Ctrl-VIO: Continuous-Time Visual-Inertial Odometry for Rolling Shutter Cameras

A wide range of sensors can be applied for accurate 6-In fact, a GS image corresponds to only one camera pose, DoF motion estimation, among which camera has become while every row of a RS image corresponds to one camera a good choice due to its low cost, light weight and intuitive pose, inevitably leading to a sharp increase in the dimension perception of the appearance information. While visual of the states to be estimated. Therefore, it is computationally odometry (VO) is able to estimate the up-to-scale camera intractable to merely estimate the poses of different rows of poses, it is prone to failure when facing challenges from RS images. A common way to cope with this problem is deficient texture, light variations and violent motion, etc. By to introduce a constant velocity model, assuming the camera additionally fusing Inertial Measurement Unit (IMU) data, moves at a constant speed between two keyframes [7-9]. Another visual-inertial odometry (VIO) can estimate camera poses way is to parameterize the continuous-time trajectory with absolute scale and becomes more robust against the by B-splines [10-13], which is a more elegant way compared aforementioned challenges compared to VO.