Creation of high-fidelity finite element models (FEMs) is an important prerequisite of finite element analysis which is widely used for assessing the health condition of structures and infrastructures. Currently, most FEMs are created by professional engineers based on as-designed condition. For aging structures and infrastructures, the as-is conditions might vary significantly from as-designed conditions due to accumulated damage. In order to capture as-is conditions, 3D laser scanners have become increasingly available because of their capability of capturing massive spatial data with high accuracy in a short period of time. However, currently, the creation of FEMs from laser scan data is still a manual process which is labor-intensive and might be error-prone. This research investigates an automatic approach of using spatial data from laser scanning along with computer vision techniques for creating high-fidelity FEMs automatically. Distinct point features from raw laser point cloud is established first. The extracted point features are then integrated with prior information available on the distributions and shapes of structural components for detecting and recognizing these components. Additional steps are proposed to extract geometric information for structural components with common shapes such as wide flange beams and cylindrical concrete piers. At the end, a methodology of converting point cloud data or extracted geometric information into all-hexahedral finite element meshes is presented.