Research And Implementation Of Three-dimensional Reconstruction In Planar Scene

Three-dimensional reconstruction is a basic research problem in computer vision and computer graphics.As a technology of data storage and virtual reality,it restores two-dimensional space to three-dimensional space,and it is used in scientific engineering such as scene modeling,medical diagnosis,and three-dimensional measurement.In addition,there is a wide application prospect in archaeology,entertainment,film and other cultural fields.The 3D reconstruction of planar scene is getting more and more attention,especially in the field of sports analysis,and it is one of the hot spots in this field to get the data of 3D scene quickly and accurately.Based on the field of sports,this paper aims to realize the three-dimensional positioning of scene target,and studies several technologies of 3D reconstruction of planar scene.In the uncalibrated scene,the geometric constraint relationship from different perspectives is restored according to the feature matching between the real scene and the planar scene model.The main work is as follows:(1)The preprocessing of scene reconstruction includes the study of planar scene and target detection algorithms.The video of scene captured by the camera contains a large amount of data information.In order to improve the reconstruction speed and restore the three-dimensional information of the target in the scene,a planar scene detection method combining color clustering and image local entropy is proposed in this paper according to the color features and texture features of the scene.Finally,we extract the semantic information of planar scene.Based on this,the background model of the scene plane is established to obtain the preliminary detection results of the target,and the foreground target pixel is complemented with the binarized result of the saliency map.In the end,the complete detection of multi-objects is achieved.(2)Research on camera self-calibration method in three dimensional scenes.In fact,the camera often needs to adjust the angle and focal length according to the requirements.In order to solve the errors caused by the traditional off-line calibration,a homography solution method based on global optimization is proposed,and the dynamic calibration of the camera is achieved by using the geometric characteristics of the planar scene.Firstly,according to the bottom-up idea,RANSAC algorithm is used to extract the geometric feature parameters in the plane,and the sequential matching strategy is used to realize the rapid matching of geometric feature points.Combined with the top-down parameter estimation method,the non-isotropic scaling of the planar scene model is used to eliminate false matching points,and the initial solution of the projection matrix is obtained.Finally,the projection matrix is optimized globally according to the inverse projection and re-projection errors of the parameters.(3)Three-dimensional reconstruction of planar scene is realized.In order to complete the dynamic updating of camera parameters and the reconstruction of scene sequences,the global motion parameter recovery method is designed in this paper.The plane transformation matrix between scene sequences is deduced based on the camera's motion.The feature points of the scene are tracked by the pyramidal Lucas-Kanade sparse optical flow,and the camera parameters are restored under the RANSAC iteration framework.Based on the camera calibration parameters,the projection model uniquely determines the position of the target in the real three-dimensional scene.At the same time,in order to make the target information more abundant,we use its color feature to train and achieve the target recognition.Combining the geometric model and color model of the scene and target,the multi-perspective visualization of the scene is realized in the modeling environment,and the recovery from the two-dimensional space to the three-dimensional space is completed.