Over the last few years, the Visual-Inertial navigation systems have attracted considerable attention mainly due to the higher accuracy that is promised by the so-called tightly-coupled scheme where visual and inertial data are integrated at a low level in a common estimation problem. However, the calibration parameters of the camera (e.g. intrinsic and extrinsic parameters) and of the inertial sensor (e.g. sensor's mounting mis-orientation) are often left to be calibrated offline that makes the developed navigation system far from an off-the-shelf product. In this work, an enhanced tightly-coupled Visual-Inertial navigation system, based on the Multi-State Constraint Kalman Filter scheme is proposed that includes the sensors' calibration parameters in the state list to be estimated along with the navigation states. Experimental results on the KITTI odometry dataset shows a considerable improvement in the odometry accuracy compared to the case where those values are obtained from the calibration file of the KITTI dataset.

Additional Metadata
Keywords Kalman Filter, Sensor Fusion, Visual-Inertial Navigation
Persistent URL dx.doi.org/10.1109/MWSCAS48704.2020.9184501
Conference 63rd IEEE International Midwest Symposium on Circuits and Systems, MWSCAS 2020
Citation
Sheikhpour, S. (Soroush), & Atia, M. (2020). An Enhanced Visual-Inertial Navigation System Based on Multi-State Constraint Kalman Filter. In Midwest Symposium on Circuits and Systems (pp. 361–364). doi:10.1109/MWSCAS48704.2020.9184501