Michael Schmid
Michael Schmid is a Data Scientist at OST’s Interdisciplinary Center for Artificial Intelligence (ICAI), where he works on applied AI, sensor-based monitoring, IoT systems, and data-driven signal processing. Alongside his technical work, he has a strong personal enthusiasm for cycling, which motivates his interest in using technology to better understand cycling conditions and support safer, more attractive infrastructure. He also teaches in engineering education and develops introductory AI courses for kids.
Session
Cities and municipalities in Switzerland, as well as in many European countries, are developing or improving cycling infrastructure. However, much of this infrastructure remains located on roads where cyclists share space with motorised vehicles.
One way to assess the effectiveness of infrastructure improvements is by measuring overtaking distance. In this context, the distance to the curb on the right side of the cycling infrastructure must also be considered, as the lateral position of the cyclist—and thus their distance to the curb—influences the overtaking distance on the left. The distance to the curb is also relevant for cyclists’ perceived sense of safety.
This project introduces Velosense, a measurement approach designed to support targeted test campaigns using instrumented bicycles and to provide direct evidence of how overtaking behaviour changes following infrastructure interventions.
Velosense uses a commercially available radar and video device from Garmin. It combines bicycle-mounted radar, video, GPS data, and AI-based image segmentation to transform raw ride recordings into structured overtaking events that can be mapped, validated, and exported. The key methodological contribution lies in estimating two cyclist-centred spatial metrics that are rarely available in practice: the overtaking distance of motorised vehicles and the cyclist’s distance to the road edge or curb. These estimates are linked with route position and video context, allowing for comparisons of overtaking behaviour during test rides before and after street space or infrastructure redesigns, lane adjustments, or other cycling-related interventions.
The core contribution is methodological: a reproducible workflow that translates everyday cycling trips into event-based evidence on safety distances and overtaking conditions. Rather than relying solely on accident data, aggregated traffic indicators, or infrastructure classifications, this approach measures the interaction itself. Additional event-based indicators include overtaking speed, vehicle deceleration behaviour, overtaking duration, and the spatial frequency of overtaking events. This provides a basis for assessing whether infrastructure changes have a measurable impact on overtaking distance and behaviour.
The presentation outlines the measurement pipeline, the structure of the derived event data, and the value of this approach for cycling research, planning practice, and the evidence-based evaluation of cycling infrastructure.