Navigation

Search This Site

23. Building the Internet of Wasted Things

Presenters Name: 
Sonali Luthar
Co Presenters Name: 
Primary Research Mentor: 
Madhur Behl
Session: 
1
Grant Program Recipient: 
Not a Recipient
Abstract: 

The U.S. faces a crisis that is burying cities in tons of landfill waste each day. Most waste management occurs downstream after commingled material has been collected from buildings. Sorting waste at that stage is expensive, slow, and manual. Due to the increased proliferation of Internet of Things (IoT) devices and their associated data, buildings and organizations are increasingly aware of energy and water usage. However, we lack the capability to track waste with the same ease and fidelity. Tracking waste and recycling provides the grounds to reduce waste within an organization. 

Motivated by real-world challenges to reduce recycling contamination, we propose building the Internet of Wasted Things‚Äîa closed-loop automated system with cameras detecting items to be disposed of and guiding the user towards the correct bin. We have built a waste object detector deep network that can classify commonly disposed items. When approaching a set of trash bins, the system detects the object and provides visual cues in real-time to help a person correctly dispose the item. The task of detecting contamination (e.g. food stains, presence of liquids, etc.) using vision is extremely challenging. 

We will describe starting from scratch by collecting and annotating large-scale video data at the UVA Link-Lab, to challenges with real-time object detection, to deployment on embedded hardware, and incorporating federated learning across multiple bin locations. Using AI-powered waste management, organizations can streamline the process from collection to disposal. This increases sustainability by reducing the volume of incorrectly disposed objects and minimizing other contamination.