Maritime Robotics (Norway)

In several motion control applications, e.g. in precise pointing devices and navigation systems, an accurate estimate of the attitude of a device in the world coordinate system is required. Readily available sensors used to estimate the attitude suffer from drift, magnetic disturbances, or a lack of information about the direction of the geodetic north. In this paper, an algorithm which estimates the attitude of a camera using camera images and a Digital Surface Model (DSM) referenced by GPS is proposed. The algorithm uses shape features of a skyline extracted from a camera image and synthetic skylines rendered from a DSM in order to estimate the roll and pitch angles, while the yaw angle is estimated using grid search. The algorithm was evaluated using roughly 600 camera images captured with varying pitch and yaw angles. The standard deviations from the ground truth, provided by a high precision pointing device, are 0.037° ,0.015°, and 0.018° for roll, pitch, and yaw respectively. The results indicate a higher precision than current camera attitude estimation algorithms using a DSM, while also providing a robust yaw estimate.