Airborne LiDAR point cloud data filtering is the most important step in the workflow of LiDAR data postprocessing. Based on the characteristics of Triangular Irregular Networks(TIN) and curved fitting filtering methods, a "from rough to fine" idea was proposed for LiDAR point cloud data filtering. In this method, strict threshold TIN was used for "rough classification" with a priority of type II error and more reliable initial ground points were obtained, then the seed points were selected with the priori information which was "rough classification" result, next Total Least Squares(TLS) algorithm was introduced to fit block terrain, and self-adaption threshold was set to deal with different area more flexibly, ultimately more refined region model was obtained. ISPRS test data and Niagara data were used for experiments, and classic filtering method and traditional curved fitting filtering method were selected for comparison. Experimental results prove that, the proposed method is practical as the filtering results are more reliable than traditional moving curved fitting filtering method, and has strong adaptability to various terrains.