The International Journal of Robotics Study, Ahead of Publish.
In this paper, we examine state evaluation of multi-visual-inertial systems (MVIS) and create sensing unit blend formulas to ideally fuse an approximate variety of asynchronous inertial dimension systems (IMUs) or gyroscopes and international and/or moving shutter cams. We are specifically thinking about the complete calibration of the connected visual-inertial sensing units, consisting of the IMU/camera intrinsics and the IMU-IMU/camera spatiotemporal extrinsics in addition to the picture readout time of rolling-shutter cams (if made use of). To this end, we create a brand-new analytic mixed IMU combination with inertial intrinsics– described ACI3– to pre-integrate IMU dimensions, which is leveraged to fuse complementary IMUs and/or gyroscopes together with a base IMU. We design the multi-inertial dimensions to consist of all the required inertial inherent and IMU-IMU spatiotemporal external specifications, while leveraging IMU-IMU rigid-body restraints to get rid of the need of complementary inertial positions and therefore lowering computational intricacy. By executing observability evaluation of MVIS, we show that the common 4 unobservable instructions stay– despite the number of inertial sensing units are made use of, and additionally determine, for the very first time, degenerate movements for IMU-IMU spatiotemporal extrinsics and complementary inertial intrinsics. Along with substantial simulations that confirm our evaluation and formulas, we have actually developed our very own MVIS sensing unit gear and accumulated over 25 real-world datasets to experimentally confirm the recommended calibration versus the modern calibration approach Kalibr. We reveal that the recommended MVIS calibration has the ability to accomplish contending precision with boosted merging and repeatability, which is open sourced to far better advantage the neighborhood.
发布者：Yulin Yang，转转请注明出处：https://robotalks.cn/multi-visual-inertial-system-analysis-calibration-and-estimation/
