A grads method is used to process lidar signal, by which the scopes and positions of abrupt change can be detected. Based on these grads messages, a translation sequence can be calculated and is added to original lidar signal, which makes the parts of abrupt change in original lidar signal removed. The abrupt change removed signal is decomposed and denoised by empirical mode decomposition, which can avoid mode mix and local surge phenomena. This denoising method is applied to real lidar data denoising. The experiment results show the effectiveness of this method. It can remove noise in signal effectively in flat part and keep abrupt change well in signal.