Linear relaxation for global pose estimation

This paper introduces a new algorithm for estimating camera pose from point correspondences. Generally, the pose problem is formulated as an optimization problem whose aim is to minimize the reprojection errors. Usually this kind of problem is non-convex and may get trapped in local minima. For dealing with this difficulty, we combine Branch and Bound with Linear Programming to generate a global optimal solution. We first use a branch-and-bound search over rotation space to find the best rotation and the problem can then be reduced to known fixed-rotation problems for which optimal solutions can be found by Linear Programming. The proposed method can obtain the global optimal solution and is very fast. It has been tested on a real data set and the results demonstrate the accuracy and high speed of this method.