Research On 3D Reconstruction Using A Digital Camera

With the development of multimedia and virtual reality, 3D model is widely used in the object recognition, industrial design, video games, digitalization of historical relics and reconstruction of historical buildings. Efficient and real-time 3D model reconstructing method has become a common important topic in various fields. Based on the analysis to several methods for reconstructing geometric shape from image sequences, this paper presented a method by using a stationary camera and a rotating stage to acquire the multiple view images of real objects, and applying visual cone intersection to build the approximate 3D model. The main points are as follows:Firstly, based on the analysis to the characteristics of existed methods for 3D modeling from images, the Shape from Silhouette was chosen to reconstruct the 3D model of object, and the mathematic model of the system was established. Then, a photography system for Shape from Silhouette was built which is composed of a stationary camera and a rotating stage. The calibrations of the camera parameters and rotating axis parameters were carried out, and the precision of the results was confirmed by a calibration example. After that, the multiple view images of real objects were taken with fixed angle interval. By image processing, the clear contours coordinates of input images were extracted from the corresponding binary images after threshold segmentation. Furthermore, Matlab program was written to process image sequences and save the silhouette data automatically. Finally, after the setting and development of Rhino, the approximate Visual Hull of the object was acquired by using the method of combining volume intersection and Rhino. Several examples were given to confirm the correctness and feasibility of the reconstruction method, and the experiment result was analyzed as well.With the rotating axis calibration in image sequences processing, the method presented in this paper do not need high placement precision for the rotating stage and camera. By combining the volume intersection and Rhino during 3D model building, the buliding process of Visual Hull can be observe directly, which is propitious to the manual control of the surface details. Compared to the conventional reconstruction methods that based on voxel, the computing complexity is not related to the resolution and the volume of the object, and the time complexity is better than the classical voxel pixel method. Meanwhile, the method can directly reconstruct the completely closed surface model, and do not need to build the surface from the discrete surface points. The experiments show the good precision, reliability and efficiency for the 3D reconstructing of convex objects.