Aiming at the problem that the accuracy of terrain simulation of initial irregular triangulation in traditional progressive triangulation filtering algorithm depends on the selection of initial seed points, an improved progressive TIN filtering algorithm is proposed. Firstly, more initial ground seed points are obtained through a moving rectangular grid, and then the local terrain of the region is simulated by using the cubic surface fitting function. Erroneous seed points are eliminated by judging the height difference between the seed points and the fitting surface. Additionally, a multi-scale strategy is adopted to gradually reduce the number of seed points used for surface fitting at each level, achieving precise simulation of the local terrain. This process constrains the influence of noise points on surface fitting, thereby achieving the goal of removing noise points and obtaining the final seed points for constructing the initial TIN. The experimental results show that the improved algorithm can obtain more seed points and enhance the accuracy of seed point extraction. As a result, the final initial irregular triangulation network more closely approximates the real terrain, thereby improving the filtering accuracy.