Inertial preintegration for VI‐SLAM by the screw motion theory

Abstract

Most smoothing‐based visual‐inertial simultaneous localization and mapping algorithms (VI‐SLAM) rely on the Lie algebra processing of the inertial measurements. This approach is limited in its decoupled update of the attitude by using SO3 and velocity increments by SE3. In addition to limitations on only point transformation between frames. We present a novel approach to handling inertial measurement unit (IMU) measurements between two camera frames by the screw motion theory. Where rigid body dynamics are concisely represented by the compact unit dual quaternion. With this approach, the limitations of point transformation are mitigated by the superior Plücker line transformation and the states update is achieved by a single coupled operation. To harness this consistent framework for a smoothing‐based VI‐SLAM, the screw motion twist parameter is based on the raw IMU measurements. Then, a consistent residual cost function with the corresponding Jacobian and covariance updates is derived for graph‐optimization algorithm respecting the screw motion paradigm. A transition method is proposed to overcome the issues of over‐parametrization by the unit dual quaternion. solving all singularity threats while saving the advantages of adopting the twist operator. Finally, the loftier performance of the proposed algorithms is attested by simulation and real‐world experiments.

A novel multi-sensor hybrid fusion framework

“…In the concept of Industry 4.0, sensor technology plays a crucial role, allowing devices to collect a large amount of data in real-time. At the same time, numerous advanced sensor technology theories have emerged [9][10][11], providing an essential foundation for realizing intelligent manufacturing and the digital transformation of mechanical equipment. Data serve not only as indicators for monitoring device status but also as the foundation for implementing predictive maintenance [12].…”