I'd start with a tiny dot of fluorescent paint on the top or bottom of the bee. Any time a bee comes through a "gate", a short flash of UV light would make these dots glow. One or more cameras could then capture the "scene" at the gate, and it would be easy enough to detect the presence or absence of such glowing dots. With a narrowband filter on the cameras, it'd be so simple that no AI would be needed, and you could probably make it work with tech available in the 1980s or earlier.
To differentiate bees from different colonies, mix dyes that glow in different colors in various ratios. Such mixtures would be added to the base paint, and a wide variety of mixture ratios could be detected with a calibrated camera setup.
The setup for all this would be rather simple, since I imagine that bees tend to fly right-side-up, so you don't need UV illumination and imaging all around the gate, only from the top or bottom - depending on where you chose to put the paint dots.
Putting paint dots on the bees would be finicky work, but still easier than messing about with gluing things to bees.
In case bees can see a part of UV spectrum, a broadband UV source would need to be filtered to exclude the "color" range that the bee can see, so as not to blind the bee temporarily due to a bright flash.
In place of a UV-fluorescing dye, ferromagnetic particles - e.g. the powder that neodymium magnets are made from - can be degaussed and mixed into paint. A precisely dosed tiny blob of such paint is deposited on each bee. Several paints with various concentrations of the particles are used to differentiate the colonies. A solenoid is installed around the gate, and connected to a custom impedance analyzer. Any time a bee with ferromagnetic paint on it passes through the gate, the quantity of ferromagnetic material can be quantified. That's assuming that the bees themselves don't contain a comparatively significant amount of naturally ferromagnetic molecules that would obscure any signal from a tiny blob of paint.
In keeping with the paint marking idea, you could use paint with one or more metal powders mixed in - such as Ni, Al, Cu, Zn, etc. These can be mixed in varying ratios. A custom XRF (X-Ray Fluorescence) system - unfortunately costing a small fortune - would produce a short but intense X-ray flash each time a bee passed trough the gate. The metals in the marker paint fluoresce in X-rays when exposed to X-ray excitation. One or more X-ray sensors would capture the energy spectrum of the fluorescing X-ray flash thus produced. A signal analysis system can then determine the ratios of elements within the sample. The metals used would need to be different than whatever occurs naturally in bees. E.g. I'd imagine there's a tiny bit of calcium in most living things, and calcium is a metal that XRF will pick up. It is another question entirely how many of such "detection flashes" of X-rays would be a lethal dose for a bee, and how much engineering time it would take to design such a system. Making a detector system that would be sensitive enough for this to work could be an "interesting" engineering problem. I'm not sure how feasible it is, but I'm sure with enough money it could be done.
The above ideas all used permanent marking. We could instead use transient marking.
One idea: see what sort of "interesting" molecules are in a bee, "interesting" by the fact that they have long magnetic, paramagnetic, or electron resonance relaxation times (minutes or more). As a bee goes through a gate that is the bore of a magnet in a small resonance system, the direction it goes (in vs out) is detected. If the bee is going out, the molecules of interest are excited. Then as the bee comes back in, a detection is made of the state of those molecules. By targeting different molecules and different resonant modes, a previous "marking" can be read back and identified. This would likely be expensive too, although at least you wouldn't be dealing with harmful X-ray flashes. On the other hand, there may be none such molecules naturally present in the bees :(. It also would be very disruptive if bees use magnetic field orientation for navigation. Passing through the bore of a magnet with 1T+ flux density would screw that up royally.
Yet another approach that may be feasible would be acoustic detection of modifications made to a bee's wings. Whether this is permanent or temporary depends on whether bees can heal small holes in their wings. To mark a bee, a fast camera system identifies the position and orientation of the wings, and a tracking laser shoots one or more small holes in one or more wings, in predetermined locations on the wing. Those holes, presumably, would slightly change the sound the bee makes when flying, and could be picked up via microphones installed in the "gate" of a hive, and identified through signal analysis. A less advanced implementation of this idea has one somehow temporarily immobilize a bee, and put some sort of a thin metal plate/foil between the bee's body and its wings. A laser coupled to a stereo microscope is then aligned with the wings, and blows the necessary tiny holes. This forgoes an expensive and complicated fast camera and laser tracking system, trading it off for manual labor.
With a microphone array distributed in a 3D volume of interest, you could track individual bees that way within the volume. Modern audio signal processing can isolate of individual sources of sound within fairly large volumes, and bees are noisy. You could use that to track interactions between the bees too - even without marking the bees, although in your case differentiating individual bees is a goal, so they'd need to be "acoustically marked" as above.
Instead of using a laser to modify the wings of a bee, it could be used to etch a tiny 2D barcode somewhere on the surface of the bee, e.g. on the back of its head. With careful design, the amount of energy can be fine-tuned to make a mark that is semi-temporary and doesn't really damage the bee. The marking system would need to track the bee as it goes through the gate. Or it could be used on an immobilized bee, in a more manual process.
A suitable illumination setup would allow a camera to see those barcodes well enough to read them.
Yet another temporary marking would involve putting a small drop/mist of a non-harmful, somewhat volatile compound on the bee that can be easily "sniffed" by a fast continuous-flow detection system. Air is sucked in from the gate area into the detector, and detection of compounds of interest is made. The compounds would be selected to maximize the sensitivity of the detection system, and would need to be mixed with a volatile carrier that is also a solvent for those compounds. That way they would be constantly evaporating from the bee. How to select such compounds to be non-harmful, and what sort of detection system to use, is a whole another story. There are continuous gas analysis systems that are very sensitive to certain compounds and can react fast enough, but - as with most other approaches - domain experts would need to be consulted to figure all of this out, and that's not cheap no matter what method you choose.
In Soviet Russia (tm), you could mix small amounts of various radioactive isotope powders or liquids into the paint. Each of those has a unique spectrum of energies of the decay products, which include primary electrons and primary gamma rays. An electron energy spectrometer (for beta rays) and/or a gamma ray spectrometer (for gamma rays) would be used to read out the radioactive signature of each bee, i.e. the ratios of various isotopes mixed into the paint the bee was marked with. Those isotopes would need to have fairly short half-lives, so that they'd emit many particles during their decay - enough that a bee "passing through" a gate could be thus identified. On the other hand, the concentration of the isotopes would need to be low enough not to harm the bee. Last thing we want is bees getting radiation sickness [shiver].
For extra credit, you could use isotopes with decay chains that end in a particle or ray pair. Say the isotope decays producing a positron. That positron won't travel long before it encounters an electron, annihilates, and two gamma rays are produced. That's how PET imagers work. These rays are coincident in time and their origin is coincident in space. With a sufficient number of expensive detectors and even more expensive instrumentation, you could track individual bees in a volume of interest that way. Double extra credit if you there's an experiment (a big detector system) at CERN that could do this job without modifications. Their health & safety would have a field day with bringing multiple bee colonies into the experiment's building. At least the buildings housing the sensors are pretty large so in theory you could bring in a bunch of pallets with soil and flowering plants, to make the bees happier.
In the western world, with sufficient budget, you could pull that off too, but you'd potentially need to destroy all the bee colonies used in the experiment to prevent dispersal of toxic radioactive isotopes, even in small amounts :(. And that just breaks my heart, and likely yours too...
