In order to improve the effect of lidar point cloud denoising, different scale algorithms are proposed. Firstly, the mean value method is used to delete duplicate points from Lidar point cloud data. Secondly, the noise points are estimated. Thirdly, the distance threshold is dynamically set for large-scale noise far away from the main point cloud, small distance is deleted the noise points in a large range, and distance value is delete the noise points in a small range. Finally, small scale noise points mixed with the main point cloud is processed with improving the bilateral filtering algorithm, the data points are gathered in the normal direction, and the position and coordinate of the sampling points are adjusted to the corrected position, which leads to that lidar point cloud denoising was smooth. Experiments show that different scale algorithm can maintain the overall and local characteristics of the scanning target, the signal-to-noise ratio index mean value is 17.56 dB, the root mean square error index mean value is 0.34, and the smoothness index mean value is 0.21, which is better than other algorithms.