Evaluation of OpenPose for Quantifying Infant Reaching Motion

Computer Vision Tasks for Ambient Intelligence in Children’s Health

“…The potential ability of computer vision to accurately characterize infant reaching motion is the topic of the paper in [130,134]. Analysing reaching motion (fast movement towards a given target, usually a toy) may contribute to the early diagnosis and assessment of infants at risk for upper extremity motor impairments.…”

“…Analysing reaching motion (fast movement towards a given target, usually a toy) may contribute to the early diagnosis and assessment of infants at risk for upper extremity motor impairments. In [130] the analysed videos obtained were from 12 infants (5 with developmental disorders) of about 12 months of age or less. The total number of reaching actions analysed was 65.…”

Computer-assisted approaches for measuring, segmenting, and analyzing functional upper extremity movement: a narrative review of the current state, limitations, and future directions

“…The output 2D keypoint representation has had considerable research interest recently, which is partially motivated by the ubiquity of RGB cameras ( 30 , 57 ). Although there are many algorithms, one notable 2D human pose estimation algorithm is OpenPose ( 28 ), which has been evaluated for utility in measuring UE kinematics ( 17 , 58 , 59 ), among other approaches. These applications involved evaluating the 2D errors of the pose predictions for reaching movements in infants ( 58 ) or extracting depth values from a red-green-blue-depth (RGB-D) image using the 2D predictions to create 3D landmarks of UE movements ( 17 , 59 ).…”

“…Although there are many algorithms, one notable 2D human pose estimation algorithm is OpenPose ( 28 ), which has been evaluated for utility in measuring UE kinematics ( 17 , 58 , 59 ), among other approaches. These applications involved evaluating the 2D errors of the pose predictions for reaching movements in infants ( 58 ) or extracting depth values from a red-green-blue-depth (RGB-D) image using the 2D predictions to create 3D landmarks of UE movements ( 17 , 59 ). Using 2D keypoint predictions followed by converting to 3D coordinates using depth from an RGB-D camera appears to be the most common use of 2D pose estimation by movement scientists because human functional motion is often tri-planar, except for assessments where uni-planar movement is specifically of interest (e.g., shoulder abduction in frontal plane ( 59 )).…”

“…The best-performing 2D pose estimation algorithms have been demonstrated to be useful for a variety of training and rehabilitation applications ( 17 , 58 , 59 , 79 , 80 ) involving gross movements. However, improvements are still needed to make 2D pose estimation comparable to gold standard motion capture systems, such as incorporating physiological constraints ( 17 , 80 ) and temporal smoothing ( 17 ).…”