State estimation and reconstruction of non-cooperative targets based on kinematic model and direct visual odometry

For automated space on-orbit servicing missions, it is essential to estimate the states of the target and reconstruct the 3D model, especially for completely unknown non-cooperative targets. Simultaneous location and mapping (SLAM) or visual odometry (VO) technology is suitable to solve this problem, but there are still some limits due to the space environment. Considering the motion pattern of space targets, this article analyses the kinematic model and defines a specific and useful application scenario. Then proposes an algorithm comprehensive using kinematic information and a direct visual odometry method. The experimental dataset is obtained by Unity3D simulation software. And the results elucidate the effectiveness of our approach in estimating the model and pose of the non-cooperative target.