基于双目图像的复杂视频场景虚拟重建仿真  

Virtual Reconstruction Simulation of Complex Video Scene Based on Binocular Image

作　　者：袁红 李瑾 黄婧 YUAN Hong;LI Jin;HUANG Jing(School of Medical Information and Engineering,Southwest Medical University,Luzhou Sichuan 646000,China)

机构地区：[1]西南医科大学医学信息与工程学院,四川泸州646000

出　　处：《计算机仿真》2022年第6期459-463,共5页Computer Simulation

基　　金：教育部高等教育司2020年产学合作协同育人项目(202002273038)。

摘　　要：传统视频场景虚拟重建方法忽略了对双目视觉的摄像机标定,无法高精度重建视频场景。为此设计基于双目图像的复杂视频场景虚拟重建方法。对双目视觉摄像机标定,并建立转换坐标系。为了降低重建的复杂度,去除冗余关键帧,匹配场景中特征点,并采用匀速运动模型估计当前帧的初始位姿,选取重建控制点。利用得到的匹配点与控制点,重建复杂视频场景虚拟。将相对位姿误差、场景特征点匹配准确性以及重建时间作为实验对比对象。实验结果表明所研究方法的相对位姿误差小、场景特征点匹配准确性高,并且所花费的时间少,证明所研究的方法能够在短时间内实现准确的场景虚拟重建。Traditionally, the method of virtual reconstruction for video scene ignores the camera calibration of binocular vision, so it is unable to reconstruct the video scene with high accuracy. Therefore, this article designed a method of virtual reconstruction for complex video scene based on binocular image. Firstly, the binocular vision camera was calibrated, and then the coordinate conversion was established. In order to reduce the complexity of reconstruction, redundant key frames were removed. Moreover, the feature points in the scene were matched. Meanwhile, the uniform motion model was used to estimate the initial pose of the current frame and choose the control points for reconstruction. Finally, the matching points and control points were adopted to reconstruct the complex virtual video scene. The relative pose error, the matching accuracy of scene feature point and the reconstruction time were all experimental objects. Experimental results prove that the proposed method has smaller pose error, higher matching accuracy of scene feature points, and less time consumption. Therefore, the proposed method can achieve accurate scene virtual reconstruction in a short time.

关 键 词：双目图像 复杂视频场景 虚拟重建 冗余关键帧 位姿误差 场景特征点匹配 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]