ABSTRACT

In the current paper a data-driven framework and related field studies on the use of Artificial Intelligence (AI), pattern recognition and smartphone sensor technologies for the detection, classification and georeferencing of roadway pavement surface anomalies are presented. The smartphone-based data collection is complimented with artificial neural network techniques, robust regression analysis and bagged-trees classification models for the classification of detected roadway anomalies. The proposed system is low-cost and sufficiently accurate, and can be used in crowd-sourced applications for pavement surface monitoring. Further, the proposed method and system architecture utilize four metrics in the analysis and have been field-tested for the detection and classification of patches (transverse and longitudinal) and potholes, exhibiting accuracy levels higher than 90%. Currently, the method is expanded to include larger datasets and a bigger number of pavement defect types (such as cracks and raveling).