Crack detection is critical to the safety and performance of engineering structures including highways, bridges, and buildings. This paper reviews the different types of image processing techniques for crack detection which includes traditional, advanced and machine learning based methods. One of the sixty works chosen and written from 1999–2023. The results indicate that machine learning is preferred simply because it can be more precise, given its automated nature. The paper reviews different methodologies applied, their performances based on the datasets employed as well as limitations governing one approach over another to realize real-time crack detection. It also highlights the need for standardized methods of categorizing and measuring cracks, as well as efficient ways to proceed in many different situations. Specifically, the study addresses potential benefits and limitations of using machine learning in infrastructure management with respect to data-driven approaches linked to big data sets as well as computing costs.