We estimate the number counts of line emitters at high redshift and their evolution with cosmic time based on a combination of photometry and spectroscopy. We predict the Hα, Hβ, [OII], and [OIII] line fluxes for more than 35 000 galaxies down to stellar masses of ~109 M⊙ in the COSMOS and GOODS-S fields, applying standard conversions and exploiting the spectroscopic coverage of the FMOS-COSMOS survey at z ~ 1.55 to calibrate the predictions. We calculate the number counts of Hα, [OII], and [OIII] emitters down to fluxes of 1 × 10-17 erg cm-2 s-1 in the range 1.4 < z < 1.8 covered by the FMOS-COSMOS survey. We model the time evolution of the differential and cumulative Hα counts, steeply declining at the brightest fluxes. We expect ~9300-9700 and ~2300-2900 galaxies deg-2 for fluxes =1 × 10-16 and ≥2 × 10-16 erg cm-2 s-1 over the range of 0.9 < z < 1.8. We show that the observed evolution of the main sequence of galaxies with redshift is enough to reproduce the observed counts variation at 0.2 < z < 2.5. We characterize the physical properties of the Hα emitters with fluxes ≥2 × 10-16 erg cm-2 s-1 including their stellar masses, UV sizes, [N II]/H a ratios and Hα equivalent widths. An aperture of R ~ Re ~ 0.5 arcsec maximizes the signal-to-noise ratio for a detection, whilst causing a factor of~2×flux losses, influencing the recoverable number counts, if neglected. Our approach, based on deep and large photometric data sets, reduces the uncertainties on the number counts due to the selection and spectroscopic samplings whilst exploring low fluxes. We publicly release the line flux predictions for the explored photometric samples.