Novel Feature Extractions for Reflection, Alligator Cracks and Potholes Road Surface Classification

摘要

Road surface inspection for cracks, distortion, and disintegration-together with appropriate surface treatments-are mandatory in maintaining the ride quality and safety of the highways. Due to especially high occurrences of `reflection', `alligator cracks' and `potholes' in Thailand, and the fact that they require markedly different treatment methods, a classifier that can distinguish among those two types of bad surface is most desirable.This paper proposed two novel feature extractions based on regional profiling and Cartesian profiling of orthogonal axes features which worked well with this particular problem, with added benefit of decoupling feature extraction from the classifiers themselves.The experimental results showed that Cartesian profiling of orthogonal axes features works well with Decision Tree (DT), and regional profiling works well with Support Vector Machine (SVM) achieving F-measures of 0.877 (0.864 Recall) and 0.875 (0.873 Recall) respectively.