Calibration of the drift error in GPS using optical flow and fixed reference station

In this paper, a sensor system using feature point tracking and optical flow to measure visual odometry is proposed to remove the GPS drift error of the hovering or landed UAVs without using an expensive INS sensor. A method of feature point tracking closest to the center of the field of view is used to reduce computational load. Measured pixel difference from the center and tracking point is converted to real displacement distance with aid of the ultrasonic range sensor. Tracking single point at a fixed position is done to measure internal error of the sensor. Whole system fully operates with low error range under a meter scale with average of 31 fps on ARM Coretex-A7 CPU-based commercial low-cost embedded Linux board. Sensor fusion with the GPS and the reference station installed at a fixed, known point helps initialization of the visual odometry, which successfully removes GPS drift errors.