This project was aimed at developing and demonstrating the application of an asphalt pavement raveling detection system based on algorithms using three-dimensional (3D) laser technology and macrotexture analysis. The capabilities of the developed algorithms include (1) data pre-processing to remove data outliers, detect pavement markings and edge drop-off, and extract the candidate pavement portion for raveling detection; (2) computation of each subsection with the newly developed 811 features for raveling analysis (each 3D pavement data file covering a 5-m pavement section divided into six sub-sections); (3) raveling classification using Random Forest models, a supervised learning technique with the known raveling classification as the learning samples; (4) post-processing to aggregate the six sub-section-based raveling classification outcomes for determining the raveling severity level for each 5-m pavement section; and (5) aggregation of the 5-m pavement section raveling to measure and report the raveling condition, including percentage and severity level, at the segment level (normally 1 mile long, based on highway agencies’ survey practices). The developed algorithms were tested and validated using Georgia Department of Transportation (DOT) pavement condition survey protocol on Highways I-85 and I-285 near Atlanta, Georgia. The 3D pavement data were collected on four test sections on I-85 (each 1 mile long) and on the entire outer lane of asphalt pavement (61 miles) on I-285. The automatic classification results on each of the test sections were compared with the ground truth (those measured by Georgia DOT pavement experts). The results showed the developed algorithms to be very promising. Tests on I-285 also showed promising results for automatic raveling detection, classification, and measurement. All pavements (with or without raveling) were 100% correctly detected and classified at the segment level (each segment 1 mile long). However, owing to the difficulty of correctly labeling all the raveling areas using videolog images and 3D pavement data and to the impact of cracking and flat-tire scratches, the raveling extent showed a certain level of variation in comparison with the manually labeled ground truth. The differences between the surveyed results by Georgia DOT pavement experts and the automatically detected and measured results were less than 15% (most actually less than 10%). Although the developed algorithms show promising capabilities for automatically detecting and measuring asphalt pavement raveling, further field evaluation is needed for implementation of the method by the DOTs.
The final report is available.