Abstract: Aiming at the problem that it is difficult to retain fine features and accurately "lock" riverbed morphology in multi-beam point cloud denoising, a multi-beam point cloud denoising algorithm is proposed for 3D registration of water flow of natural resources. On the basis of KD tree search, statistical filtering theory is introduced to classify multi-scale noise and eliminate large scale noise. Based on the principle of minimum information entropy, the optimal neighborhood is determined based on the principal component analysis (PCA) algorithm, and the curvature information entropy is constructed to optimize and improve the bilateral filtering factors, so as to achieve the purpose of denoising and preserving fine features of underwater terrain point cloud. Experiments show that the proposed algorithm is feasible, can effectively ensure the fine features of underwater terrain and applied to the 3D registration of water flow of natural resources.

Key words: 3D registration of water flow, multi-beam point cloud, curvature information entropy, optimal neighborhood, bilateral filtering