摘要： 针对半标定图像的极线校正问题，结合图像特征点的尺度和主方向性质，提出一种新的误配剔除方法。根据全局的缩放和旋转约束对误配点进行粗剔除，结合特征点的尺度、主方向和空间布局信息对剩余的误配点进行精剔除，使得到的精确匹配点满足极线约束。根据此约束，利用最小二乘方法估计未标定参数，并将半标定的图像极线校正转换为已标定的图像极线校正。实验结果表明，该方法生成的校正图像对精度较高，图像变形较小。

关键词: 立体视觉, 误配剔除, 一致性约束, 极线校正, 半标定, 最小二乘法

Abstract: A novel mismatch removal method is proposed to deal with the problem of epipolar rectification in the semi-calibrated case. It discards the mismatch with global scaling and rotation consistency, and rejects more incorrect matches accoding to the information of scale, orientation and spatial layout. The rest correspondences satisfy epipolar geometry constraint, so the method uses least square to estimate the uncalibrated parameters. Experimental results show this mismatch removal method works well in the problem of semi-calibrate epipolar rectification. The rectification precision is high and the image loss is low.

Key words: stereo, mismatch removal, consistency constraint, epipolar rectification, semi-calibrated, least squares method

中图分类号: 

• TN911.73