-
-
Processing Stella with light fixture.
-
Shelf simulation ready for beer analysis.
-
Preparing screen for the mount.
-
Static camera with fisheye camera example implementation.
-
Raspberry Pi + screen cardboard mount
-
Top-view bottle view used for analysis
-
Setting up cooler enclosure prototype.
-
Preparing for Shock Top analysis.
-
Underneath view camera and light
-
Custom Raspberry Pi + Touchscreen mount. Thanks Domino's Pizza!
-
GIF
Snapping pictures and uploading with the Pi
-
Bottle branding detection
-
Product description
-
Website homepage
-
Software used
-
Current beer shelf status
Inspiration
In order to help stores improve beer sales we believe it is important to collect as much data as possible about the quantity and variety of beers available for purchase.
What it does
Shelfie will monitor the bottles on shelves to track availability. Our solution is able to analyze individual bottles using machine learning to determine what types of beverages are on the shelf.
How we built it
We attached a camera onto a Raspberry Pi which we then mounted onto a cardboard box that would act as the cooler unit.
The Pi and screen are mounted using a custom cut cardboard attachment from a used Domino's pizza box. We also attached a bright light to the Pi that we have mounted to the inside of upper box.
Our Pi runs an application that takes a picture every second, uploading it to Dropbox.
We then have a server running OpenCV which takes those images, processes them by taking in the tops of the bottles and then analyzes them to see what type of bottle it is. In order to train our AI, we went through multiple iterations and cuts of the photos to reject or accept the images as a certain brand.
The status of what bottles are available is then propagated through the cloud to our web UI where we display how many bottles of each type are in the cooler system.
Team Members
Asheik Hussain
Saj Arora
David Maiman
Also thanks to Steven Yoo who was here in spirit.
Log in or sign up for Devpost to join the conversation.