【摘要】 提出一种立体视觉中应用极几何结合仿射变换来解决图像匹配问题的方法。首先利用两幅图像间的极几何关系,确定一些可靠匹配点作为控制点;进而构建全局仿射变换,把像面的特征点投影到另一幅像面上。再利用极几何约束和变换后的图像点位置关系,进一步搜索匹配点。最后是重匹配,分区域对控制点进行优化更新,经过迭代运算,得到最终结果。实验表明,这种算法能较快地收敛,有效地剔除误匹配点和提高匹配精度。更多还原

【Abstract】 A method for image matching in stereo vision is proposed in this paper,which uses epipolar geometry and affine transformation comprehensively.Firstly,find some reliable matching points with epipolar constraint as initially matching.Then a global affine transformation can be constructed and the feather points on one image can be transformed to the conjugate one.Constrained by epipolar geometry and transformed points’ coordinates,more matching points can be obtained.Lastly,choose and optimize the control points for next matching,substitute the control points adaptively.The final result can be obtained after this iteration.Experimental results show that this algorithm converges fast and can delete most wrong matching points effectively.更多还原