Development of a Semi-Automated Tool Based on Computer Vision Methods for Safety Assessment of Micromobility Users

The operational behavior of micromobility users is a key indicator of the safety performance and design quality of cycling infrastructure; yet, existing video-based methods either require intensive manual processing or locate users coarsely through the centroid of the bounding box. This study presents and validates a semi-automated computer-vision tool that extracts the lateral position and instantaneous speed of micromobility users from bird’s-eye-view video recordings acquired with a single camera. The tool, implemented in Python 3.10.11, integrates bike lane segmentation, background-subtraction-based detection, multi-object tracking, and a heatmap-based contour extraction that places the measurement point at the wheel–pavement contact, providing a physically meaningful reference at predefined control sections. Validation was conducted in controlled tangent and curved sections, against physical distance references and previously verified e-scooter speed readings. In the tangent section, lateral position estimates showed a negligible bias, with a mean error below 1 cm and 99% of observations within ±5.0 cm, while over 70% of speed estimates were within ±2.0 km/h tolerance. In the curved section, the tool slightly underestimated lateral position and overestimated speed, with errors remaining within the practical tolerances. These results support the use of the tool for operational and safety studies of micromobility infrastructure under controlled conditions, reducing processing time without requiring trained detection models.