business strategy, data science

How a Postal Operator Made the most of Computer Vision to Create New Sources of Revenues


When an international and domestic postal operator got in touch with us, they were examining innovative ways to make greater profit out of their “last-mile” delivery, i.e.the final leg of the journey where a product lands in a consumer’s hands.

At that time, they were exploring the idea of adding cameras to their delivery vehicle with the objective to collect road data. What to do with it and how to make it pay was another story. It was right at that time that we entered the game.


To us, seeking alternative sources of revenue streams by using the existing activity made sense in this context. As a matter of fact, when a delivery vehicle is driving, it encounters a variety of data, including road signs. How to use data as a source of value creation applied to road signs? As Flemish municipalities in Belgium are required to maintain their road signs on an annual basis, we felt it was a niche market to address.

So making the most of this data and automate it represented value creation for our client.


Thanks to cameras in delivery vehicles, we were able to capture a large amount of data in Flemish municipalities. We used computer vision methods to detect and classify traffic signs. Backed by annotated images, we could build algorithms that reliably addressed this challenge.

It was an exciting project as we created annotation applications so that our client’s employees could generate an annotated dataset. The objective of it is to give the algorithm samples of the results we expect. In other words, we detect a sign on an image by framing it before classifying it and giving it a code. This method allows our algorithm to automatically recognise a sign.

During our mission, we also went through a data reduction stage, turning videos into images.

Finally, we used a model to recognise people and cars and eventually blur them to keep them anonymously.


We are now developing an automatic system to identify the condition of traffic signs. Basically, if they are not in good condition, our system sends a notification to the traffic sign manager.

We are also extending data collected from the images to road markings.

By and large, there are endless possibilities as we can apply our model to central berms, lampposts or road conditions.