Investigation of multiple-baseline stereovision for state estimation of unknown dynamic space targets

Estimating the pose, motion, and structure of noncooperative dynamic targets using on board sensors is a challenging problem. This work suggests using multiple-baseline stereovision for noncooperative dynamic target relative pose, motion, and structure estimation, with a particular emphasis on space applications. A computer-vision feature-matching algorithm is designed, which produces input data for a recursive filtering algorithm. A newly-developed initialization scheme is proposed, which decreases the ambiguity in the target center of mass location. An extensive numerical study is provided, comparing between two estimators and different relative motion models. It is shown that a simple kinematic model can work well when combined with initial structure estimation. The effect of varying the number of cameras and stereo-rig geometry is investigated. The proposed vision-based relative motion estimation method was validated at the Technion's Distributed Space Systems Laboratory. A scalability analysis indicates that the proposed method may be potentially useful for space applications.