SoftPOSIT Enhancements for Monocular Camera Spacecraft Pose Estimation Jian-Feng Shi Carleton University Ottawa, Ontario K1S 5B6, Canada Email: jianfeng.shi@carleton.ca Steve Ulrich Carleton University Ottawa, Ontario K1S 5B6, Canada Email: steve.ulrich@carleton.ca Abstract—This paper proposes several enhancements to the softPOSIT algorithm with applications to spacecraft pose estima- tion using a monocular camera. First, the proposed enhancements include a technique for reducing false matches as result of local minimum trapping. Second, this paper provides two strategies for iteration control parameter initialization by using the trace of the correspondence distance, and by using image centroid matching. The method of image centroid matching allows the world model center of geometry to align with the image centroid. The alignment result in reasonable correspondence weighting values used for match optimization. The various algorithm enhancements were tested on 26, 180 simulations with varying geometries and initial pose conditions. Results show a significant increase in accuracy when compared with the original method. I. I NTRODUCTION T HE technology of using camera images for robotic and unmanned aerial vehicle guidance, navigation, and control (GNC) have significantly matured in the recent years. For example, the advances in vision techniques can benefit multi- spacecraft formation flying missions [1] using monocular optical and thermal cameras as proximity operation navigation sensors. Pose estimation using monocular camera images determines position and orientation of the target spacecraft relative to the observer camera coordinate system. Many techniques are available in solving the pose estimation problem although the most advanced techniques based on complex non- geometric features requires significant computational resources and are less efficient for some space applications where the target image can be approximated by simple geometries [2]. A more suitable approach is by finding correspondence between the internal 3D model with the 2D camera image hence inferring the target pose. This is called the model-to-image registration problem or simultaneous pose and correspondence problem. The image point correspondence is also known as the Perspective-n-Point (PnP) problem. [3] For the unknown space vehicle, the focus remains to solve the arbitrary n point problem. Some selected solution to the PnP problem include the use of numerical scheme solutions such as RANdom SAmple Consensus (RANSAC) [4], Iterative Closest Point (ICP) [5], Newton-Raphson Method (NRM) [6], and non- iterative solution such as the EPnP[7] solution that is on the order of O(n). David et al.[8] solve the PnP problem by combining Simulated Annealing (SA) and Scaled Orthographic Projection (SOP) [9]. This method is termed SoftPOSIT (Softassign [10] and Pose from Orthography and Scaling with ITerations). Soft- POSIT is an iterative point correspondence scheme minimizing a global energy function based on the 2D and 3D projection points differences. There has been several implementations of the SoftPOSIT method in terrestrial applications. For example, Jager et al.[11] used SoftPOSIT to determine the pose of a ground vehicle based on thermal camera images, and Diaz and Abderrahim [12] used it to estimate the pose of a spinning spacecraft model. Once converged, SoftPOSIT can produce accurate pose estimation results with low computation re- sources; however, some shortcomings of the algorithm include local minimum trapping and iteration control parameter to correspondence compatibility. This paper proposes several techniques to address the above mentioned issues. First, a novel checking criteria is introduced to find the best initializa- tion orientation. A proper initialization minimizes the chances for the optimization to converge into a local minimum. Second, two strategies are proposed on initializing the iteration control parameter by using the trace of the correspondence distance, and by approximating target object center of geometry with image centroid. Both strategies eliminates non-viable itera- tion control parameter prior to match optimization. Finally, optimization reset conditions will be discussed for practical considerations to numerical anomalies. This paper is organized as follows, Sec. II formulates the SoftPOSIT methodology and provides definitions for frames and algorithm parameters. Sec. III provides the enhancement formulations for global minimum search and strategies for iteration control parameter selection. This includes the deriva- tion of the centroid matching technique. Sec. IV provides descriptions of models and ICs used in the algorithm valida- tion. Sec. V provides results and discussions of the simulation findings. Finally, Sec. VI concludes the study by comparing the enhancements to the original SoftPOSIT algorithm. II. SOFTPOSIT FORMULATION This section provides problem definition and formulates the SoftPOSIT [8] methodology. Define a tracker body F SB equipped with a single camera F VW pointed towards a target object F CB . The frames F SB and F CB are located at the object Center of Geometry (COG). The camera frame F VW has its z axis pointed outwards from the boresight of the