the brightness of images occurs due to pavement wear-related changes in texture depth and spacing (wavelength). The traffic induced pavement wearing process was simulated by gradual smoothening of the modeled surfaces and then images corresponding to each wearing stage were generated. The theoretically predicted variation of the image brightness due to wear was experimentally verified using images from a gradually worn out concrete specimen. Finally it was illustrated how the brightness evaluation of wheel path images has the potential to be a screening tool to monitor the degradation of macrotexture and hence the skid-resistance of pavements at the network level.