局部特征描述改进的LK跟踪注册方法

Improved Lucas Kanade tracking and registration method based on local descriptor

针对KLT跟踪方法抗光照变化和抗遮挡较差的问题,提出一种使用局部特征描述改进的LK跟踪注册方法(DF-LK)。使用ORB特征点求解初始位姿,通过自适应非极大值抑制重新划分特征点,选择均匀分布的特征点作为LK方法跟踪的控制点集。相邻帧图像之间的单应性矩阵通过在DF描述后的图像上使用LK方法进行求解,跟踪的结果由向前向后错误检测进行评估,由单应性矩阵和初始位姿求解出当前帧的摄像机位姿,并叠加虚拟信息。实验结果表明,该方法在光照变化、部分遮挡和透视变化时均有较好的稳定性和鲁棒性。

To solve the problem that KL tracker is not illumination invariant and occlusion robust,an improved Lucas Kanade tracking and registration method(DF-LK)based on local descriptor was proposed.The initial camera pose was calculated using ORB keypoints,keypoints were repartitioned through adaptive non-maximal suppression and the well-distributed keypoints were chosen as the aggregation of control points tracking using LK method.The homography matrix between two successive images that described by descriptor fields was computed using LK method and the tracking result was estimated by forward-backward error detection.The current camera pose was calculated with homography matrix and initial camera pose.And virtual information was added.Experimental results show that the proposed method is robust and stable against illumination changes,partial occlusion and perspective distortion.