This paper summarizes the important research achievements of pavement defect automatic detection technology, and analyzes the research progress of key technologies in this field, including pavement defect data acquisition system, pavement defect automatic recognition method and pavement defect automatic classification method. The results show that: Firstly, in terms of data acquisition methods, the research of pavement defect detection has experienced the development process from two-dimensional camera measurement acquisition system to three-dimensional laser scanning acquisition system. The three-dimensional structured light scanning technology has stronger environmental adaptability and higher recognition accuracy; Secondly, the authors first reviewed 78 machine Learning (ML) based defect detection methods to identify the current trend of development, pixel-level crack segmentation. However, different recognition algorithms still lack of scientific performance index evaluation methods. In the future, a set of objective evaluation system of defect recognition algorithm should be established, in which the important elements should include consistent performance evaluation indexes and pavement defect data sets with different road environment. Thirdly, the corresponding supervised classification methods and unsupervised classification methods are also summarized for the automatic classification methods of pavement cracks. The accuracy of defect classification largely depends on the integrity of the data source and automatic recognition. With the improvement of intelligent level, it can also be predicted that the game between the deep learning (DL) based defect segmentation models and distress characteristics-based defect image classification models will become the research hotspot of the next generation detection technology.

road engineering; 3D Pavement; automatic recognition; deep learning; distress classification.