Fast horizon detection for airborne visual systems

In this paper a fast horizon detection algorithm is introduced, which is run on an airborne visual system. It is important to run the horizon detection as fast as possible, and also with low power consumption. Hence we developed an algorithm, which has low complexity. The horizon detection uses attitude information provided by the flight control computer of the aircraft and it corrects the horizon based on the visual information from our airborne camera system. All calculations are run onboard, and the detected horizon is used in the sense-and-avoid task for sky-ground separation. The performance of the algorithm is measured in flight tests, which shows that it has superior performance compared to the horizon calculated from the attitude provided by the autopilot and also compared to the horizon calculated from the Euler angles given by our Kalman filter based attitude estimator. The measurements were run on videos from two cameras next to other sensors' information, which were recorded in November 2015. The videos were annotated by hand, which were used as a ground truth.