3DCrack

We introduce 3DCrack, a new 3D pavement image dataset to support deep learning-based crack detection. Data was collected using the Georgia Tech Sensing Vehicle (GTSV), equipped with dual 3D line laser sensors. These sensors capture a 4-meter-wide pavement profile at speeds up to 60 mph. Each frame produces a dense 1,000 \u00d7 2,080 point cloud, with 1 mm \u00d7 1 mm resolution after interpolation, and 0.5 mm height precision. The dataset covers diverse cracking conditions, including varying geometries such as transverse, longitudinal, alligator, and compound cracks, as well as challenges like pavement joints and other visual distractions, ensuring robustness in trained models across real-world scenarios.