TL;DR
- D4T analyses the entire lift at 40 images per second. For a lift lasting three seconds, that is 120 images.
- Many systems take a few photos per lift. That way you miss what lies under the top layer and a measurement becomes less reliable.
- More images means more detections, more accurate percentages and image evidence for every alert.
How many images an AI system analyses per lift determines how much of the waste it sees. D4T from Rematics analyses the entire waste dump at 40 images per second. That is a deliberate choice, and it has consequences for everything that happens after the lift.
A few photos or a full analysis
There are roughly two ways to look at waste during collection. You can take a few photos at fixed moments. Or you can analyse the whole lift continuously.
"Many systems take a few images during a lift. We analyse the entire waste dump," says Bram Peetermans, CEO of Rematics. "The more images we analyse, the more relevant the insights we deliver."
Why one photo is not enough
Waste falls in layers. Anyone who takes a photo at one moment mainly sees the top. A gas cylinder or a pile of cardboard halfway down the container then remains invisible.
During a lift, the contents shift constantly. What was hidden at first comes to the top a moment later. At 40 images per second, the system captures those moments. If a lift takes three seconds, D4T analyses 120 images of the same container.
What more images deliver
More detections. An item that is half hidden in one image is often visible in the next. This matters especially for small, hazardous items such as batteries.
More reliable percentages. To measure how much non-conforming material is in a stream, such as film in cardboard, you need a view of the whole stream. A few photos give a sample. A full analysis gives a measurement.
Image evidence. Every detection is stored with the image in which it occurred. A customer can see for himself what was in his container. A mistake by the system stays visible and can be corrected.
What it asks of the technology
Analysing 40 images per second with several AI models requires a lot of computing power. That used to be possible only in a data centre. D4T does it on a rugged AI computer in the truck itself. As a result, no video needs to be sent and the analysis keeps running, even on a weak connection. On the product page you can read which cameras and hardware D4T uses.
From images to figures at scale
Per container, it is a matter of dozens to hundreds of images. Across a whole fleet, that soon adds up. The more than 70 refuse trucks with D4T together analyse around 1.75 million lifts per year, good for around 157,500 tonnes of waste.
Because every lift is analysed in full, the figures are not an extrapolation from a sample. In commercial waste, D4T measures on average 40 to 50 per cent recyclable material. Taking the lower limit, that comes to around 63,000 tonnes of recyclable material per year that ends up in the incinerator today. How that measurement translates into CO2 can be read in 5,900 tonnes less CO2 per year and on the impact page.
The question to ask every supplier
Comparing systems for waste recognition? Then ask how many images they analyse per lift, and whether they follow the whole lift or a few moments. The answer says a lot about how reliable the data is.
Want to see what 40 images per second can deliver for your waste streams? Request a demo.