Visual-inertial odometry system with simultaneous extrinsic parameters optimization

For visual-inertial odometry in a robotic system, the performance of the resultant localization can be improved with effective initialization. In this paper, an initialization pattern is proposed based on the monocular visual-inertial system, which can optimize extrinsic parameters between the camera and the inertial measurement unit simultaneously. Firstly, rotation extrinsic, gyroscope bias, metric scale, and gravity vector are estimated by visual-inertial information. Then, the translation extrinsic parameters are optimized in visual-inertial alignment stage and the velocity is calculated via previous state estimation. Since accelerometer bias has little effect with respect to initialization when the rotation angle is not large enough, it is not estimated in the initialization stage and is handled in the following stages. The performance of the proposed method is validated by comparative experimental results.