Relative Computer Vision-Based Navigation for Small Inspection Spacecraft

摘要

This Note presents an approach for relative navigation between two spacecraft using only computer vision information from a planar fiducial target. This approach was designed specifically to work for a small inspection spacecraft that has a limited amount of physical space for a fiducial target, but is also applicable to other types of systems with similar constraints. The difficulties associated with using the minimal number of point features were discussed along with how this drove the selection of the globally convergent iterative solution to the perspective projection exterior orientation problem. A multiplicative extended Kalman filter approach was used to "smooth" the measurements with rigid-body dynamics, estimate linear and angular velocities, and to implement probabilistic outlier rejection. The principal contribution of this work is the combination of these two methods in such a way that works reliably for four coplanar fiducials, which is the minimum number. Experimental testing was performed using SPHERES with a computer vision upgrade and a four-point fiducial marker. An analysis of the experimental results leads to the conclusion that the described fiducial-based navigation system provides an upper bound position, orientation, and linear and angular velocity estimates that are on the same order of magnitude as the SPHERES ultrasonic global metrology system that was used as the "ground truth" relative navigation system.