Quaternion-based Orientation Estimation Fusing a Camera and Inertial Sensors for a Hovering UAV

Orientation estimation in quadrotors is essential for low-level stability control and for high-levelnavigation and motion planning. This is usually carried out by fusing measurements from different sensorsincluding inertial sensor, magnetic compass, sonar, GPS, camera, etc. In indoor environments, the GPSsignal is not available and the Earth’s magnetic field is highly disturbed. In this work we present a newapproach for visual estimation of the yaw angle based on spectral features, and a fusion algorithm usingunit quaternions, both applied to a hovering quadrotor. The approach is based on an Inertial MeasurementUnit and a downward-looking camera, rigidly attached to the quadrotor. The fusion is performed by means of an Extended Kalman Filter with a double measurement update stage. The inertial sensors provideinformation for orientation estimation, mainly for roll and pitch angles, whereas the camera is used for measuring the yaw angle. A new method to integrate the yaw angle in the measurement update of the filteris also proposed, using an augmented measurement vector in order to avoid discontinuities in the filterinnovation vector. The proposed approach is evaluated with real data and compared with ground truthgiven by a Vicon system. Experimental results are presented for both the visual yaw angle estimationand the fusion with the inertial sensors, showing an improvement in orientation estimation.