The calibration of the ATLAS $b$-tagging algorithm in environments characterised by high multiplicity of jets is presented. The calibration uses reconstructed $t\bar{t}$ candidate events collected by the ATLAS detector in proton-proton collisions at LHC with a centre-of-mass energy $\sqrt{s}$ of 13$\,$TeV, with a final state containing one charged lepton, missing transverse momentum and at least four jets. The $b$-tagging efficiencies are measured not only as a function of the most relevant kinematic quantities, such as the transverse momentum or the presudo-rapidity of the jets, but also as a function of quantities that are sensitive to close-by jet activity. The results extend the regions where data-to-simulation $b$-tagging scale factors are derived when using dilepton $t\bar{t}$ events.