Flexible generation-on-demand multistation for non-cooperative 6DOF pose measurement

Abstract

Multidimensional recognition of scene targets has been widely used in the perception and control of autonomous units, such as satellite docking, automotive assembly, and robot guidance. In this paper, dynamic virtual measurement stations with desired quantity and distribution are proposed using a rotating-prism-embedded camera, capable of flexible six-degree-of-freedom (6DOF) pose estimation for non-cooperative targets. It includes virtual-multistation-based stereo computational imaging and coordinate-alignment-based efficient absolute orientation. By flexibly assigning stations for multi-viewpoint directional cross-positioning, this framework not only allows the reconstruction of high-quality three-dimensional (3D) points but also is suitable for fast pose extraction of dense large-scale points. Simulation studies show that this method can improve accuracy with more points and focal-length values while suppressing the resulting range loss and time consumption. Experiments demonstrate that the architecture outperforms mainstream cooperative methods in accuracy and robustness, enabling non-cooperative 6DOF acquisition with means errors better than 1 mm and 0.2°.