Robust Range Camera Pose Estimation for Mobile Online Scene Reconstruction

We present a multi-sensor camera tracking method for a real-time 3-D reconstruction on mobile devices. Our approach combines the iterative closest point (ICP) pose estimation using the low-resolution range maps delivered by a PMD pico-flexx time-of-flight camera with the 6-DOF pose estimates given by inertial tracking. In contrast to prior approaches, we do not rely on additional sensors, such as 2-D visual odometry. We fuse the results of the inertial tracking with the extrapolated ICP pose estimates using an extended Kalman filter (EKF). Subsequently, the output of the EKF is used as initial guess for the next ICP-based pose estimation. This approach yields an efficient and robust pose tracking for the spatially and temporally low-resolution range data given in mobile applications, and, at the same time, it results in a consistent geometric reconstruction, as the final pose minimizes the error with respect to the scene geometry.