Do we have a problem?
In Germany, only 32 percent of what ends up in the residual waste actually belongs in the residual waste bin (Umweltbundesamt, 2020). Also in Karlsruhe, local waste management institutions face poorly separated trash bins, such as recycable material (Wertstoffe) which can contain up to 50% of residual waste (Restmüll) according to Amt für Abfallwirtschaft Karlsruhe. Recyclable materials are recycled and reused, while residual waste is usually incinerated, exported to developing countries or otherwise disposed of, which leads to environmental pollution. Current endeavours to improve the quality of trash separation, like information sheets of public authorities, technical articles lack the desired effect possibly because they appear not to be focused on the citizens enough.
What can we do about it?
We propose a novel computer-vision-based approach that automatically detects waste objects and determines the correct trash bin. Our solution consists of the one-stage object detector model YOLOv5 as a backend and an interactive user interface as a frontend (see figure 1) connected via a powerful API. WasteDetective is supposed to help the residents of Karlsruhe collectively improve trash separation quality. It can be incorporated in various end-user focused applications for example:
A vision-based smartphone application that tells you what trash bin a trash object belongs to.
A Smart Trash Bin which automatically detects trash objects before separating it into the right bin.
Throughout the project, we have created a working prototype running on a NVIDIA Jetson Nano and thought of strategies of how to introduce our service to the market.
What have we learned?
Throughout the course of the project, the project team has learned some key lessons about building an AI prototype from scratch. For example:
Data plays a crucial role: Quality, labelling conventions and class balance were some of the key aspects in our project
Don’t underestimate the effort needed to deploy an AI model once it is developed
Organize your teamwork and use smart tools that make your life easier like e.g. Roboflow for data labelling and management.
Business aspects like customer base and the exact use case determine the direction of the AI-project development.