Whole-Body Magnetic Resonance Imaging Enables Assessing Spatial Accuracy and Precision of Skeletal Joint Locations Inferred from Motion Capture Systems

Abstract

Motion capture systems can be used to infer skeletal joints from three-dimensional surface information for various human poses. However, to-date it remains unclear how well the estimated joint coordinates coincide with the anatomically-correct joint positions. The aim of this study was to determine the localization accuracy and localization precision of inferred joint positions using Magnetic Resonance Imaging (MRI). Color and depth information (RGB-D), and skeletal information of an athlete in static pose standing upright were acquired. A whole-body 3D tomographic scan was also recorded using a 3T MRI scanner. The deviation of the joint location was the largest for the left upper leg (4.1cm±0.2cm) and the smallest for the lower arms (0.2cm±0.01cm). The mean surface point distance averaged 2.2cm±1.3cm (left upper leg), 1.8cm±1.3cm (left lower arm), and 1.5cm±1.0cm (right lower arm). To our knowledge, this is the first attempt to use MRI as a gold standard to validate skeletal joint locations of a motion capture system. MRI provides a suitable means to validate skeletal joint localization for any motion capture system (markerless and marker based). However, advanced software solutions are required to validate and correct Kinect™ skeletal joint localization in the future.