Abstract. Currently, traffic law enforcement agencies rely mainly on the license plate to identify illegal vehicles. But for those vehicles with no license plates or license plates are obscured or slipped, the existing techniques do not work. In this paper, a method to reconstruct the external dimensions of vehicles from 2D images is therefore proposed to identify those illegal vehicles above. The external characteristics of vehicles and methods to model vehicles are analyzed. Based on the analysis, we simplified vehicles as a combination of a rectangular parallelepiped and trapezoidal polyhedron which can be produced with corner points from 2D images. The process to establish the vehicle model with corner points is studied. Finally with Visual C++ and SolidWorks API, the 3D model of a vehicle is established based on a 2D image. The vehicle's primary outlines information including length, width and height are restored, and the information can be used to identify the vehicle.