Telraam: counting traffic with AI on affordable Edge devices

Published on: May 25, 2021

Rat-run traffic, air pollution and traffic safety give city planners huge headaches these days. Systematically counting traffic and collecting traffic data can provide more insight into a possible solution for these problems. Together with traffic sensor specialist Telraam, Kapernikov is now helping cities to count traffic more efficiently with the help of their citizens.

When cities install new roads or change traffic plans, they don’t do this randomly. Decisions like this are usually based on citizen feedback and on hard data. Before they can decide anything, planners want to know what the traffic situation is in the targeted area. 

Citizen science

So, how do you organize those traffic counts efficiently? Enter Telraam, a Belgian traffic measurement specialist and provider of high-tech and reliable traffic measuring equipment.  Telraam has developed a traffic sensor specifically for use in citizen participation projects.

The company has made it easy for city residents to set up their own traffic counters. Participants can mount units on their own street-facing windows and as such become part of a citizen science traffic counting project. All collected data then goes to policymaking and research, but also to residents and interested parties.

Low threshold

Usually, cities and communities don’t have endless money to spend, so they need to make wise choices on how to allocate tax payer’s money. Measurement technology therefore needs to be accessible, and – since it’s used in citizen participation projects – setting up the device needs to be easy as ABC. 

Traffic measurement technology for citizen science projects
needs to be accessible and easy to set up

The challenge for Telraam is therefore not to seek the highest level of performance and technical sophistication, but instead to be creative in making use of technology that is available within a limited budget. Again, that’s exactly what Telraam is doing. According to the Belgian measurement technology specialist, cities should have access to low-threshold technology and be able to benefit from the power of numbers. The more measurement sensors can be deployed by participating citizens, the completer the traffic picture will be. 

Is it a bird? Is it a plane?

However, developing a sensor does not come without challenges. To improve the usability and performance even more, Telraam recently teamed up with Kapernikov and IoT specialist Bagaar, a project which was approved for Flemish funding (VLAIO). Could the company’s current visual sensor, based on a Raspberry Pi system, be improved within budget and performance restraints? Well, we couldn’t wait to get our hands dirty on that challenge. 

Today, the sensor counts traffic with an accuracy rate of 85% to 90%. Not bad at all, if you compare this to industry standards. The sensor now distinguishes between cars, trucks, bikes and pedestrians (not birds or planes, sorry). 

But the sensor still has a few detection issues. For example, groups of cyclists are often detected as cars, and special cases, such as a car carrying a bicycle, often lead to false or missed detections. And we are ambitious: what if we want to classify and distinguish between more types of traffic participants, such as bicycles, cargo bikes and kick scooters (instead of just bicycles), or between vans, buses and large trucks (instead of just large motorized vehicles)?

Today’s state-of-the-art deep learning techniques are quite successful in handling this. But this usually takes a lot of processing power. The question is: can this be done with limited processing power and within budget limits? To achieve this, Kapernikov is currently overhauling the detection algorithm and uses AI on the edge – with the image processing directly on the device – so there’s no need to transfer all sensor images to the database, only the detection information. 

The hardware selection is still ongoing, but we’re confident that we’ll be able to classify traffic in more traffic types with higher detection accuracy. It should be possible to distinguish a group of cyclists from a car, although we probably won’t be able to count the exact number of cyclists in a dense group using this kind of low-power hardware. 

Installation? Easy as 1, 2, 3.

Another thing the team wanted to tackle was the sensor installation. The Telraam sensor needed to be connected to the user’s home Wi-Fi network. For many users, this turned out to be a real struggle, which resulted in too many support calls. 

Connecting to the user’s home Wi-Fi network turned out to be
a real struggle. The Telraam team wants to drastically
simplify the entire process.

The team is now looking into drastically simplifying the entire process. The goal is to bypass the Wi-Fi network altogether and instead let the communications run over Thingstream’s MQTT Anywhere, supporting 2G, 3G, LTE, and LTE-M across more than 600 cellular network carriers in 190 countries around the world without the need for cellular data. This will make the installation much more straightforward. 

We also want to register the sensor’s position and viewing direction by means of an integrated GPS receiver and electronic compass. Again, this does not involve any interaction from the user. The sensor only needs to be attached to the window, and it’s ready to detect.

A different ball game 

Constraints often make people more creative. That’s certainly the case for this project. As much as we love to work with the latest and greatest technology, it’s just as challenging to look for a creative solution within performance and budget limits. Nevertheless, the selected low-power hardware still offers state-of-the-art detection capabilities. 

Telraam wants to keep its traffic sensor for citizen science projects at a low cost and suitable for roll-out in large numbers, and we’re excited to help them achieve this. In fact, it’s a challenge many manufacturing and Industry 4.0 companies are faced with as well. If this project has proven anything, it’s that recent hardware developments have made it possible to integrate advanced AI on low-power or battery-operated devices.   

This project proves that it’s possible to integrate
advanced AI on low-power or battery-operated devices.

Want to see how this story ends? Then keep an eye on this blog for more news on our collaboration with Telraam. 

Author

Maarten De Munck

It has been a long time since Maarten was playing with his MSX computer and Fischertechnik construction toys as a young boy. Back then, it was all about making cool constructi ...