Abstract: In order to improve the speed of response to landslides and the efficiency of early warning, reduce equipment losses and costs, and ensure the safety of operators, a method of monitoring and early warning using images was proposed.Taking the displacement of the image point with the same name in each time period as the main criterion, combined with the comprehensive evaluation of parameters such as the gradient direction angle and the adjacent weighted average distance between the adjacent feature points.Establish a Gaussian distribution model to verify whether the pivot amount formed by the displacement of the same name image point in each period exceeds the confidence intervalso as to determine whether the displacement is affected by the gross error.If it is not due to gross errors, the second step is to count the weighted average distance between adjacent points and the weighted average of displacement in the two sets of points with the largest number of angles in the gradient direction. The two reasonable thresholds are set according to photo-grammetry technical specifications and practical experience.When the weighted average distance between adjacent points exceeds the threshold, it means that the relative positions of the moving feature points are relatively scattered.Similarly, the weighted average value of the displacement exceeds the corresponding threshold, indicating that the target has obviously slipped and there is a potential danger.Comprehensive analysis of the above two conditions can automatically distinguish whether the movement of the target is accidental decentralized man-made damage or real landslide, and make corresponding early warning measures.The results show that the mathematical model constructed by this method is more reliable,it can accurately identify the real landslide and saves the long time cost of calculating the coordinates with multi-view images. This method can be applied to emergency monitoring of sudden landslides.