双摄象机大转角立体图象的深度计算

RECOVERY OF 3-D INFORMATION OF A POLYHEDRON FROM TWO PERSPECTIVE VIEWS WITH GREAT VARIATION OF VIEWPOINT

本文就具有普遍性的“大转角双目视觉”的三维恢复问题,提出了一种包括自动匹配问题在内的求取物体三维坐标的系列最小二乘法.这种方法不仅能够克服由于大转角变化及遮挡等原因造成的求取图象对应点的困难,而且在数据准确性、鲁棒性、计算复杂性、保证旋转矩阵正交性及三维运动的直接恢复程度方面都优于现存的方法;用实际图象数据进行计算,结果满足实际要求.

A new algorithm based on point-to-point correspondence and least-square solution is proposed to deal with the recovery of 3-D information from two perspective views with great variation of viewpoint. The algorithm is aot only able to overcome the difficulty of Image-Matching caused by great variation of viewpoint and occluding, but also has advantages over the existing algorithms in numerical accuracy, computing complexity, orthogonality of rotation matrix, robustness and the amount of recovered information. It is demonstrated that the precision of algorithm is up to 3.3% for images of real world scenes.