COMPUTER VISION FOR REAL-TIME RELATIVE NAVIGATION WITH A NON-COOPERATIVE AND SPINNING TARGET SPACECRAFT Setareh Yazdkhasti, Steve Ulrich, and Jurek Z. Sasiadek Carleton University, Ottawa, Ontario, K1S 5B6, Canada ABSTRACT Relative navigation for spacecraft has received a great at- tention recently because of its importance for space ap- plications. Many space missions require accurate relative position knowledge between space vehicles. This paper proposes a vision-based relative navigation system only based on stereo gray-scale camera. An effective Image processing algorithms was developed for estimating the relative position between non-cooperative, and spinning space objects. The experimental test-bed was used for the simulations of service missions of satellites to test the performance of the developed vision and navigation algo- rithms. Simulation results in a dark room with spinning scaled model of spacecraft, demonstrated the effective- ness of the proposed algorithm. Key words: Uncooperative spacecraft; Stereo camera; Image processing. 1. INTRODUCTION In reality, a majority of the failure satellites are spin- ning or tumbling in an orbit without light beacons and available state information which called non-cooperative satellites. When the satellites are non-cooperative, with- out available interactions between non-cooperative vehi- cles, the problem of relative state estimation becomes more complicated. Thus, the pose determination of a non-cooperative satellite becomes a tremendous chal- lenge. Over the last decade, vision-based navigation sys- tems have been extensively used to address the prob- lem of relative motion determination, due to their low cost, mass, and power requirements, compared to active sensor-based techniques (e.g., laser range finder) [1]-[2]. Various researchers have investigated on pose estimation of an object and some of these techniques use vision- based sensors for measuring the motion of targets in the space environment. Their methods vary depending on their research objectives and assumptions. In particular, laboratory experimentations of cooperative vision-based navigation systems relying on known fiducial markers in- stalled on the target vehicle were recently reported by Romano et al.[3] at the Naval Postgraduate School, and Tweddle and Saenz-Otero [4] at the Massachusetts Insti- tute of Technology (MIT). Such cooperative vision sys- tems were also extensively used on actual on-orbit mis- sions. For example, the Space Vision System (SVS) [5] monitored and tracked a pattern of special dots installed on the International Space Station. As the Space Station moved, the system tracked the changing position of the dots, and calculated the relative motion between the two vehicles. However, when the target object has no fiducial markers, such vision systems cannot be used. It is there- fore required to develop a strategy that does not rely on such markers, i.e., a relative navigation system applica- ble for unknown, uncooperative, and possibly spinning, target spacecraft. Most of existing work in the literature that address the problem of uncooperative relative vision- based navigation assume the existence of a CAD model of the unknown object. For example in [6] the algorithm used stereo camera and 3D model matching, Then ap- plying matching algorithm called iterative closest point (ICP), in order to match Existing 3D model to 3d mea- surement point obtained from stereo matching, and uses time series of images to increase the reliability of the rel- ative attitude and position estimates. In [7] the ICP algo- rithm implemented to register range data and model from stereo camera. In [8] stereo image of satellite model with white mat surface on the model took and the ICP algo- rithm was applied to matched known model to measure data points. Geometric Probing combining a voxel tem- plate set and a binary decision tree has been proposed to improve the weakness of ICP of potentially falling into a local minimum depending on the initial relative attitude between the model and measured data [9]. However, this method is suitable when prior knowledge about the geometry of the satellite is available, without such knowledge, the problem become more difficult. An- other disadvantage of ICP algorithm is convergence to false local minima. Tomashi et al. [10] developed a fac- torization method. This method takes multiple images taken by monocular camera to measure the pose. But it also requires prior knowledge, such as the shape and di- mension or the proportional dimension of the satellite. In addition, when the pose parameter is calculated, so- lutions of nonlinear equations will be involved, which makes computation more complicate. A vision-based rel- ative navigation and control strategy for inspecting an un- known an, no cooperative object in space using a stereo