Enhancing indoor navigation accuracy using a multiple camera configuration and IMU data

Abstract Visual Odometry (VO) plays a vital role in indoor navigation and modern architecture, particularly with the increasing interest in low-cost navigation systems. This study proposes a multi-stereo VO system exploiting the integration of multiple cameras and Inertial Measurement Unit (IMU) measurements. Different positioning and integration algorithms are tested to investigate the performance that can be provided by each algorithm. The testing process utilizes real field data over an indoor closed path of 419 m length. These data were collected by a portable configuration that included four cameras and an IMU. The results show that using four cameras for triangulation can significantly improve the navigation performance. In addition, the best performance metrics can be provided by integrating magnetometer and gyroscope heading information through an Extended Kalman Filter (EKF) to combine with the pose changes derived from the multi-stereo VO. Along the tested trajectory, this technique can achieve a mean absolute error of 0.41 m and a Root Mean Squared Error (RMSE) of 0.43 m. The closing error can be maintained at 0.18 m, while the error in the navigated distance is limited to 0.47 m (0.11%). These promising results indicate the reliability and robustness of employing the proposed low-cost system in indoor navigation.