Zettelkasten - Why use the visual-inertial sensor combination?

Source: Mur-Artal 2017 VI-ORB

Cheap but also with good potential
Cameras provide rich information but are relatively cheap
IMU
- provides self-motion info, helps recover scale in monocular applications
- enables estimation of the direction of gravity –> renders pitch and roll observable

Source: Forster 2017 IMU Preintegration

Visual-inertial fusion for 3D structure and motion estimation
Both cameras and IMUs are cheap, easy to find and complement each other well
Camera
- exteroceptive sensor
- measures, up to a to-be-determined metric scale, appearance and geometrical structure of a 3D scene
IMU
- interoceptive sensor
- makes metric scale of monocular cameras, as well as the direction of gravity, observable

Source: (Wu 2018) Image-based camera localization

Cameras provide rich information of a scene
IMU provide odometry
- self-motion information and
- accurate short-term motion estimates at high frequency

Source: Mirzaei 2008 A Kalman Filter-Based Algorithm for IMU-Camera Calibration: Observability Analysis and Performance Evaluation

Motivation Inertial navigation systems (INSs) combine IMU with GPS. However, GPS is not always utilisable (e.g. indoor locations) –> cameras/visual sensors are used as an alternative. Points in favour of cameras: small size, light-weight, passive sensors that provide rich information at low cost.

Cameras and IMUs are complementary in terms of accuracy and frequency response.

An IMU is ideal for tracking over
- short periods of time
- motions with high dynamic profile
A camera is best suited for
- state estimation over longer periods of time
- smoother motion profiles

Backlinks