A clustering algorithm for road barrier detection is proposed,and it is used to deal with laser point cloud with anisotropic distribution.Considering the real-time change of the point cloud space distribution,a hierarchical clustering method based on online learning merging-threshold is proposed to determine the upper limit of search range of the cluster number and the selected point set of initial cluster centers.Then the method of maximizing the product of distances is proposed for initializing the point set,which improves the clustering result by incorporating spatial distribution of point cloud and determines the initial cluster centers of K-means algorithm effectively.Finally the optimal number of clusters is determined by the analysis of validity index.This adaptive online learning algorithm is used to cluster the 2.5D laser point cloud in the real experiments,and it can cluster most barriers with anisotropic distribution and close connection correctly in comparison with two other algorithms.