Terminal docking is an important step towards long-term underwater residency of Autonomous Underwater Vehicles (AUVs). An important part is to correctly estimate the relative position between the AUV and the docking station. While there are many solutions to this problem, it is unclear how they perform with respect to each other in terms of accuracy and computational performance. We propose a side by side comparison of a Rao-Blackwellized particle filter (RBPF) with a Maximum-A-Posteriori (MAP) method in a vision-based terminal homing scenario. Both methods are evaluated in a simulation study based on performance under different uncertainties. Subsequently, they are validated using real-world data from field tests. The comparison shows that in the simulation study, the smoothing performs more accurate than the RBPF, whereas on the experimental data, they perform equally. However, the smoothing requires less computational power compared to the RBPF.