3d Reconstruction Method Based On Silhouettes

The ultimate goal of computer vision is to understand 3D objects, that is to achieve certain functions of human visual system. The research content of the 3D computer vision is using the 2D projection image of object to restore its 3D features. In this thesis, we use the object contour of 2D projection image , with the help of the constraint relations between corresponding image contour and the geometry feature to calculate the camera parameters, and then get the minimum of space objects -visual hull, which is close to the real object, to 3D reconstruction of objects. This thesis includes the following aspects:(1) Research of camera model and different camera calibration methodsWe introduce a mature and commonly used imaging models, linear model, in the current field of 3D reconstruction. And summarize some commonly used methods of camera calibration, including traditional camera calibration method, camera calibration method based on active vision and camera self-calibration method.(2) The study and design of active contour modelsActive contour models use the advanced ability to identify of human beings, make the method focuses on the region of interest nearby, resulting in improving the reliability and accuracy of getting contour out of the object image. We introduce the principle of active contour model, and its basic mathematical model was derived. Subsequently, we analyze the advantages and disadvantages of active contour models. Finally, the algorithm for active contour models have been studied and implemented.(3) Multi-planar mirror 3D reconstruction methodThe plane mirror imaging principle tells us that when using the plane mirror imaging, we will get an additional virtual observation point of the target object. When using two angled plane mirror imaging, the object will have multiple virtual images, if we shooting with the camera at this time, the resulting image will contain the object and its virtual image in mirror, it is the same as imaging the target object from different angles at the same time. Therefore, we propose the use of an angled plane mirror for calibration, thus achieving 3D reconstruction.