Abstract: According to the characteristics of urban LiDAR point clouds, we proposed a knowledge-based filtering algorithm with adaptive TIN models. The main strategies are: ① taking object-oriented segmentation for raster data interpolated regularly; ② separating terrain objects from off-terrain objects by using iteration Otsu clustering method; ③ constructing the initial TIN form classification results and adjusting the parameters of the ground point criterion adaptively in the aim of improving the filtering quality. We did experiment with the real data of ALS50 system and also assessed the results quality with traditional algorithm. The result shows that knowledge-based filtering method can further improve the quality of point clouds filtering.