We can already see where in the store you are.
It’s much easier and cheaper to just record what shelve/height/area products are placed in and reconcile that data with sales. The data will be more noisy, but you can capture so much more valuable information for the same investment. You can also usually apply the models in retrospect, since most larger chains already record where products are supposed to be.
Cameras are really only used at corporate test stores in small numbers. Camera data is usually looked at manually and only focuses on a specific product or area.
I've worked on systems designed to send in-store customers custom, extremely time limited (< 1 hour) coupons based on where they are in the store and where they linger. Linger time being used as an indicator of purchase intent for a specific item. This does require that the customer has the store's app installed and running.