A fast and robust vision-based horizon tracking method

In order to achieve accurate attitude measurement and smooth navigation controlling of unmanned surface vehicle (USV), a visual horizon tracking method based on Kalman filter has been proposed. Firstly, the approximate parameters of the visual horizon are estimated according to the predictive camera attitude, and the candidate image pixels are obtained in a small region of interest around the predicted horizon location. Then, within the candidate image pixels, the candidate line segments are determined by repeated interior points selecting and least squares fitting, and the final visual horizon is picked out according to geometric properties of the candidate line segments. Lastly, the optimal prediction of the camera attitude is updated by Kalman filter. As the horizon detection is only performed in a local region, compared with traditional horizon detection method, the proposed method takes a short time and performs robust to the false alarms. To validate the performance of our method, an experiment is performed based on the frames captured by the camera which is fixed on an USV. The experimental results indicate that our method has higher tracking accuracy and speed.