Litter is trash that’s not where it belongs — in a landfill or trash bin. It creates significant problems for the environment, for animals, for humans, and the entire world. It’s dangerous, costly, and unsightly.
We propose a litter eradication effort in three phases: - First, to use crowdsourcing citizen-science to build an open database of litter images. - Second, to use image recognition technologies to develop computer vision models to identify litter in context, and to incentivize users to clean detected litter. - And third, to use the highly-accurate and tested models with autonomous drones and camera technologies to automate litter extraction.
Litter can be identified with pattern-based algorithms according to subtle differences in surrounding environment according to color, size, evidence of text, reflection, etc.
An app or web service can be developed that brings in enough interested users to serve as citizen scientists.
The costs of hosting, servers, database maintenance, and model development computing requirements are not beyond the scope of the projects resources.
Find litter examples, photograph them, upload them to the service, and build a large enough database from which to build machine learning models.
Phase 1 constraints consist of the ability to develop a working image data collection app, and an engaging user account platform to incentivize the photographing process in the first place. Phase 2 constraints have to do with the development of the litter detection model itself, the data science process in not only modeling the computer vision components, but in developing a production algorithm that can use the model in real-time. Phase 3 constraints are significant and revolve around the ability to incorporate the advancements from Phase 2 in a meaningful way to autonomous drone technologies that would include camera sensors, autonomous flight control, and even (ideally) an ability to collect detected litter for drop-off to an appropriate waste or recycling facility.