Project description

If people are asked, why they are scared of cycling in the city, one of the most common answers is, that they are afraid of cars passing them by too closely. Even though there are legal regulations on how much distance cars need to keep when overtaking a cyclist, there is no law enforcement and no data on that matter in Germany. Our goal was to change this and add some hard facts to this often very emotional debate.
In an interdisciplinary team of journalists, physicists and machine learning experts we developed an app and a sensor that allow to measure these distances. We built 100 of these sensors ourselves to distribute them among 100 volunteering readers. The publication concept of the project comprised several serially published chapters in an interactive \”Swipe story\”, a format especially developed for this investigation. The first chapters included a call for participants and were published on August 20th 2018. The story went viral and 2500 readers volunteered. The selection was made using an algorithm in order to select test riders as diverse as possible from different postcode areas, cyclist types and age groups.
The results were published in December after two months of data collection and three month of data analysis and visualizations. On over 13.000 km of recorded distance we measured 16.700 overtakes, 56 percent of which were illegally close. We investigated various correlations in the data like wearing helmets, gender or infrastructure. All in all, Radmesser consisted of eight chapters on the website as well as several additional articles in the newspaper. The project was funded by the “Medieninnovationszentrum Babelsberg”, which supports new and innovative ideas in journalism.

What makes this project innovative?

Project Radmesser combined classical reporting with data journalism, citizen science, polling and innovative technologies such as machine learning and the hardware development of own sensors. The whole project is unique. So far, there had not been any reliable data on the safety distances to cyclists in Berlin nor anywhere else in Germany. Now, thanks to the team's invention, this data was collected. And thanks to 17,000 measured overtaking processes, a previously subjective feeling of fear can be proven: Tight overtaking is a problem on Berlin's streets. The results were narrated in an innovative interactive swipe story, using a playful web design that focuses on facts and statistics: Less text, more images, but especially infographics were designed as core elements of the storytelling. Additionally, the project used many publication channels such as podcasts, printed articles, video, social media stories and newsletters to get the story out there. Finally, Radmesser is a part of what we believe will become a new field in journalism: “The Journalism of Things”. With having sensor technology being able to collect data, journalists can use that. Radmesser proves that sensors and networked things could change the way journalists deal with technology – from reporting about it to reporting with it.

What was the impact of your project? How did you measure it?

The response to the first publications of Radmesser was overwhelming: 2,500 volunteers registered for the 100 places as test drivers. A total of 5,000 people filled out the corresponding survey about biking in Berlin. Local as well as national radio and television stations reported on the project, national print media took up the topic as well. The Hashtag #radmesser was launched overnight by readers. On Facebook, every chapter could be shared individually, whereof alone Chapter 7 of the swipe story got almost 2000 likes and shares. After the publication of the dramatic results, local politicians demanded changes in infrastructure and law. Members of parliament demanded answers to the problem from the federal government.

But it was not only national and international media which showed interest in the project. From everywhere came requests to repeat the project in other cities. The team got very concrete inquiries for collaborations in cities all over Europe. Although we never talked about renting out or selling the equipment, more than 30 associations, media companies, municipalities, political parties and individuals contacted us and asked for cooperation.

Source and methodology

We developed a sensor-kit consisting of a smartphone app and a self-built sensor module with which we can measure the overtaking distances of cars relative to bicycles. The technology did not exist before. The project team built over 100 of these sensors and distributed them among 100 volunteering cyclists in Berlin. They recorded data during their daily rides through the city for about two months. In this way we recorded 13,500 km of tracks and around 17,000 overtaking maneuvers. The data recorded by these volunteers was the basis for the analyses. For a Berlin specific geo analyses we further used open data provided by the Berlin administration on roads and bicycle infrastructure ( The resulting analysis was published as a new Open Data Set, that has been used by various third parties since. The working principle of the sensor-kit is twofold: Two ultrasonic distance sensors measure the distances to the left of the cyclist. Depending on which sensor records a close object first or last, an algorithm can distinguish who was passing by whom. We did not want to record cars waiting in front of a red traffic light while the cyclist passes them on the right-hand side. If the sensor records an object overtaking on the left-hand side, the app triggers a photo on the smartphone. This image is important to verify that the object was indeed a car and not for instance a dove, a running person or the like. Using image recognition libraries, it was then possible to identify the type of object that was passing by. For this setup to work the sensor needed to be attached to the bicycle and the smartphone to be mounted on the handlebar using a standard smartphone mount. The sensor further measures the distance to the right in order to analyze how much distance cyclists keep to parking cars.

Technologies Used

The app was developed using the React-Native framework for cross-platform apps. It uses a weather API to correct the ultrasonic sensors, the smartphone camera and GPS tracking. The sensor module was built using Arduino microcontrollers, ultrasonic-distance sensors, Bluetooth low energy modules and various electronics parts. For the assembly we used e.g. soldering irons, drills and CNC mills. The server backend for storing the data was written with the Django python framework. The data analysis consists of two major parts: detecting overtaking events and the analysis of geo information. The detection stage consists of an image detection step and a step that decides which measurement is a real overtaking event. The image detection is done using the convolutional neuronal network approach of „YOLO V3“ on a NVIDIA GPU using "CUDA" for speedup. The decision step relies on the image information form the detection and on various python technologies like "jupyter lab", "pandas", "scikit-learn", "xgboost", "seaborn", "folium" and others. The actual decision is done using the machine learning technique "gradient boosted decision trees" of the "xgboost package". Apart from a training set of 3000 manually labeled overtaking maneuvers, we checked 7000 critical events manually to make sure, the facts are straight. The geo analysis stage used technologies like "shapely python package", the "HERE GmbH map matching API" and "EPSG coordinate systems". It mainly focused on cleaning, aggregating and anonymizing the data. For this we also used data sets on streets and cycle infrastructure provided by the Berlin City administration. For the visualizations we relied on the javascript frameworks "d3" and "leaflet" for animated infographics and interactive maps.

Project members

Andreas Baum, Martin Baaske, Michael Gegg, Hans Hack, Fabian Altenried, Jakob Kluge, Hendrik Lehmann, David Meidinger, Ekkehard Petzold, Hannes Soltau, Helena Wittlich


Additional links


Click Follow to keep up with the evolution of this project:
you will receive a notification anytime the project leader updates the project page.