Honestly, it sounds really, really complicated.
If you can make them speak, differentiating voices can work. Although two days ago I had some kind of throat, and it sounded like a frog.
But trying to differentiate with some pattern recognition will be tough. Imagine that the system would have to say that a man in a long coat with long hair is a man, and a woman with one of these women-flaps with a bulky sweatshirt and jeans is a woman that a cowboy hat is not long hair - it seems very difficult to do with any rule system.
You may be able to do machine learning, apply it to millions of images, but I am skeptical about too big a pose / clothes / etc.
I am an American, and I was asked several times in western clothing in India why "I was in men's clothing." My friend, an American woman, was dressed in an embroidered top, which no one wore in the United States. But I found out that it was men's clothing (kamiz) in Pakistan.
And think about the social consequences of guessing the "wrong." Unfortunately, most women often encounter this experience, since the man is defaulted, but at least in the USA, assuming that the man in front of your kiosk is a woman and is trying to sell him panties or something that will cause a negative reaction with side of many men.
Finally, I wonder what “men's products” and “women's products” are. The last 3 non-food items I purchased were “a gallon of motor oil, a USB cable and batteries,” and I'm a woman. I think that the jockey shorts are “men” and the tampons are “women,” but in fact, do you think anyone wants these objects to flash on a giant monitor as they pass by?
What about a more fun, engaging and technically feasible experience? - how to drop product photos, and let people swim in space - so if I swim to the table saws, I get carpenter material. And if I go to the pram, they offer me baby food. Perhaps I am a woman who does her own redistribution, and perhaps I am a new dad and need a pram.
This whole project looks like a social disaster awaiting a place.