Answers are grounded in this paper and cite their sources.
Prior velocity and acceleration distributions are often applied in one point initialization, where the track state is initialized based on a single measurement. This distribution is often approximated by a (first two moments preserving) Gaussian distribution. These parameters are derived as a function of the problem dimensionality and target trajectory assumptions.
No takes yet. Share an insight, caveat, or question.
Mušicki et al. (2013) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: