How does the teaching with only one image work?
The convenient application can be explained in 2 steps: You take a picture of your dish and the visioncheckout recognizes it accurately and reliably during checkout. So fast, so good.
But what most people really mean when they ask us this question is: “How can the visioncheckout recognize all dishes if it has only seen one picture of them?” To answer that, we need to elaborate a little further.
The magic behind our AI
During our development phase, we trained an algorithm with hundreds of images of dishes, glasses, drink bottles, retail goods and much more. You can picture it like this: For every single picture, we told the visioncheckout where which dish is located, “Here is a burger”, “This is a lasagna”, “There are noodles”, until the algorithm recognized where dishes are located and where the relevant items are on a new picture. As a result, it has become so good at recognizing identical dishes that only one image is sufficient to assign the matching images.
This is how the visioncheckout sees
But what if the plate placed under the camera is not exactly as in the example picture? The visioncheckout filters out the information that is most important to him from the photo: the plate, and the meal on the plate. These three pictures therefore all look the same to him:
What do I have to pay attention to when teaching?
All items that are to be sold during the day must be listed in the menu, which is automatically imported from the POS system. If an item or dish has previously been sold, the pictures are already saved. In this case, you don’t need to worry about anything and can get started right away.
When a dish is offered for the first time, it is highlighted red in the auvisusmanager and you cannot yet activate the menu. A bit like: “You want to sell stew today, but I don’t know what it looks like yet. Please teach it to me first”. To do this, simply take a picture of the dish. As soon as a picture is saved for each dish, you can get started.
What if a guest wants to change the side dish?
No problem, we have a solution for that too. In the morning, when teaching the dishes, all you must do is take photos of 3-5 variations and the AI learns to concentrate on the main component. This way, the dishes are recognized accurately and reliably even with a free choice of side dishes.
So, in the eyes of the visioncheckout, plates that have turned…
…all look the same.
Does the recognition also work with large and small portions?
Recognizing large and small portions is also no problem. Simply teach an item for both sizes and the visioncheckout can distinguish between the two dishes and books the correct size into the POS system.
Do I have to label all items individually?
Can I also implement the visioncheckout in our company kiosk?
The visioncheckout can be used in your company kiosk and recognizes all taught articles with the help of computer vision without having to see the barcode.
However, there is a small peculiarity here: unlike plate dishes, the packages of retail articles have at least two very different sides. Therefore, a photo of each side of the article must be taken for the learning process, which the visioncheckout can use as a reference image. For a chocolate bar, it would look like this:
For all those who now think that this is too time-consuming: don’t worry. The visioncheckout saves all images, so they only have to be taught once. Since the assortment in kiosks is usually rarely completely replaced, but rather supplemented, the use of the visioncheckout is also worthwhile. In addition, the visioncheckout gives you new opportunities to use your working hours in a targeted way. If the visioncheckout takes over the cash register during off-peak hours, you can extend your opening hours without additional effort and thus offer your customers a better service.
Is there a limited number of articles?
From a practical point of view: a clear no. In any case, more articles can be saved in the visioncheckout than you can sell. To theoretically provide an infinite number of articles, we would need infinite computing capacity. And how can we put it: that doesn’t exist. Just as there are no infinite grains of sand on the beach. But counting them would still take quite a long time.
What if the prices change?
Then they also update automatically in the visioncheckout. Since all items are imported directly from the POS system into the auvisusmanager,prices are also transferred and always remain up to date. The payment process is still a transaction directly between the POS system and the customer, the card reader is connected to the POS, not to the visioncheckout. If the correct price is saved in the POS system, the correct price will be displayed and debited.