结合分支定界法和线性规划的摄像机位姿估计

Camera pose estimation using branch and bound method with linear programming

介绍了一种新的利用对应点估计摄像机位姿的算法。通常情况下,摄像机位姿估计可以转化为一个最优化问题,现有算法将问题转换成一个序列二阶锥规划问题,通过对旋转矩阵所在空间进行分支定界搜索来求取全局最优解。对现有算法进行改进,通过将二阶锥约束松弛为线性约束,提出了一种结合分支定界法和线性规划方法的全局优化算法。该算法不仅能够求得全局最优解,而且算法速度较现有算法提高了一倍以上。最后通过模拟数据和真实数据对该算法进行了验证,结果表明了该算法的准确性和高效性。

In this paper, we introduce a new algorithm for estimating camera pose from point correspondences. Generally, the camera pose problem could he formulated as an optimization problem. The current methods transform the problem into a set of second order cone programming （SOCP） feasibility problems which obtain the global optimal solution by searching the rotation space. In this paper, by relaxing the second-order cone constraints to linear constraints, we propose an improved method that combines branching and bounding with linear programming （LP）. Our method cannot only get the global optimal pose but also runs two times faster than the current best method. Our approach has been tested on a number of synthetically generated and real data sets, and the results demonstrate the accuracy and the high speed of the proposed method.