Indoor localization and mapping using camera and inertial measurement unit (IMU)

Abstract

This paper presents a monocular camera and inertial measurement unit (IMU) fusion technique using Extended Kalman Filter (EKF) with delay in landmark initialization to address the simultaneous localization and mapping (SLAM) problem for single smartphone. The dynamic model of the EKF is chosen to be constant acceleration while the velocity of the system is constantly monitored in order to have enough overlap between consecutive camera frames. Moreover inconsistency in SLAM algorithm due to heading error is removed by utilizing magnetometer measurement model. The use of data association technique ensures that the final map solution is robust and consistent even in complex environment. For fast and robust features matching, the Speed-Up Robust Features (SURF) extraction algorithm followed by random sample consensus (RANSAC) method is applied on camera frames. The extracted features from SURF algorithm are related to ground plane, since the system moves parallel to the ground. The experimental results illustrate the performance of the monocular-IMU SLAM over long walked trajectories in indoor environment.