A novel initialization method based on genetic algorithm for simultaneous pose and correspondence estimation

Simultaneous pose and correspondence estimation problem is used to determine the pose of a 3D object from a single 2D image when corresponding relation is unknown between 3D object points and 2D image points. The problem arises in many areas of computer vision and some algorithms have been presented. However, all the state-of-art algorithms rely on appropriate initialization and the correct solution may not be reached to in many times with the traditional initialization method which starts randomly. We derive a novel method which estimates the initial value based on genetic algorithm, considering the influences of different initial guesses comprehensively. Using this initialization method, the proper initial guess could be calculated and the simultaneous pose and correspondence problem could be easily solved. Simulation results and experiments on real images prove the effectiveness and robustness of our proposed initialization method.