基于图像非刚性匹配的物体空间转角检测方法

Object space rotation angle detection method based on image non-rigid matching

转角是物体空间位姿的重要属性之一。对于物体旋转轴线与摄相机光轴线垂直的特殊情况,本文提出了利用图像相似性检测物体空间转角的方法。论文定义了基于匹配特征点的图像相似度系数。考虑到由于图像非线性形变导致的误匹配,本文采用基于向量场一致性的特征点筛选算法,用能量函数的向量场表示形变,在图像匹配中剔除误匹配点,提高图像相似度计算的准确性。实验结果表明,向量场一致性算法可以有效剔除误匹配点,在一定角度范围内,通过标定相似度系数与安全门转角的关系,可以估计物体空间位姿转角。

The corner is one of the important attributes of the object space pose.For the special case where the axis of rotation of the object is perpendicular to the optical axis of the camera,the similarity coefficient of the image feature point is defined in this paper,with a method for detecting the spatial angle of the object using image similarity proposed.Considering the mis-matching caused by the nonlinear deformation of the image,the feature point filtering algorithm based on the vector field consistency is used in this paper.The vector field of the energy function is used to represent the deformation with the false matching points eliminated in the image matching to improve the accuracy of image similarity calculation.Experimental results show that the vector field consistency algorithm can effectively eliminate the mismatched points.In a certain angle range,by adjusting the relationship between the similarity coefficient and the security door rotation angle,the object space pose angle can be estimated.