As an important geographical environment resource in China, the lake area is of great significance for ecological protection and environmental governance by obtaining high-precision terrain in this area. In order to obtain high-precision topographic data in the lake area, a variety of filtering methods under the point cloud data were studied. Meanwhile, considering the characteristics of progressive encryption triangulation network filtering and cloth simulation filtering itself, the progressive encryption triangulation network filtering and cloth simulation filter fusion filtering methods were introduced to reduce the influence of seed point selection and terrain undulation on the filtering algorithm. With the help of noise point constraints, largescale and small-scale noise were firstly processed respectively to obtain relatively clean point clouds. Secondly, CSF multi-scale acquisition of seed points to be determined was used to complete the acquisition of final seed points by TPS interpolation. Finally, the triangulated network encryption was carried out to obtain ground points. Taking three special areas (construction area, vegetation area, and fish pond area) of Weishan Island as examples, the applicability of the three methods was compared and analyzed. Among them, the fusion filtering algorithm was used in the three sets of data, and the overall error were 0.19%, 2.96%, and 7.32%, which were reduced by 22.44% and 3%, 13.57% and 1.39%, 24.02% and 14.14% compared with the PSD and CSF filtering algorithms. Although the total error of the three algorithms increased in the fish pond area, the accuracy of the fusion filtering algorithm was still the best, and the results showed that the fusion filtering method had the highest accuracy, which was more suitable for multi-type complex regional point cloud filtering.