Simulated infant movements: comparing Microsoft Kinect with VICON for 3D Motion Capture.

Research Report
Population: Pediatric

Linda Pax Lowes, PT, PhD, Principal Investigator, Nationwide Children’s Hospital

Lindsay Alfano, PT, DPT, PCS, Research Physical Therapist, Nationwide Children’s Hospital

Mike McNally, MS, CSCS, Lab Manager, The Ohio State University

Rachel Bican, Student PT, Student Physical Therapist, The Ohio State University

Jill Heathcock, MPT, PhD, Director of Infant Lab, The Ohio State University

Keywords: Prediction, Infant Movements, Motion Capture, Kinect

Purpose/Hypothesis: Commercial cameras that measure 3D movements, such as the Microsoft Kinect, have been used with success in measuring arm, leg, and trunk movements of adults with typical and atypical movement patterns from neurological disease. Despite its portability and low-cost, the Kinect has not been used to measure movements in infancy. This is likely because its skeletal tracking mode is only able to track movements for people > 4 years of age. Laboratory based 3D motion capture systems, such as VICON are accurate, reliable, and able to capture differences between typical and atypical spontaneous and purposeful movements in infants. VICON is not portable and the high cost is prohibitive for most clinical application. The purpose of this study was to compare a portable 3D measurement technique using Microsoft Kinect to a gold standard laboratory (VICON) for use with infants.

Subjects: Simulated infant movements were captured

Materials/Methods: A custom designed computer program called ACTIVE mini (Abilities Captured Through Interactive Video Evaluation) uses the Kinect depth and color tracking systems to record 4 distinctly colored wraps on each extremity. Synchronized recordings of the ACTIVE mini and Vicon system of a simulated infant and protractor were captured as the infant and protractor were moved through space in preset directions (i.e. mimicking arm and leg flexion, extension, horizontal abduction, and a random 3D movement when an “infant” is in supine).

Results: Our results revealed good correlation (r = 0.98), moderate agreement (>85% of results within the 95% confidence interval) and small to medium average distance errors (flexion and horizontal adduction: 3 mm, abduction: 18 mm) between peak distances generated by the Kinect and VICON methods. Bland-Altman results showed high inter-tool agreement across all movements. Our findings join others in reporting that Kinect may overestimate movements. Specifically, in this study, Kinect overestimated the maximum excursion in the antigravity direction by 3 – 18 mm. Kinect was more accurate for movements that had greater excursion in the antigravity direction, specifically flexion and horizontal adduction and less accurate in the antigravity direction for movements that had less antigravity excursion such as shoulder abduction.

Conclusions: The Kinect was found to be reasonably effective in measuring simulated infant limb movements, indicating its potential for use in research and clinical settings.

Clinical Relevance: Commercial 3D cameras have the potential to accurately measure 3D limb position (specifically endpoint kinematics) in infants. These preliminary data suggest that modest differences in measurements between Kinect and VICON may be systematic and therefore mathematically correctable.

Lowes, Linda Pax, PT, PhD; Alfano, Lindsay N, PT, DPT, PCS; McNally, Mike , MS, CSCS; Bican, Rachel , Student PT; Heathcock, Jill C, MPT, PhD. Simulated infant movements: comparing Microsoft Kinect with VICON for 3D Motion Capture.. Poster Presentation. IV STEP Conference, American Physical Therapy Association, Columbus, OH, July 17, 2016. Online.