Point cloud data filtering of airborne light detection and ranging (LiDAR) is the focus in the current study of point cloud data processing field. In order to deal with the difficulty of point cloud data filtering at present, an improved filtering method based on hierarchical pseudo-grid and parallel computing is presented. In this method, hierarchical pseudo-grid is established by point cloud data, and the grid is multi-scale decomposed. The original gross error points of LiDAR data are eliminated. The ground point and planimetric point are obtained. According to the principle of double threshold filtering, more refined ground points are obtained by filtering process gradually with the order from big to small mesh scale. And the parallel programming process for point cloud data is combined to reduce the error accumulation of filtering algorithm. Experimental results show that the filtering accuracy of the improved algorithm is enhanced greatly compared to other classical filtering algorithms. The type II errors are controlled effectively in different terrain conditions. Meanwhile, the total errors are decreased, the efficiency of filtering process and the reliability of digital elevation model (DEM) are enhanced.