Boundary point extraction is a crucial step for 3D reconstruction,and most existing methods extract boundary points using one criterion.As a result, the efficiency of extraction is low or the results are unsatisfactory.To solve aforementioned problems,a fast and accurate algorithm is proposed to extract b oundary point cloud in the paper.It consists of two main steps:coarse extracti on and fine extraction.In coarse extraction, scattered points are organized by using Kdtree.For a point,unit normal vectors are constructed based on the poin t and its neighborhoods searched by Kdtree.Then the composed normal vector is c alculated by summing above unit normal vectors.After that coarse boundary point s are extracted according to the ratio of modular length of composed normal to t he number of neighboring points.In fine extraction,the projected plane is esti mated for each coarse boundary point using its neighborhoods. Then neighborhoods are projected onto the projected plane and angles of the judging point and its neighboring points are calculated.After that the accurate boundary points are e xtracted using the maximum angle.Airborne and terrestrial point cloud are both used to validate the proposed method.Experimental results show that boundary po ints are extracted accurately by using the proposed method for these two kinds o f datasets.In addition, the efficiency is improved 6.8times for airborne point cloud and 2times for terrestrial point cloud.The proposed method can be used to accurately extract boundary points with high efficiency,which benefits to 3D reconstruction.