Video analysis in swimming is valuable but time-consuming. Marking joint points, defining phases and measuring angles are still often done manually or with specialised systems.
A study published in August 2026 examines how Human Pose Estimation can automate this work.
Computer vision becomes difficult underwater
[bild id="XX"]
On land, modern systems can now identify human body landmarks remarkably well. In the water, refraction, bubbles, reflections, changing light and partly obscured body segments are added to the problem.
These conditions make swimming a particularly demanding test environment for automated movement analysis.
From video to metric
The long-term goal is attractive: a normal or standardised video recording could automatically identify joint positions and derive technical information from them.
Coaches would spend less time manually placing points and could compare multiple athletes or several training dates more quickly.
Automation does not replace interpretation
An algorithm can measure an angle. Whether that angle is problematic for a particular athlete and race distance is a different question.
Automatic models also need to work reliably across different body types, strokes, camera positions and water conditions.
The research is therefore moving in an interesting direction: it is not the camera that is being reinvented, but the analysis. If these systems become robust enough, biomechanical analysis could move from the specialist laboratory closer to the everyday pool deck.
Data quality determines usefulness
Automated analysis can only be as good as the source video. Camera angle, water clarity and obscured joints remain practical challenges. Before a system is used routinely, it must therefore be clear under which conditions its measurements are stable enough.
SCIENTIFIC SOURCE
https://pubmed.ncbi.nlm.nih.gov/42610410/