Camera-based race bicycle dimensions verification
We investigate how we can reduce cost, deployment effort, human intervention, as well as time in measuring certain bike dimensions before a race, for compliance checks with the Union Cycliste Internationale (UCI) regulations. In this direction, we developed a Computer Vision-based concept framework that receives as input a still image from an off-the-shelf ‘action’ camera, uses Convolutional Neural Networks and a Conditional Random Field formulation for bike parts and keypoints detection with spatial topology constraints (context as strong inductive bias), and outputs the desired measurements. The challenge is achieving keypoint localization within 1 mm error/tolerance range, per UCI regulations. This implementation is part of a cyber-physical system, where bike and camera are set on a custom-built platform for reducing ambiguity, enabling easy ‘extrinsic’ calibration, and avoiding the bottleneck of depth estimation. In our experiments we show that vision-based metrology is a promising alternative to existing, relatively cumbersome procedures, that can maintain measurement errors within the acceptable range. We accommodate varying settings (external conditions), application constraints, and user needs. For the latter, we take into account simplicity and ease of use, also incorporating interaction with the displayed results for reporting and tackling potential disputes during pre-race checks.
Authors
- Gavriil Tsechpenakis (ORCID: https://orcid.org/0000-0002-8864-3435)
- Jan-Anders E. Mansson (ORCID: https://orcid.org/0000-0002-1972-8857)
- Andrei Krasilnikau
- Emmanuel Brunet
- Gurvan Le Boeuf
- Evan David Stack
- Austin Han-Ju Lee
- Adeline Nora Ripberger
- Bryan Zhang
- Gerard Thibault
- Colin Andrew Scifert
Institutions
- University of Indianapolis (US)
- Purdue University West Lafayette (US)
- Purdue University in Indianapolis (US)
- Indiana University – Purdue University Indianapolis (US)
Publication Details
- Journal
- Sports Engineering
- Published
- 2026-11-18
- DOI
- https://doi.org/10.1007/s12283-026-00581-4
- Primary Topic
- Advanced Neural Network Applications
- Type
- article
- Field-Weighted Citation Impact
- 0.00