A New Loose-Coupling Method for Vision-Inertial Systems Based on Retro-Correction and Inconsistency Treatment

摘要

Real time pose estimation of a mobile rigid-body in an unknown environment and without adding constraints (markers, antenna, ultrasound, Radio Frequency IDentification (RFID)...) is a crucial issue for Augmented Reality (AR) applications. One of the most advanced indoor/outdoor pose estimator is the Simultaneous Localization and Mapping algorithm (SLAM) based on monocular or binocular images. The complexity of this algorithm and his processing time present the main drawbacks of this type of pose estimator. It is difficult to use SLAM on mobile or embedded devices in real time applications, because they suffer from low computational resources. On the other hand Iner-tial Measurement Units (IMU) allow indoor/outdoor localization without, important time processing but require signal integration which generates drifts over longer periods of time. In this paper we propose a new method for coupling SLAM with IMU. In this approach we take into account the SLAM processing time in order to avoid incoherences of timestamps in fused SLAM and IMU poses. To this end we propose a retro-correction method that synchronizes all poses with a general clock timestamping all events. In addition the quality of the poses is improved with detection and treatment of inconsistency. For this purpose an Adaptive Complementary Filter (ACF) was developed. Finally simulated and experimental results validate the efficiency of the proposed method.