Price tags tend to be small. Hard to read from any significant distance, unless you’ve got a really high res camera. In which case you’re going to be drowning in data as you try to process all those huge video frames.
Then there’s the issue of occlusion, where higher shelves may block the ability to see price tags on lower shelves, and vice-versa. You need to take pictures at multiple different heights to be sure you’ve got coverage. This also includes what do you do if someone is blocking your picture of that shelf at the moment, so that you have to either sit there and wait until they move, or you go ahead and move on and then come back.
Think about breaking the process up into multiple parallel stages. How many cameras can you stack together to capture multiple different heights simultaneously? How many sides can you scan at the same time?
How much intelligence can you build into the camera rig so that it not only captures the pictures of the products, but also accurately captures, recognizes, and OCRs the values on the tags?
What about location information? How do you accurately record precisely where you took which picture, so that you can properly place that image relative to all the others, and so you know what is located where?
How about recognizing the content of the picture that you took? Sure, you might know what’s on the price tag, but can you be sure that’s the right product with that tag? This also relates to how can you tell if a product has been mis-placed, perhaps by someone who took it off the shelf and then put it back on the shelf in the wrong place?
Look very closely at the Amazon Go stores and the kinds of solutions they have to have in place.
