Video-Based Analysis and Reporting of Riding Behavior in Cyclocross Segments

Area of Science:

• Sports Science
• Computer Vision
• Biomechanics

Background:

• Video-based trajectory analysis is common in sports like soccer and basketball.
• Its application in cycling, particularly cyclocross, is less explored.

Purpose of the Study:

• To present a video processing pipeline for extracting riding lines in cyclocross races.
• To enable detailed analysis of rider behavior and performance metrics.

Main Methods:

• Utilized an Alphapose skeleton detector and a spatiotemporally aware pose tracker to identify and track riders.
• Enriched pose data with meta-information like rider posture (sitting/standing) and team affiliation.
• Developed a post-processor to consolidate information and propose rider lines with associated meta-data.

Main Results:

• Successfully extracted riding lines and rider behavior from video feeds.
• Enabled insights into intra-athlete ride line clustering and anomaly detection.
• Provided detailed breakdowns of riding and running durations within race segments.

Conclusions:

• The developed methodology offers valuable data for performance analysis in cyclocross.
• Insights can enhance sports storytelling and facilitate automatic race summarization.