Real-time Motion Estimation for UAVs Based on Dissimilar Multi-sensor Data Fusion

Large time-delay usually exists in the visual localization results of unmanned aerial vehicles (UAVs), which lowers the accuracy of motion estimation. To solve this problem, a real-time motion estimation method based on multi-sensor data fusion is proposed. Firstly, attitude information provided by the inertial measurement unit (IMU) is fused into the monocular visual localization algorithm, so that the delay in localization results can be reduced. Then, the delay of the visual localization results is considered during the motion estimation using Kalman filter, and the measurements of the accelerometer are used to compensate the delay. With the proposed method, motion estimation results with high accuracy are obtained in real time. The method is validated by field experiments on a quadrotor system. By comparing the motion estimation results with the method without delay compensation as well as the ground truth data, the effectiveness of the proposed method is verified.