Three-dimensional Model Calculation Of Natural Scene Based On Stereoscopic Vision

Stereoscopic vision is an important branch in computer vision,and has been widely used and developed in many fields such as space exploration,medical diagnosis,industrial measurement and digital education.The core problem in stereoscopic vision is the accuracy and efficiency of the stereo matching algorithm.Fast,accurate stereo matching method has always been a key technology in all areas of stereoscopic vision.So the continuous optimization and improvement of the three-dimensional matching technology is very important.A basic stereoscopic vision system includes image acquisition,image correction,stereo matching and three-dimensional reconstruction.Through the binocular stereoscopic vision system,the two cameras were used to photograph the objects,and the corresponding parallax graphs were obtained by the stereo matching algorithm.According to the correspondence between parallax and depth,we can restore the three-dimensional information about the scene.This paper is to build a complete three-dimensional visual system,and recover the 3D model through the glasses-free display device.Firstly,we summarize the research background of stereoscopic vision technology.Including the application of stereoscopic vision in various fields,as well as the technology and application direction researched at home and abroad in recent years.And then we analyze the principle of the entire stereoscopic vision system.Mainly it can be divided into camera systems and stereo matching system.For the camera system,the geometric imaging principle of the camera and the spatial coordinate transformation are analyzed in detail.By using matlab toolbox to complete the camera calibration work,we obtained the internal and external parameters of the camera system,and completed the correction of the image.For the stereo matching system,the basic model of binocular stereoscopic vision system is introduced,and the relationship between depth and parallax and the basic framework of stereo matching algorithm are expounded.By comparing the two kinds of local stereo matching methods,adaptive weight method and guided-filter method,an optimized framework of stereo matching is proposed.Different from the previous local matching algorithm,which mainly optimize the matching cost and cost of aggregation method,we put the focus of the algorithm on the parallax refinement step to simplify the amount of computing.In the post-processing step,the weights are calculated by the fast-guided filter,and the weighted-median-filter is used to optimize the parallax graph,which greatly improves the accuracy and efficiency of the stereo matching algorithm.According to the disparity map obtained by the matching algorithm,this paper adopts two multi-view expansion methods different from the previous DIBR technology.One is based on the phase enhancement,which can expand the viewpoint without disparity maps.The other is based on Fourier slices technology.According to the Fourier projection transformation,we get a projection in a particular slice and obtain the new viewpoint by inversing Fourier transform.