Abstract: Remote real-time monitoring of surface subsidence is a crucial technical requirement for managing engineering risks, such as surface collapse and geological landslides, induced by underground engineering. To address the key challenges of complex environmental impacts on accuracy and high real-time requirements in surface subsidence monitoring tasks, a surface subsidence analysis method integrating digital image correlation (DIC) technology and deep learning is proposed. First, YOLO V8 and self-developed targets are used to realize multi-target positioning and matching, and DIC is adopted to achieve precise tracking of measurement points. Then, a sub-pixel search algorithm based on quadratic surface fitting is used to improve monitoring accuracy, and a homography transformation method is applied to calculate real displacement values. Based on the proposed algorithm, a computer vision remote real-time monitoring system for surface subsidence is constructed, which establishes a complete monitoring system integrating remote image acquisition, real-time image transmission, real-time displacement calculation, data management, and risk early warning. This system is both practical and scalable. Simulation tests show that the sub-pixel search algorithm can improve monitoring accuracy; the homography transformation method effectively reduces monitoring errors caused by excessive angles between the camera axis and the target surface. The proposed method can avoid occlusion and track large displacements accurately. Field tests demonstrate that compared with traditional DIC algorithms, the proposed method exhibits stronger adaptability and robustness in complex natural environments such as strong light and heavy fog. These findings contribute positively to the advancement of new technologies for surface subsidence monitoring in engineering disaster management.