Perceived traffic safety influences cycling behavior, yet its value relative to travel time and other route attributes remains unquantified. Building on the stated choice experiment and street-level image dataset of Terra et al. (2025), we extract safety perception scores from cycling-perspective images using a computer vision model and estimate mixed logit models to test whether perceived safety affects route choice after controlling for visual street-level features. Cyclists are willing to accept 64 additional seconds of travel time for a one-unit increase in perceived safety (scale: \(-2\) very unsafe to \(+2\) very safe). Safety preferences vary across demographic groups: older cyclists, recreational cyclists, and those with positive cycling attitudes place more weight on safety, while commuters prioritize speed. These willingness-to-pay estimates enable planners to quantify perceived safety improvements for cost-benefit analyses and to score existing cycling networks for targeted infrastructure upgrades. (Replication code: https://github.com/koito19960406/cycling_safety_perception).
