Airborne lidar has allowed the rapid generation of high-resolution digital terrain models of large areas, but it is still difficult to automatically identify ground points and non-ground points in areas covered by dense buildings or dense vegetation. This paper proposes a mobile curve fitting least squares iterative algorithm automatically and quickly filters Lidar data. This method uses moving window weighted iterative least squares method to select seed points, and based on adaptive thresholds, it gradually filters and classifies non-ground points and ground points. Experiments in four study areas show that the new filtering method can separate the urban area and the ground and non-ground points covered by dense vegetation. For type Ⅰ errors, the error range of the new algorithm is 4.08% to 9.40%, for type Ⅱ errors, the error range is 2.48% to 7.63%, and for total errors, the error range is 5.01% to 7.40%.