Research On Digital Modeling For Car Styling Based On Single View Images

Actually, the procedure of the styling of automobile is a conversion process which begins from 2D image to 3D model, that is to say, from the stylists’design sketches or renderings, through CAS (Computer Aided Styling) engineers turning the sketches or renderings into 3D digital model, at last with a series of ponder and discussion getting the last 3D digital model or clay model. However, in the conversion process which turns 2D to 3D, the CAS engineers may misunderstand the stylists’design ideas and modify the model again and again, which is professional and time-consuming, lengthens the development cycle of the car body styling and reduces the efficiency of design. Therefore, how to quickly reconstruct the 3D digital model of automobile from the 2D image, as well as achieve the three-dimensional modeling of data reuse effectively and improve the efficiency of design must be a significant work.For effective digital modeling of automobile styling based on single-view (side view) image, in this paper, we utilize the modeling approach of data-driven which is a research hotspot at present; firstly, we define reasonable the characteristic curve network template of car styling with consistence representation and corresponding surface models, and with these put forward a reasonable 3D reconstruction method based on statistical morphable model from side view. And we achieve effective reconstruction 3D digital model from single view (side view) and propose the constraint deformation of 3D surface models. By the data effective reuse of 3D modeling data, creative design can be achieved. In this paper, the research work is mainly from the following several aspects:(1) Define characteristic lines template of car body and build model library. By analyzing the lines characteristic of the styling of car body, we define more reasonable and complete characteristic lines of car body, includes the front, the rear, the top and the side view. And by analyzing the correspondence relationship between four views and 3D curve of car body, we define the same topological structural 3D characteristic lines template of car body. Each of form characteristic lines is represented as a cubic Bezier curve so that achieve a consistent parametric representation. We establish more than 50 3D characteristic curve networks manually as database, which provide database for the reconstruction of 3D curve network and shape expiation. (2) 3D surface modeling based on sparse reconstruction. Analyze the existing method of reconstruction 3D model based on single view (side view) and we propose a 3D reconstruction method base on statistical morphable model from side view. This method is based on database and reconstructs the 3D curve network of car body through given characteristic lines or sparse feature key points. Then with the second development of Siemens NX, the surface model of car body is generated automatically based on template. Because of the 3D reconstruction based on single view is an ill-posed problem and the reconstruction is focus of the estimation accuracy of depth information, so we also proposed a 3D reconstruction method given sparse prior depth information. By numerical experiments and error analysis, we prove that the modeling method is fast, robust, high-efficiency in this paper.(3) Constraint deformation of the 3D surface models. Discussed how to utilizing morphable model method to achieve free deformation, deformation with feature preserved and deformation with layout constraint (wheelbase variant) of 3D car body. Reconstruct the 3D curve networks effectively given sparse key points. Achieve a creative design platform of a car for flexible editing of 3D surface models.(4) Development of software modules. Finally, based on C++ and OpenGL, the software module of the reconstruction of 3D curve network of automobile body, the software module of constraint deformation DCiP-AutoMorpher/Styling are introduced.