1: \begin{abstract}

2: This paper revisits the problem of continuous-time structure from motion,

3: and introduces a number of extensions that improve convergence and efficiency.

4: The formulation with a $\mathcal{C}^2$-continuous spline for the trajectory

5: naturally incorporates inertial measurements, 

6: as derivatives of the sought trajectory.

7: We analyse the behaviour of split interpolation on $\SO3$ and on $\R3$, and a joint interpolation on $\SE3$,

8: and show that the latter implicitly couples the direction of translation and rotation.

9: Such an assumption can make good sense for a camera mounted on a robot arm, but not for hand-held

10: or body-mounted cameras.

11: Our experiments show that split interpolation on $\RplusSO3$ is preferable over $\SE3$ interpolation in all tested cases.

12: Finally, we investigate the problem of landmark reprojection on rolling shutter cameras,

13: and show that the tested reprojection methods give similar quality, 

14: while their computational load varies by a factor of 2.

15: \end{abstract}

16: