We want to give retail stores an incentive to have the correct planogram. We want brands to occupy their own shelf, with the brand name facing outwards. We also want to make sure that the shelves are restocked as often as possible.
What it does
We take a picture of the shelf at a set interval. We then use an image recognition AI to see if the shelf is following the correct planogram, if there are empty spots on the shelf, and if there are other brands that are currently occupying our shelf space. If the grocery store offends any of these criteria, we upload the pictures to an internal UI to help the company inspect these stores.
Challenges we ran into
The image recognition AI needs a lot of photos to make accurate predictions. Unfortunately, we did not have enough time to take photos of the fridge. In addition, we ran into problems getting the webcam connected, and therefore most of our pictures in the database consists of photos from our phones, instead of the webcam. Therefore, when we use the webcam to take a picture of the shelf, the accuracy is much lower than expected because it is being compared against the phone pictures (which has higher quality, better lighting, etc...).
Accomplishments that we're proud of
We got the webcam working as well as the image recognition AI. We believe that given enough pictures, the model will be able to give us pretty accurate predictions of whether a shelf is planogram-compliant, empty, presentable, etc...
What we learned
We learned how to leverage hardware, divide responsibilities and still keep communication lines open. We also learned how to use an image recognition AI (Clarifai) to predict the status of the shelf.
What's next for Best Buds For Lite
Since we are using a machine learning AI, the next steps would be for us to keep on uploading pictures to the AI so that it can continue to learn.
Log in or sign up for Devpost to join the conversation.