In order to realize 3D reconstruction from huge laser scanning point cloud, this paper proposes an improved algorithm for the simplification of the laser point cloud, based on dynamic grid k-nearest neighbors searching. Firstly, the dynamic neighborhood method is used to quickly find the k-neighbor point in the search process. Then, the data point curvature, the average vector angle between the point and its k-nearest neighborhood points, and the average distance from the point to its neighborhood points are calculated for each k-nearest neighborhood. According to the three parameters, the discriminant parameter and characteristic threshold are defined to extract the feature points. Finally, using the bounding box method to simplify the non-feature points, and to splice the reduced non-feature points. The experimental results show that the proposed algorithm improves the accuracy and computational efficiency, compared with other k-nearest neighbors search methods. Furthermore, this method can not only preserve the geometric features of the model, but also avoid the generation of large-scale blank areas, combining with the feature extraction and the secondary reduction.