Open data for bicycle classification for better planning and simulation, and for better detection from motorized vehicles (OpenBike-CD)

We are witnessing rapid changes in the transport landscape that have resulted in a gap between the quality of data available and what is needed from stakeholders. Therefore, this pre-study aims to increase the knowledge about bicycle flow by, together with stakeholders and the industry, identify and define: (i) data needs for better planning; (ii) data collection activities; and (iii) research and cooperation possibilities. To answer these questions a series of digital interviews and e-mail questionnaires were conducted with relevant stakeholders and industry actors. The results show that to better plan for bicycling, stakeholders need more detailed bicycling data, meaning that they need to know the type of bicycle, bicyclists’ socio-demographics, and route choice. Ideally, stakeholders need to know how the bicycling flow/exposure is changing over time, which means they need to follow the percentage share of each type of bicycle (e.g., e-scooters, e-bikes, cargo-bikes), the percentage share of different bicyclists’ groups (e.g., age, gender), and which links/streets are used the most. The bicycle flow-counting industry is trying to cope with and adapt to the rapid changes in the transport landscape and can today mainly classify e-scooters with 3D cameras, with a good degree of accuracy. Other types of bicycles are more challenging to classify, as many different branches exists, and new models are being introduced almost every year. Bicyclists’ socio-demographics variables are disregarded, with GDPR as a great concern or barrier to deal with. The car industry, however, focuses on detecting bicyclists in time at crosswalks/crossings and on right-turn maneuvers, with challenges in detecting bicyclists at adverse weather conditions, e.g., darkness, twilight. Furthermore, there is also a need to vary approaching angles and detection distances, when bicyclists interact with vehicles, so the car industry can train its detection algorithms in a range of real-life scenarios.