Congratulations to School of Computing students Sejal Gupta, Nicole Green, Hari Krishnakumaran, and Dhrue Raja, who won the Best Poster prize, chosen by staff and students, at this year’s Computing Poster Fair.
Their project, Recycling Made Easy, was produced by the team as part of the COMP6000 Group Project module and was supervised by Dr Daniel Soria.
Recycling Made Easy aims to create an application that can inform the user how to correctly recycle their household waste. The app, entitled Dumped, is targeted towards University of Kent students living off-campus , as well as Canterbury residents. The user scans either the barcode or the recycling symbol of the desired item to be recycled, and the app shows them which bins are appropriate and the correct recycling information according to the correct local council regulations.
The app uses machine learning in two ways: 1) by detecting the symbols shown by the user, and then presenting the user with a pop-up/new screen informing them of how to recycle the item appropriately and 2) by detecting the barcode information, followed by a popup of information about that item. All of this information is received from a database populated with the item name, barcode and symbols. The application has additional features, including the calendar which shows local waste collection dates, an extensive list of different recycling symbols, and a map to show real-time directions from the user’s current location to the nearest recycling location.
The iOS application uses a CoreML model that has been trained with a comprehensive data set of recycling symbols to ensure the greatest level of accuracy. The model then performs a cross-check within the back end of the application to ensure the scanned image matches an image within the data set. The Android application uses TensorFlow Lite, whereby a model is trained with a set of images. When the camera detects a recycling symbol it will return a probability of what symbol it could be. Both iOS and Android use the same data set, with both achieving high accuracy.
The barcode scanner recognises the numbers on the item and matches against the products populated in the real-time Firebase database. The Android application uses MLKit to scan and recognise the items, and iOS imports UIKit and AV foundation. If the item does not already exist in the database, the user can add products into the database themselves allowing the database to expand.
The team says: ‘We are incredibly glad to receive this award as a team and would like to thank our supervisor, everyone that voted for us and those that supported us at the poster fair.’
