To solve the problem of low efficiency and weak stability in searching the -nearest neighbors of a large-scale scattered point cloud

a fast algorithm for finding -nearest neighbors is presented. First

the point cloud data is divided into different sub-spaces by using a space block strategy. Second

the variation of the search step length is controlled dynamically. The accuracy of the algorithm is ensured by the minimum distance from the point to the small cube boundary. Finally

the infinite loop problem due to improper initial values in existing algorithms is avoided by altering the right-side threshold

which controls the number of pre-screening points. The experiment results show that the proposed method obtains not only a good stability for the initial searching step

the step increment

and the sampling density at different topology structures

but also a better performance than the existing algorithms.