In order to obtain the boundary feature points and boundary lines quickly and efficiently in the scattered point cloud, a point cloud feature regularization algorithm is proposed by means of the fusion of improved field force and judging criterion. An improved k-d (k-dimensional) tree method is first used to search the k neighbors of a sampling point. Then this sampling point and its k neighbors are used as the reference points to fit a micro-cut plane and project to this plane. The local coordinate system is established on the micro-cut plane and the three-dimensional coordinate is transformed into the two-dimensional coordinate. The boundary feature points are identified by use of field force and judging criterion. These boundary feature points are sorted and connected according to the vector deflection angle and distance. The boundary lines are smoothed by the improved cubic B-spline fitting algorithm. The experimental results show that the proposed algorithm can used to extract the boundary feature points quickly and efficiently, and the deviations of the fitted boundary lines are in the level of 10 -5 m, indicating a relatively high precision.