Advances in Research

In recent years, with advances in computer vision, vision-based approaches for real-time landslide monitoring and early warning have become a research focus. Building on OpenCV-Python, this study evaluates three representative moving-object detection algorithms—frame differencing, background subtraction, and optical flow. Centering on key characteristics such as tracking the number of rock fragments before and after a landslide and performance in complex environments, we construct an experimental comparison framework using landslide video sequences to systematically assess each algorithm’s adaptability, accuracy, and practical deployment potential in early-warning scenarios. The results show that frame differencing achieves the fastest response and relatively high detection accuracy, making it suitable for rapid warnings in high-risk areas, though it risks missed detections in cluttered backgrounds; background subtraction is more sensitive to small deformations and is appropriate for applications that require monitoring the detailed evolution of landslides; and although optical flow can characterize motion trajectories and is useful for trend analysis, its detection accuracy is limited under complex backgrounds. Finally, this paper proposes a scheme that works together based on the respective advantages of the frame difference method, background subtraction method, and optical flow method. This work examines the applicability of multiple moving-object detection methods to landslide early warning and aims to inform technology selection and optimization for real-time geological-hazard monitoring systems.