The performance of a direction-based Bayesian filter in the orbital tracking problem

The space debris tracking problem from a series of angles-only observations can be viewed as an example of Bayesian filtering. Bayesian filtering is easy to implement if the joint distribution of the state vector and the observation vector is normally distributed. Under Keplerian dynamics, the propagation of an initial normally-distributed point cloud does not tend to remain normal in various standard coordinate systems. Hence we propose using an “adapted structural (AST) coordinate system”, which preserves approximate normality much more successfully. We analyse the performance of a Bayesian filter in this new coordinate system.