Research on Segmentation and Clustering Algorithms of 3D Point Clouds for Mobile Robot Navigation

The segmentation and clustering of 3D point clouds play a pivotal role in the navigation of mobile robots. In this study, algorithms for segmenting and clustering 3D point clouds were investigated to enhance the navigation capabilities of mobile robots. Firstly, the algorithm proposed by Himmelsbach based on line-fit is widely used in 3D point clouds segmentation. However, the original method has the problem of under segmentation. Therefore, an improved algorithm combining line-fit, noise removal and non-uniform partition method based on Concentric Zone Model was proposed to solve the above problem. Secondly, Euclidean clustering is a commonly used obstacle clustering algorithm, but it is difficult to determine the appropriate distance threshold, which requires a considerable amount of time to adjust the parameter to achieve satisfactory clustering results. A new approach to determine a reasonable distance threshold and prevent excessive clustering was proposed in this paper. Finally, the results of segmentation and clustering experiments showed that our improved segmentation algorithm has higher accuracy and segmentation speed, and our improved Euclidean clustering algorithm has better clustering performance and convenience of threshold parameter setting compared to the original algorithms.