Ambiguity in the calculation of normal vectors during Poisson surface reconstruction of 3D point clouds can lead to errors in the reconstructed surface. This paper introduces a modified Poisson surface reconstruction algorithm designed to address the ambiguity of normal vectors. The algorithm calculates the normal vectors of all input point clouds, selects the point with the smallest curvature as the origin, defines its normal vector as the positive direction, searches for neighboring points using the KD tree, reorients the normal direction of neighboring points, and then regards its neighbor points as the new origin to correct the ambiguity of the normal and complete the overall orientation of the point cloud, and finally the algorithm reconstructs the surface according to the obtained normal vectors. Observations and quantitative analysis show that the algorithm significant reduces the ambiguity of normal vectors and can reconstruct a better surface, reducing the maximum deviation distance errors of three models by 66%, 86%, and 95% respectively.