The classical moving surface filtering algorithm demonstrates a significant effect of filtering on various terrains. However, the gross errors still exist when the lowest point of the grid is used as the ground point in the moving surface grid algorithm. This study proposes a confidence interval test method to select the best initial seed points by using the residual, mean square error, and confidence probability as the reference values. The grid overlap method is used to solve the problem of the fracture layer between adjacent grids. Moreover, the hierarchical clustering adaptive threshold determination method is used to determine the elevation difference threshold. For the condition where special grids with insufficient special seed points cannot fit the surface, planes are established and thresholds are set to determine whether the plane can meet the condition. This study compares the improved moving surface filtering algorithm with the classical moving surface filtering algorithm by qualitative and quantitative experiments. The findings demonstrate that the improved moving surface algorithm can obtain a better filtering effect than the classical moving surface algorithm.