基于直线光流场的三维运动和结构重建  

Reconstructing 3D Motion and Structure from Straight-Line Optical Flow

作　　者：陈震 [1] 高满屯 [2] 曲仕茹 [2] 杨声云 [1] 

机构地区：[1]南昌航空工业学院人工智能与图像处理研究中心,南昌,330034 [2]西北工业大学机电工程学院,西安,710072

出　　处：《计算机学报》2004年第8期1046-1055,共10页Chinese Journal of Computers

基　　金：国家自然科学基金项目(60275037);江西省自然科学基金(0411049);航空科学基金(03I53062);江西省测试技术与控制工程研究中心开放基金项目(ZX200328008);南昌航空工业学院科研基金项目(ZX200304001)资助

摘　　要：利用直线间运动对应关系 ,将像素点光流的概念和定义方法应用于直线 ,提出了直线光流的概念 ,建立了求解空间物体运动参数的线性方程组 ,利用三幅图像 2 1条直线的光流场 ,可以求得物体运动的 12个参数以及空间直线坐标 .但是在实际应用当中 ,要找出这 2 1条直线的光流场是很困难的 ,因此该文提出了运用解非线性方程组的方法 ,只需要 6条直线的光流 ,就可以分步求出物体的 12个运动参数 ,并根据求得的 12个运动参数和一致的图像坐标系中的直线坐标 ,求得空间直线的坐标 ,从而实现了三维场景的重建 .A mathematical definition of straight-line optical flow is first presented in this paper, and according to the conception and definition of pixel optical flow, we define the time derivative of image line coordinates as straight-line optical flow. Then linear algorithm for reconstructing 3D motion and structure from straight-line optical flow is presented. The method useS optical flow of 21 straight lines in 3 consecutive image frames to establish a set of equations about 3D motion parameters. After solving the above equations, 12 motion parameters of 3D object and coordinates of the 3D straight line could be obtained. In fact, it is difficult to get the 21 straight lines and their correspondence. Another non-linear algorithm is presented. The method useS optical flow of 6 straight-line in 3 consecutive image frames to solve the problem of structure from motion. When applying the method of solving non-linear equations, only 6 equations are needed to get rotation and translation parameters. Then the 3D coordinates of the straight line can be recovered.

关 键 词：直线光流场 重建 三维运动和结构 图像序列 计算机视觉 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]