Adaptive fusion of different platform point cloud with improved particle swarm optimization and supervoxels

Abstract

Fusion of point cloud from different platforms is crucial for enhancing spatial information completeness in large-scale scenes, particularly in urban 3D modeling. To address redundancy, noise, and accuracy degradation in direct registration of point cloud across platforms, we propose an adaptive fusion method utilizing supervoxels. Initially, a high-precision point cloud is selected as the reference point cloud (RPC),and we apply a coarse-to-fine registration approach to unify the RPC and the target point cloud (TPC). Registration parameters are optimized using Improved Particle Swarm Optimization (IPSO), enhancing automation and precision of the fine registration. Subsequently, supervoxels are constructed for the registered RPC and TPC. Finally, within each corresponding supervoxel, redundancy and noise are eliminated by applying alpha-shape and Laplacian, considering the data quality and density distribution of the RPC. Experimental validation was conducted with data acquired from three distinct platforms. The proposed method significantly enhances registration precision. Compared to RANSAC-ICP, our method reduced average RMSE by 36.35%, average MAE by 34.85%, and average Frobenius Norm by 84.48% across three experimental groups. The proposed fusion method improves data completeness, reducing the point cloud count by about 30% compared to direct registration. Moreover, it effectively preserves the detailed features of the fused point cloud, serving as accurate data sources for constructing 3D urban models.

Keywords

• Point cloud
• Particle swarm optimization
• Supervoxel
• Density
• Adaptive fusion