多特征点拓扑确定位姿测量算法研究

为解决单目视觉位姿测量时，由目标特征点较多导致图像点与目标点拓扑关系未知的问题，提出了一种多特征点拓扑确定位姿测量算法。较多特征点可在目标进行大角度运动时保证足够的特征点进行位姿解算，与较少特征点相比提高测量精度。该算法将拓扑确定的过程和位姿求解的迭代过程进行嵌套，同时进行拓扑确定和位姿计算。位姿计算的迭代过程基于平行透视投影模型，不需要目标重心投影点坐标作为迭代初始值。拓扑确定的过程转化为分配问题的求解过程。每次位姿迭代的过程中进行一次拓扑确定，拓扑确定的结果可以计算更优的位姿估计。通过多位姿测量实验和精度对比实验结果证明:该算法适合大范围、高精度的位姿测量，在-120~120范围内，位姿测量均方根误差为0。272。

Study of pose estimation based on multiple feature points topological determination

In monocular vision pose estimation, the topological relationship between objective feature points and image feature points where there are multiple feature points is difficult to determine. An algorithm based on multiple feature points topological determination was proposed to solve this problem where the correspondences are unknown. By mounting multiple feature points on the object, enough proper feature points for pose computing were guaranteed when the object is large-scale moving，which can improve the precision of pose estimation. The algorithm nested the iteration process of topological determination and the iteration process of pose computing into one iteration loop, solving them simultaneously. The pose estimation iteration process was based on para-perspective projection model, where the coordination of the projection of the object gravity center used as the initial parameter of iteration is not needed. The iteration process of topological determination was transformed into a solution of assignment problem. Each topological determination can obtain a better pose estimation in every pose estimation iteration loop. The results of multiple poses experiment and precision comparison experiment prove that the algorithm is qualified for the high precision pose estimation of 3D object with large scale motion, with the root mean square error 0.272in the range of -120-120.