Motion cameras agree on some movements and diverge on others
A peer-reviewed comparison of movement-recording systems finds that agreement depends on what is measured. In 12 people performing Taijiquan movements, pelvis measurements matched relatively well, while trunk orientation and measures derived from movement speed were less reliable across systems. Calibration corrected some stable joint-angle offsets without making every feature interchangeable.
Science··Night
Two systems record the same Taijiquan movements
A single peer-reviewed study compared two ways of recording human movement: Kinect V2 infers body position without attached markers, while Qualisys tracks physical markers. For the reanalysis, 12 participants had been recorded simultaneously by both systems while carrying out 13 standardized gestures from Taijiquan, a Chinese martial art. Rather than assign one accuracy score to a device, the researchers evaluated individual joints and movement features.[1]
Pelvis agreement does not extend to every joint
Pelvis measurements agreed well across systems. Knee and elbow angles showed strong relative consistency but systematic offsets, while hip results changed with the matching of anatomical landmarks. Forward-and-back trunk orientation had negligible cross-system association. Measures of speed and movement smoothness also remained distorted across sensitivity checks. These differences separate tracking a changing pattern from reproducing the same absolute measurement.[1]
Calibration corrects stable offsets selectively
The analysis included 1,815 paired segments and accounted for repeated movements from each participant when estimating uncertainty. Calibration kept participants separate between training and evaluation. It reduced stable knee and elbow bias but could not restore features with weak underlying agreement. The findings apply to these devices, people and gestures, rather than providing a universal rating for markerless motion capture.[1]
