River cleaning and VITO Engage citizens in the Waste Watchers project to detect trash using drones. The AI ​​will then identify where the trash is located, making cleanup more efficient.

ITdaily was allowed to take a closer look along the Scheldt in Temse. We met with Thomas de Groote, founder of River Cleanup, and Klaas Pauly, researcher at VITO and project leader of Waste Watchers.

Our day begins in Temse on the towpath along the Scheldt, where we sit in the shade of the famous Boelwerf crane on a sunny day. This Hensen crane is the only one of its kind in Belgium and a tribute to all shipbuilders. The crane rises 48 meters high above the Scheldt, where many cyclists pass by on this beautiful day.

A beautiful place, but that is not the main reason why we agreed here. “Temse is one of our test locations for the drone project,” says Thomas de Groote, founder of River Cleanup. In collaboration with the research institute VITO, the non-profit organization is mapping the garbage along the Scheldt in a very special way. “We fly drones along the banks of the Scheldt to photograph the garbage. Thanks to the combination of technology and artificial intelligence (AI), we can find out where we need to clean up,” says de Groote.

Through the eyes of a drone

We look around for rubbish on the towpath. Here and there you see a can or cigarette butt, but that is only a fraction of all the rubbish along the Scheldt. “There is a lot of rubbish hidden between the tall grass or on the banks of the Scheldt that we cannot possibly see with the naked eye from the towpath,” says de Groote. “Where do you start to clean up?” This requires an extra pair of eyes that sees the surroundings from a different perspective.

With a drone you can visualize the surroundings and the garbage from a completely different perspective.

Klaas Pauly, researcher at VITO and project manager Waste Watchers

Is it a plane, is it a bird? No, it’s a DJI Mini 4 Pro that you see in the sky. Klaas Pauly, researcher at VITO, takes a black bag from his luggage carrier Blue bike and shows us the drone that does all the work. “This DJI drone is very easy to use and does not require a drone license. “During our flight days as part of the Waste Watchers project, no less than two hundred enthusiastic volunteers flew similar drones without any problems or plane crashes,” says Pauly, laughing.

Ready for takeoff

To understand exactly how it works, we are shown a live demonstration. Pauly first draws his blue Waste Observer-Wear a fluorescent vest with the inscription: “I am a Waste Watcher, my drone looks for garbage.” A little later, passers-by immediately notice us. “You are from River Cleanup, what you are doing is fantastic!”, they say. The cardigan immediately proves its worth. With this nice compliment in mind, the parameters of the drone are set correctly and the drone is ready for use lift up.

These parameters were not chosen at random. The drone is sent into the air at a height of fifteen meters. The camera is then pointed vertically downwards and the aircraft flies over the towpath at a constant speed (about 10 km/h). A higher speed could affect the quality of the images. During the flight, the 12 MP camera takes a photo every five seconds so that the images are neatly connected.

“It is important that you can still see the drone with the naked eye in the air,” says Pauly. As soon as our eyes have difficulty locating the drone, Pauly decides to return the device. Photos are also taken on the return flight, this time with part of the Scheldt in the picture. After a flight of several minutes, the drone has collected about 250 meters of photos of the towpath.

Waste Observer

To map a large part of the Scheldt, many pilots, drones and batteries are needed. “On average, the battery of a DJI mini drone lasts about twenty minutes,” says Pauly. Fortunately, Pauly and de Groote are not alone in working at all of the research project’s locations. River Cleanup organized several flight days as part of the Waste Watchers project.

These flight moments were primarily intended to introduce people to drones and the technology behind them. “About 200 volunteers flew a drone for the first time in their lives,” says de Groote. In addition, there were about twenty citizens who contributed to the project with their own drones. “They flew their drones in different locations based on our protocol and sent regular images,” says Pauly.

We collected around 10,000 photos in three months.

Thomas de Groote, founder of River Cleanup

The pilot study of the Waste Watchers project was tested in Antwerp, Hemiksen and Temse in the spring. The flight days were organized in Dendermonde, Wichelen, Ghent, Oudenaarde and Avelgem in the summer. Volunteers flew to many other locations with a drone. Between July and September they collected around 10,000 photos. “This shows the power of the combination of AI and technology. This means you can search a much larger area and organize targeted clean-up operations,” says de Groote.

AI makes work easier

Pauly explains to us in understandable language what the next, more technical step in the process looks like. “Once the images are taken, something has to be done with them,” Pauly begins. “First we wanted to work with a Convolutional Neural Network that can recognize objects and images, such as a cigarette butt. “However, it is almost impossible to do this for every type of waste,” says Pauly. For this reason, the VITO researchers began looking for an alternative.

“We have switched to multimodal models. However, the best known of these models, such as ChatGPT, are not open source, whereas Meta’s Llama model is,” Pauly continues. “External researchers have further developed this Meta model and trained it on the basis of aerial images, so that the recently derived RS-LLaVa model is suitable for detecting objects from the air,” says Pauly.

The AI ​​model recognizes where the garbage is located.

Klaas Pauly, researcher at VITO and project manager Waste Watchers

“We chose this model to analyze our drone images. In the future, we want to refine everything even further,” says Pauly. The model is currently able to indicate whether or not there is waste, but not what type of waste it is. The model has an accuracy of 60 to 80 percent.

Find hotspots

Before the photos are checked by the AI ​​model, they must be located on a map. The images are first stored in software developed by VITO, where the quality of the images is checked. The images are then uploaded to VITO’s MAPEO platform and stored in the cloud. “Using the coordinates, the photos are placed in the right place and then glued together to form a larger whole,” says Pauly.

Once all the images are in the map, it still has to be cut into small pieces. “The AI ​​model cannot process such large files. To do this, we divided the map into pieces of one square meter each,” says Pauly. These individual pieces can then be poured into the model, which in turn recognizes whether there is garbage or not.

In the picture above you can see the result of the drone detection in Temse in one go Heat mapHere, the hotspots with the highest concentration of waste are clearly visible. Even after the clean-up work in this zone in recent months, a significant decrease in waste can be observed (343 positive m² in May 2023 compared to 176 positive m² in August 2024, or more than 50% reduction).

refinement

The project is a world first. “Never before has there been citizen participation on this scale, with citizens coming together to track down garbage with the help of drones and AI. This is how we won the audience and jury awards Flanders Geospatial Awards“, says Pauly proudly. “We want to refine the model even further in the future to get even more accurate results.” De Groote replies: “We are working on a new part of the project that uses smartphones. These photos will help us find out more precisely what type of waste it is. Everyone also has a smartphone in their pocket, so we would get even more data with which we can refine the model.”

VITO’s technical expertise combined with River Cleanup’s army of volunteers creates a great citizen initiative with an important social goal: fighting garbage.