Development of a Ghost-Mode Enhanced Video-Based Biomechanical Screening Tool for Sprint Performance: A Kinovea Bland-Altman Validated Framework
Description
This dataset and computational framework provide a high-fidelity, non-invasive video-based biomechanical analysis system specifically engineered for evaluating sprint kinematics (with focus on the 100m sprint start and acceleration phases). By integrating markerless computer vision pipelines (MediaPipe Pose) with anthropometric constraints, mathematical filtering, and interactive visualization layers (including a ghost-mode skeleton overlay), the system extracts precise performance indicators and injury risk profiles directly from monocular video footage. The architecture is statistically validated against industry-standard 2D software (Kinovea) using Bland-Altman agreement metrics. Methodological Architecture & Key Computational Engines: Multi-Point Scaling & Calibration: Employs an advanced multi-point anthropometric calibration model utilizing standardized body segment proportions (BODY_PROPORTIONS) combined with median-pooled scale factors to ensure precise pixel-to-meter coordinate transformations. 3D-Aware Spatial Stride Tracking: Computes spatial stride lengths using a 3D-aware ankle-to-ankle Euclidean vector matrix, projecting foot placements directly onto a calculated horizontal ground plane to correct for perspective distortions. Temporal Phase & Ground Contact Time (GCT) Engine: Utilizes a Savitzky-Golay filtering pipeline (savgol_filter) for numeric smoothing of velocity profiles, running a specialized peak-detection algorithm to determine discrete touchdown, takeoff moments, and localized GCT events. Automated Interventions & Benchmark Engine: Features a diagnostic comparison engine calibrated against normative elite baselines (e.g., Usain Bolt performance data) to provide rule-based athletic recommendations. Pathology & Injury Risk Matrix: Quantifies real-time bilateral asymmetries and dynamic knee valgus angles (frontal plane deviations projected from sagittal data) to flag prospective musculoskeletal injury risks. Potential Reuse and Scientific Applications: This research repository serves sports scientists, computational biomechanists, track and field coaches, and rehabilitation experts. The code assets and structured tracking datasets can be leveraged to validate markerless tracking software in outdoor field settings, build localized benchmarks for athletic acceleration, or refine automated classification models for human running gaits. Core Academic Foundations Reflected: Nakano, N. et al. (2020). Evaluation of 3D markerless motion capture accuracy using OpenPose with multiple video cameras. Savitzky, A., & Golay, M. J. E. (1964). Smoothing and differentiation of data by simplified least squares procedures. Winter, D. A. (2009). Biomechanics and Motor Control of Human Movement. Stenum, J., Rossi, C., & Kline, T. B. (2021). Two-dimensional video-based analysis of human gait using pose estimation.
Files
Institutions
- Sabaragamuwa University of Sri LankaSabaragamuwa Province, Ratnapura