Comparative Analysis of Cameras and Software Tools for Skeleton Tracking

Skeleton tracking is a valuable tool for monitoring user performance. In this study, we present a methodology for comparative analysis of cameras and software tools for skeleton tracking, aiming to assess their performance in quantifying 3-D motion accurately. The methodology provides a systematic way to compare tools for skeleton tracking, enabling the prioritization of different features depending on the specific study objectives. It contains three main steps: experimental design, feature extraction, and feature analysis. The methodology is demonstrated for two case studies to analyze upper and lower extremity features using an experimental design comparing the performances of three cameras (RealSense (RS), ZED2mm, and ZED4mm) and three skeleton tracking algorithms (PyZED, Nuitrack, and MediaPipe). An experiment was performed with 16 participants walking a 6-m path while moving their hands horizontally. Ground truth was measured using a 3-D marker-based motion capture system (a Vicon Motion System). Two methods are employed: a comparison of root-mean-squared error (RMSE) values and a statistical test applied to a linear mix model. Results indicate that ZED-2i cameras outperform the RS camera across features, and ZED2mm exhibits superior performance. This study demonstrates that ZED cameras perform better for both upper and lower extremity features. This article contributes a methodology along with valuable insights on the specific tools tested for selecting appropriate hardware and software tools for skeleton tracking applications for features-based assessment.