I need to count the number of facings that different products have in a supermarket shelf. Instead of trainning the model on product image samples I was wondering whether using the texts on product packages (and maybe their relative position x,y) has been tried before.
The algorithm could use OCR to read product texts from different frames of a video (in my case the source is a video) and then try to match the texts retrieved to a list of apriori standard product descriptions.
i'm assuming that other features (like color) are not relevant in this case.
Anyone has heard of something like this before?
