基于改进的ORB算法与姿态估计的跟踪注册方法研究

传统的特征提取算法在图像匹配过程中易出现误匹配现象,本文在ORB算法的基础中融入一种最小平方中值估计法-LMedS方法,利用ORB算法的特点和LMedS方法去除可能存在的外点,消除误匹配现象,从而得到正确的匹配特征对,使特征匹配率有很大的提高。采用基于非线性最小二乘进行姿态估计,通过迭代算法估算相机姿态完成虚实注册。实验结果表明,本文的方法无论是在特征点匹配还是在实际场景中都具有很好的鲁棒性,在不同尺度角度、部分遮挡的情况下,同样具有良好的性能,准确、实时地完成跟踪注册。

Research on Tracking and Registration Based on Improved ORB Algotithm and Pose Estimation

The traditional feature extraction algorithm is prone to error match in image matching.this paper combine estimation of least median squares with ORB algorithm,using characteristic of ORB algorithm and LMedS to remove the outer point, eliminate error match,and result in the correct match,improve matching rate.we proposed a method of nonlinear least squares to estimate pose,and then estimate the camera pose to complete the tracking registration by iterative algorithm.Experimental resultes show that the method proposed in the paper has good robustness in feature matching and actual scene,and can complete accurately and real-time the tracking registration in the case of different scale angle and partial occlusion.