Reduced Featured Based Projective Integral for Road Cracks Detection and Classification

This paper presents an enhanced and robust approach to detect and classify pavement cracks from captured images. The approach was based on three stages: pre-processing, feature extraction and classification. In pre-processing, we carried out several algorithms to compensate the impact of quality distortions during image acquisition. Then, features are retrieved from projective integrals computed on edge images. These features fed machine learning algorithms to classify the type of crack that may appear in a pavement image. The obtained results proved the relevance of our reduced features. We achieved the best successful classification rate of 93.4% using the Support Vector Machine (SVM) classifier and an accuracy of 94.7% for crack detection.