Context. Lunar occultations (LO) provide a unique combination of high angular resolution and sensitivity at near-infrared wavelenghts. At the ESO Very Large Telescope, it is possible to achieve about 1 milliarcsecond (mas) resolution and detect sources as faint as K approximate to 12 mag. Aims. We have taken advantage of a passage of the Moon over two crowded and reddened regions in the direction of the inner part of the Galactic bulge to obtain a high number of occultation light curves over two half nights. Our goal was to detect and characterize new binary systems, and to investigate highly extincted and relatively unknown infrared sources in search of circumstellar shells and similar peculiarities. Our target list included a significant number of very late-type stars, but the majority of the sources was without spectral classification. Methods. An LO event requires the sampling of the light curve at millisecond rates to permit a detailed study of the diffraction fringes. For this, we used the so-called burst mode of the ISAAC instrument at the Melipal telescope. Our observing efficiency was ultimately limited by overheads for telescope pointing and data storage to about one event every three minutes. Results. We could record useful light curves for 184 sources. Of these, 24 were found to be binaries or multiples, all previously unknown. The projected separations are as small as 7.5 mas, and the magnitude differences as high as Delta K = 6.5 mag. Additionally we could also establish for the first time the resolved nature of at least two more stars, along with an indication of circumstellar emission. We were also able to put upper limits on the angular size of about 165 unresolved stars, an information that combined with previous and future observations will be very helpful in establishing a list of reliable calibrators for long baseline interferometers. Conclusions. Many of the newly detected companions are beyond the present capabilities of other high angular resolution techniques, but some could be followed up by long baseline interferometry or adaptive optics. From estimates of the stellar density we conclude that statistically the influence of chance alignments appears to be negligible. We infer that most cases are probably giant-dwarf pairs.
