Research On Camera Calibration And3D Rendering Technology In3D Reconstruction

With the development of3D techology,3D reconstruction based on images becomes a hot topic in scientific research filed. Its task is to reconstruct3D scene using the information extracted from their2D images.3D reconstruction based on images relates to a lot of technologies, such as camera calibration, feature point detection and mathing, fundamental matrix estimation, projective reconstruction, affine reconstruction and3D display. In this paper, we mainly research the camera calibration technology and3D display technology in3D reconstruction based on images.Camera calibration with the duty of obtaining camera parameters is a key technology in3D reconstruction based on images. The precision of camera calibration is a strong guarantee to do follow-up works. Camera calibration technologies can be classified into two categories: traditional camera calibration technology which requires to observe a special calibration object and camera self-calibration technology which can finish camera calibration with the information between natural images. In this paper, we mainly dicuss the traditional camera calibration technology and present a simple camera calibration method with intersecting lines on a plane. The calibration object used in the presented technique is easy to make, and can be got easily from life scenes. Between different images, the object can be moved freely, which makes the calibration process very simply.Imaging process is a projective transformation, which keeps points colinearity and curves intersection, namely for space collinear points, their corresponding pixels in image plane are also collinear, for space intersecting curves, theirs corresponding curves in image plane are also intersecting at the corresponding pixel of space intersection. Traditional calibration technology mostly uses these properties in projective transformation. But by noises, etc., these properties can not be maintained in image plane. In experiments, we handle the calibration objects images with a pre-processing technology, adjusting pixels to let them meet points colinearity and curves intersection in original space. Experimental results show that, with pre-processing technology, the accuracy of existing calibration algorithm has been greatly improved.The results of3D reconstruction should be displayed by the form of3D, which relates to3D rendering technology, especially texture mapping. In this paper, we also research texture mapping technology in3D rendering. High image quality and efficiengy are usually conflicting goals in texture mapping, especially for real-time rendering. Mipmap is a typical texture mapping filtering method for texture mapping. Due to simplicity and efficiency, it has been popularly used in current graphics hardware. However, its implementation eficiency is at the cost of image rendering quality. We first present an improved sampling scheme using the same Mipmap pyramid. In this scheme, for each screen pixel in the scanline, instead of just using the texture scaling ratio in the single direction to determin the Mipmap pyramid level for sampling, we take a row sampling and column sampling respectively in different. Mipmap pyramid levels to generate its value. These two pyramid levels are determined independently by the texture scaling ratios in the row and column directions. The double samping scheme can overcome the aliasing and overblurring problems caused by just sampling using the texture scaling ratio in a single direction. Motived by this, we propose a new txture filtering method called Dipmap (Double Mipmap) to further improve image rendering quality. In Dipmap, we establish a row pyramid for rwo sampling and a column pyramid for column sampling to makes these samplings completely independent of each other. Experiments show that Dipmap can diliver much improved image rendering quality than the state-of-the-art texture filtering methods like Mipmap, Ripmap and Fast, while being nearly as efficient as Mipmap.