NHA12D: A new pavement crack dataset and a comparison study of crack detection algorithms

This paper conducts a comparison study to evaluate the performance of state art crack detection algorithms quantitatively and objectively. A more comprehensive annotated pavement dataset (NHA12D) that contains images with different viewpoints pavements types is proposed. In study, were trained equally on largest public collected evaluated proposed (NHA12D). Results: models generally fail distinguish cracks from concrete joints. Detecting still has huge room for improvement. Domain adaptation can be used boost unseen images.