A Hybrid Genetic Algorithm For Relative Pose Estimation Captured By Hand-Held Camera

Relative pose estimation consists of determining the relative position and orientation of two cameras. A hybrid genetic algorithm is presented for relative pose estimation of two overlapped imageries captured by hand-held camera. It makes use of the global convergence characteristic of the genetic algorithm and the local convergence of the gradient-based algorithms. The estimates are achieved by solving a nonlinear least squares problem without initial approximations when more than 5 point correspondences available. The algorithm can effectively overcome the premature of simple genetic algorithm and converge faster. It is a more accurate method compared with the 8-point essential matrix estimation algorithm. Experiments demonstrate the performance of the proposed method.