Urban waste | Since 2024

A cleaning vehicle that inspects its own route

Since 2024 a system on board our client's vehicles locates and classifies, along the usual route, waste next to containers and alerts their platform.

Technologies

Computer vision

Detection and classification

On-board anonymisation

Georeferencing

On-board processing

Client platform

The idea and its impact

Before the commercial launch, a pilot on the route refined the requirements and the sensor mounting. The commercial launch came in the third quarter of 2024.

What our client was looking for

Detect automatically, along the route, the waste left next to containers, with anonymisation done on the vehicle itself.

The fleet today

Today 13 service vehicles run Seevia Waste in Madrid, Valladolid, Zaragoza and L'Hospitalet de Llobregat. The next planned deployments are Bilbao, Vigo, Granada and the Canary Islands.

Illustrative diagram

What Seevia brings

01

On-board processing

The cameras and the computer are mounted on the service vehicle and analyse the images along the route. Only the alert reaches the platform.

02

Alerts with judgement

Each alert arrives with place, time and an encrypted image when there is an incident to deal with, integrated with our client's platforms.

03

Privacy by design

We store no images, blur pedestrians and vehicles and crop the area of interest.

04

From detecting to classifying

Besides detecting the waste, the system tells what type it is, so the right service can be sent.

Contenedores y residuos junto a ellos, detectados por Seevia Waste; los coches aparecen difuminados

Waste next to the containers, detected on board.

Both side cameras on the route: people and plates are already blurred.

How it has evolved

Running around the clock since late 2024. In 2025 we improved precision in the cases that were most often confused. In 2026 detection improves with more data and the system classifies the type of waste.

Measured results

93.51% good positives in good visibility

Measured on morning, afternoon and night sequences. With poor visibility, 88.88%.

Manual review of alerts against the image · October 2024

Night model: 89.33% against 72.32%

It gets more right and detects 2.5 times more incidents over an equivalent period.

Night images reviewed by hand · March 2025

Classification: about 75% precision on the main classes

Measured on the most frequent waste classes.

Our own daytime control set · March 2026

Current detection: average precision of 88.2%

Disproportionate alerts fall to 9.6%, the project's lowest.

Daytime and night images reviewed by the team · April 2026

Milestones

Before the third quarter of 2024

Detection pilot for containers and waste, integrated with our client's platform.

Third quarter of 2024

Commercial launch, with industrialised on-board equipment.

Late 2024 and early 2025

Around-the-clock operation, at night too.

Second and third quarter of 2025

More precise detection and connection to Mawis.

November 2025

Classification of the type of waste begins.

2026

Improved detection with more data and industrial-grade equipment.

Questions

Can it be applied to my city or site?

Yes. Each case is studied and adapted to its environment and the way it works.

How do we start?

With a study of your case and a scoped pilot before extending.

Have a similar case?

Write to us and let's talk.

Next

A major engineering group