一种基于直线特征的单目视觉位姿测量方法

A mono-vision method of measuring pose based on line features

提出了一种基于直线特征的单目视觉位姿测量方法SoftNewton。构造了新颖的目标直线与图像直线匹配评价函数,避免检测图像中直线的端点,最终通过软决策技术确定直线特征匹配关系,并采用高斯牛顿迭代算法基于全透视成像模型解算目标位姿。和POSIT算法相比,高斯牛顿迭代算法保持了旋转矩阵的正交性,提高位姿解算精度。仿真图像实验中,在干扰直线和噪声存在的情况下算法经过29次迭代解算得到正确的直线特征匹配矩阵,姿态误差小于0.2°,位移误差小于0.5 mm。仿真图像和实际图像实验结果均表明SoftNew-ton具有较高解算精度和较强的鲁棒性。

A new mono-vision algorithm to measure pose based on line features,called SoftNewton,is proposed.The new evaluating function for the matches between image lines and object lines is constructed,which avoids detecting the lines endpoints in image.The algorithm applies Soft-Assign technique to determining the correspondence of lines feature between image and object.Futhermore,GaussNewton iterative algorithm,that computes object pose under a full-perspective camera model,keeps the rotation matrix orthogonal and gets higher precise results.Experimentally,the simulated images with disturbing lines and noise are measured by SoftNewton.After 29 times of iteration,the accurate match matrix of lines feature is achieved.As a result,the attitude error is less than 0.2°,and the position error is less than 0.5 mm.Experiments involving synthesis images as well as real images demonstrates the robustness and accuracy of SoftNewton.