Abstract: Machine vision technology has significant advantages in joint and fracture detection on tunnel faces due to its simplicity and efficiency. This paper reviews the research progress of two mainstream methods: digital image processing and deep learning. First, the specific algorithms applied in each stage of the “image preprocessing-joint and fracture segmentation-skeleton extraction” workflow using digital image processing methods are elaborated, and their limitations in complex tunnel environments are summarized. Second, the application scenarios of commonly used classification and semantic segmentation models in deep learning-based methods are analyzed. Then, the calculation method of real joint size on tunnel faces is discussed, together with the computation of joint occurrence parameters, trace length, spacing, and grouping. Finally, future development directions are proposed, including the adoption of semantic segmentation models as the core framework, integration of skeletonization techniques from digital image processing, enhancement of 3D reconstruction theory, and improvement of the efficiency and accuracy of 3D joint feature extraction, so as to better support tunnel construction.