Fusing visual tags and inertial information for indoor navigation

Abstract

We present a navigation system based on a monocular camera and an inertial measurement unit. The system detects visual tags and fuses the measurements on the image plane with inertial signals to perform pose estimation and localization using a Sigma-Point Kalman filter. The tags are detected by edge-based feature extraction and channel codes. During periods in which tags are not visible, epipolar constraints, arising from past views, are exploited to significantly reduce the position error growth rate. The experimental results in an office building indicate capabilities for indoor navigation.