Abstract: According to the characteristics of photon counting laser point cloud, this paper studies and implements the point cloud denoising algorithm in ICESat-2 ATL08 technical document and carries out the accuracy evaluation. Firstly, the total number of points in the neighborhood of the target point is calculated point by point as the local density value. Since the signal is usually more concentrated than the noise in the spatial distribution, the density histogram often exhibits the double peak characteristic of noise on the left and the signal on the right. Then, initial parameters of noise and signal Gaussian function are calculated by peak calculation and distance judgment, and Gaussian fit and expectation-maximization algorithm are adopted for fitting two exact waveform position. Finally, the intersection position of the double Gaussian function is taken as the denoising threshold. The points smaller than the threshold are marked as noise, and the points greater than the threshold are signals. The experimental results of MABEL data show that when the data exhibits typical bimodal distribution characteristics, the algorithm has excellent denoising effect, and the accuracy is better than 98%. When the data does not meet the bimodal characteristics, the effect becomes worse.

Key words: photon counting, LiDAR, spatial density, denoising algorithm, accuracy evaluation