Abstract: Landslides and other geological hazards are characterized by their sudden onset and complex triggering mechanisms,making accurate prediction particularly challenging in complex geological environments such as fractured rock masses and unconsolidated soil deposits. Such events frequently result in substantial casualties and economic losses. Therefore,the development of real-time,high-precision landslide deformation monitoring technology is of considerable practical importance. This study proposes an algorithm that integrates machine vision and centroid analysis,and based on this approach,a software system has been developed with capabilities for intelligent deformation feature recognition,real-time displacement monitoring,and early warning. The system enables automated acquisition,continuous processing,and intelligent analysis of video data from landslide monitoring points. It accurately computes displacement vectors using high-performance image processing algorithms and dynamically outputs motion trajectory diagrams of feature points. As no significant deformation has been observed in the study area to date,indoor physical landslide model tests were conducted to validate the system's recognition accuracy,timeliness,and reliability. A comparative analysis of identification results under different simulated deformation scenarios was performed. The test results demonstrate that the system can stably and efficiently capture the complete landslide deformation process,accurately identify deformation characteristics,including displacement direction,magnitude,and movement patterns of key points,and exhibits strong anti-interference capabilities against factors such as lighting variations,rainfall,and fog. It also achieves near-real-time response performance. This system provides a cost-effective technical solution for the early identification and continuous tracking of landslide hazards,while also supplying valuable data for risk assessment and disaster prevention decision-making.