The identification of jets coming from the hadronization of $b$-quarks ($b$-tagging) is instrumental for many physics analyses performed at the Large Hadron Collider (LHC). ATLAS has $b$-tagging capability starting at High-Level Trigger stage, where $b$-tagging algorithms can process reconstructed tracks. Physics processes with $b$-jets in the final state, but no lepton or missing transverse energy, can benefit from using $b$-tagging at trigger level to maximise the signal efficiency. In Run 2, the $b$-jet trigger efficiency is measured using a likelihood-based method applied to data enriched in $t\bar{t}$ di-leptonic events. The $b$-tagging efficiency, for trigger algorithms as well as for the combination between trigger and offline algorithms, is extracted for jets in a transverse momentum range from 35 to 600 GeV, and data-to-simulation scale factors are derived.