Abstract
Competitive swimming is a demanding sport that requires rigorous training to achieve technical perfection. Research on computer simulations of flow have improved our understanding of how thrust and drag can be optimized for better performance. However, for a swimmer translating this information to technical improvement can be difficult. In this paper, we present an analysis and visualization framework for swimming motion that uses virtual reality to display 3-dimensional models of swimmers. The system allows users to digitize their motions from video sequences, create personalized virtual representations by morphing prototypical polygonal models, visualizing motion characteristics and comparing their motions to other competitors stored in a library. The use of virtual reality alleviates many problems associated by the current video-based visualization methods for analyzing swimming motion.
Introduction
In competitive swimming, success is often determined by hundredths of a second, making technical perfection critical. While researchers utilize Computational Fluid Dynamics (CFD) to study water flow, these simulations are often too computationally expensive and specialized for daily coaching use. On the pool deck, coaches typically rely on 2D video analysis. However, 2D video suffers from perspective distortion, loss of depth information, and frequent occlusions caused by bubbles and splashing. Furthermore, traditional motion capture systems fail underwater due to light refraction. This research presents a Virtual Reality (VR) framework that bridges this gap by transforming standard 2D underwater videos into personalized, interactive 3D models.
Methodology
The framework consists of three primary components designed to move from raw video to a clean, interactive 3D environment.
Motion Capture from 2D Video
Since underwater sensors are impractical, the system employs a semi-automated digitization process. Users mark joint locations on synchronized 2D video frames from multiple angles (front, side, and top). To recover the missing 3D depth ($z$-coordinate), the system uses a reference "T-pose" image of the swimmer to establish real limb lengths ($r$). By comparing the projected limb length in the video ($p$) to the real length, the depth is calculated using the Pythagorean identity: $d = \sqrt{r^2 - p^2}$.
Human Body Shape Parameterization
To ensure the visualization is realistic and useful for future CFD analysis, the system morphs a "prototypical" human mesh to match the specific athlete. It utilizes 33 standardized anthropometric measurements. For missing data points, it employs linear regression based on the 1988 Anthropometric Survey of US Army Personnel. The final mesh is deformed using Radial Basis Functions (RBF) with a biharmonic spline to ensure a smooth, personalized representation.
Interactive 3D Visualization
The resulting data is exported to a VR environment where users can rotate, zoom, and toggle between rendering modes (solid, silhouette, or skeletal). The system automatically calculates joint angles, velocities, and accelerations, providing quantitative data that is difficult to extract from raw video.
Results and Discussion
The system successfully transformed 2D breaststroke and butterfly footage into digitized 3D skeletal motions. By moving from video to a 3D mesh, the system effectively "cleans" the data, removing the visual noise of bubbles and splashes. This allows a clear view of the stroke from any angle—including angles that would be impossible to film in a real pool.
A key insight of this work is the power of normalization for comparison. By storing motions in a library, an amateur swimmer's motion can be overlaid directly with an Olympic champion's. This makes subtle technical flaws, such as the amplitude of a leg kick or hip displacement, immediately apparent regardless of the swimmers' actual speeds.
Conclusion
The presented VR framework offers a significant advancement over traditional 2D video analysis by providing individualized, depth-accurate 3D reconstructions. While the current motion acquisition process is labor-intensive due to manual joint marking, it provides a foundation for future automated computer vision techniques. The framework is general enough to be applied to other cyclical sports, such as cycling or baseball pitching, where technical form is paramount.
Summary updated