A method for point cloud optimization includes point cloud data preprocessing steps, point cloud sharp feature recovering steps and point cloud sharp feature enhancing steps, wherein the point cloud data preprocessing steps include large-scale random scanning point cloud simplifying, noise reducing, exterior point removing, homogenization, normal vector calculating and space structure division preprocessing, the point cloud sharp feature recovering steps include further projection and normal vector calculating and point cloud sharp feature recovering, and the point cloud sharp feature enhancing steps include point cloud sharp feature up-sampling and point cloud sharp feature enhancing. By enhancing sharp features of random scanning point cloud, a three-dimensional model finally achieved is up to the requirement of practical application. Meanwhile, optimized point cloud is more concise and neat, so that the existing art can be more suitable for adjusting parameters and capable of improving efficiency. In other words, under the condition that existing point cloud three-dimensional reconstruction processes are not changed, by optimizing raw materials of point cloud, degree of automation, production efficiency and quality of products are improved.