next up previous
Next: Experiment 1: Egomotion, Models Up: 3D Structure from 2D Previous: Biased Measurements


To test our recursive structure from motion system, a variety of simulations were carried out with synthetic data. These tests are detailed in [2] and demonstrate the accuracy and reliability of the method. Structure, 3D motion and focal length were stably recovered in situations that included increased noise levels, orthographic projection, increased noise bias and degenerate rotational motion. The estimator performed accurately despite extreme levels of noise and was robust to bias as well. In addition, it properly handled the orthographic and degenerate rotation cases which typically cause problems for other techniques and camera models. In addition, an extensive Monte Carlo analysis was performed with thousands of trials to confirm the stability of the technique and show smooth degradation in increasing noise conditions.

In this section, we present results from applying our estimation formulation to two sequences of real imagery. Additional real imagery results can also be found in [2].