Recovering 3D Shape and Motion from Image Sequences Using Affine Approximation

3D Shape and Motion from image sequence is acquiring much attention. This paper presents a novel method of 3D structure reconstruction that combines the affine factorization and bundle adjustment subtly. This means allow trade-offs between redundancy, computation and implementation effort. The novel method which was solved linearly based on affine model, followed by optimization based on perspective model not only has improved the accuracy in the case of affine case, but also dramatically decreased computational complexity compared with the traditional perspective case. This method resolves the contradiction of accuracy and robustness in conventional algorithms. Its principle and implementation is presented, as well as real image test show the new method gives an accurate and robust result, and it can be applied to aerial surveying and mapping, virtual reality and military reconnaissance, etc.