Pavement Defects Detection and Classification Using Smartphone-Based Vibration and Video Signals

摘要

Presented herein is a big-data driven methodology for the detection of roadway anomalies, utilizing smartphone-based data and image signal streams. The methodology uses a vibration-based method and artificial intelligence for the detection of vibration-inducing anomalies, and a vision-based method with entropic texture segmentation filters and support vector machine (SVM) classification for the detection of patch defects on roadway pavements. The presented system pre-processes video streams for the identification of video frames of changes in image-entropy values, isolates these frames and performs texture segmentation to identify pixel areas of significant changes in entropy values, and then classifies and quantifies these areas using SVMs. The developed SVM is trained and tested by feature vectors generated from the image histogram and two texture descriptors of non-overlapped square blocks, which constitute images that includes "patch" and "no-patch" areas. The outcome is composed of block-based and image-based classification, as well as of measurements of the patch area.