Skip to main content
Computer Vision AI Automation Computer Vision AI Automation

Waste detection and river cleanup with a Computer Vision AI system

Dredging company DEME Group partnered with Robovision to develop a computer vision AI-based automated system that detects and identifies waste in a river, after which an autonomous ship collects it. The AI-powered waste detection system gets increasingly better at recognizing waste– a big plus for the environment and a huge efficiency gain for DEME.

Fast
The processing power of the model is so high that the response of the automated ship is almost instantaneous.
Scalable
Can be used in a variety of natural situations like lakes and river.
Accurate
With each object it identifies, the model learns and becomes more accurate.

The challenge: waste detection in oceans and rivers

We don’t have to convince you that polluted oceans and rivers are a problem. For the animals living in them, for the surrounding ecosystems, and basically for the whole planet. Dredging company DEME Group puts its expertise to work in turning the negative impact of polluted waters around. We’ve tried to help them do just that—but faster.

We really enjoyed the collaboration with Robovision, as they proactively came up with solutions, which is for us a key factor in a good partnership.

Magali Bruggeman, Business Development Manager at DEME Group

Cameras attached to a bridge

While developing a solution, we virtually split the river into two parts. In the shallow part, a net ensures passive waste collection. In the deep, where boats and canoes or suppers float by, we work with active waste collection. Two cameras attached to a bridge over the river have this deep part of the river in sight. Thanks to Robovision AI, they can identify whether an object is waste or not. If it is, an autonomous ship collects the object and neatly deposits it in the passive collection net.

High degree of variability

We faced quite some challenges in this case. There was no off-the-shelf algorithm for this specific question but there was the high variability. The weather has a big impact on the surface of the water, as well as on the objects floating in it. A detector existed, obviously, but the dataset was missing. But if we didn’t love a challenge now and then we’d be in the wrong business, pioneering in the application of AI! We started with an unsupervised approach, identifying everything that isn’t waste, and continued to train the model in a supervised manner. Meanwhile, the intelligence on what is and isn’t waste is growing increasingly sturdy.