A Connection-Based Point Cloud Segmentation Method Using Bipartite Graph

Point cloud segmentation is a fundamental but necessary step for many real-life applications. However, most of the existing segmentation methods suffered from the multiple types of surfaces and noise data, which leads to the ‘over-’ and ‘under-’ segmentation, and inaccurate boundaries. To solve these problems, a new robust technique is proposed for segmenting the point cloud into planar or curved primitives in this study. First, the point cloud is decomposed into structural supervoxels. We employ the local dimensional feature to improve the performance of the supervoxel segmentation method near the boundary area. Second, a connection-based merging algorithm is proposed to cluster the adjacent supervoxel based on an optimal matching method. Comprehensive experiments demonstrate that the proposed method obtained better performance than other baseline methods on outdoor samples with low computation costs.