Automated Pavement Patch Detection and Quantification Using Support Vector Machines

Abstract

Pavement condition evaluation provides transportation authorities with decision support tools for the selection of appropriate repair or replace actions, thus preventing the possibility of transportation networks disruption. The current costly, time-consuming, and subjective pavement assessment tactics require automation, combined with the application of low-cost technologies for more widespread deployment. Presented herein is an automated vision-based method for detecting and quantifying pavement patches; a critical aspect of pavement surface valuation and rating. The proposed system uses road surface video frames acquired either by a smartphone or an external camera, positioned respectively inside and outside of a moving passenger vehicle. Support vector machine classification applied to feature vectors, generated from the image and defined by two texture descriptors plus the histogram of nonoverlapped square blocks, characterized image blocks as parts patch or no-patch areas. The output consists of block-based and image-based classifications, while applications of the method to test video frames demonstrates a detection accuracy of 87.3 and 82.5% respectively. Additionally, the patch area is quantified with a percent absolute error of 11.4%.