Detecting exoplanets by direct imaging is a difficult task that systematically requires the use of advanced post-processing algorithms in addition to an efficient adaptive optics and coronagraphic system. The upcoming thirty meters class telescopes will enable exploring the inner stellar environments where the typical contrast levels to reach will require long exposure times only achievable by combining several observations conducted days, weeks, or months apart. For such observations, the orbital motion of exoplanets will no longer be negligible, and a proper orbital modeling will be crucial to combine multi-epoch observations without drastically degrading the detection confidence and the achievable contrast. In that context, we recently developed the Pacome algorithm which efficiently combines observations of the same star while accounting for the Keplerian motion of the sought exoplanets within an end-to-end statistical detection formalism. The sensitivity and reliability of the proposed method constitute major advantages in the field to detect new companions at a statistically grounded confidence level. In this paper, we consider a case-study example of thirteen VLT/SPHERE-IRDIS observations of the HD 95086 system, and we illustrate the efficiency of the Pacome by (re-)detecting HD 95086 b with a very high multi-epoch signal-to-noise ratio of 53.7. We also derive its orbital elements posterior distributions.