The above methodology can be extended to multiple images (i.e. more than 2) which becomes a batch optimization over the whole set of measurements. The relative motion must be estimated for each additional image but the structure is rigid and fixed so the system becomes more constrained. This approach was developed through work by Kumar et al. [34], Szeliski and Kang [52] and Weng et al. [61] where the SfM solution is found via Levenberg-Marquardt nonlinear minimization.