Infant upper body 3D kinematics estimated using a commercial RGB-D sensor and a deep neural network tracking processing tool

Quantitative biomarkers of infant motion may be predictive of the development of movement disorders. This study presents and validates a low cost, markerless motion tracking method for the estimation of upper body kinematics of infants from which proper biomarkers may be extracted. The method requires a single RGB-D sensor, a 2D motion tracking software publicly available and a purposely developed algorithm for the estimation of the 3D coordinates of points tracked from the RGB images. The algorithm deals with various sources of errors in reconstructing the 3D coordinates of the tracked points and allows to estimate kinematic variables to be used to identify potential biomarkers. Both simulated and actual infant's motions were recorded. The infant's motion was recorded at 4, 5 and 6 months of age. Anthropometric measures are also estimated to validate the method on both simulated and actual infant's motion. Known point kinematics were obtained from a doll, with size and shape of an infant, lying on a turntable rotating at $33^{1}/3$ rpm. The doll's motion was recorded from two angles: parallel to the turntable rotation plane and angled at $\boldsymbol{45^{\circ}}$ with respect to it. The latter presents occlusions of tracked points similar to those expected during the recording of an infant's motion. The errors in estimating the selected anthropometric measurements during the infant's motion resulted to be similar to those obtained during the simulated infant's motion. The range of the elbow and shoulder angles estimated during the infant's motion resulted to be well above the error found during the turntable recordings. Similarly, the length of the hand path and mean velocity recorded during the infant's motion resulted to be much greater than the error found in the simulation. Moreover, changes over time of both anthropometric and kinematic variables may be appreciated. Therefore, the proposed method may be effectively used to explore biomarkers of early development of movement disorders. More accurate estimates may be expected if more performing hardware and tracking software are available.