Table 1.
Overview on analyzed datasets.
In each dataset, the same subjects underwent painful trials and control trials. For the BioVid dataset, several videos were excluded from analyses, because participants left the camera’s field of view or visual review revealed obvious pose measurement errors. Abbreviations: M = mean, SD = standard deviation.
Fig 1.
Egocentric rotation angles describing orientation of the head in degrees (DEG).
Pitch quantifies down- or upward head orientation, yaw quantifies left or right head turn, and roll quantifies right or left head tilt.
Fig 2.
Head posture time series with corresponding video frames showing reaction to a painful heat stimulus from BioVid dataset (high temperature plateau lasting from second 0 to 3.5).
The subject moves her head downwards and followed by a little upward movement, which is reflected by the increase and following decrease of the pitch angle.
Table 2.
Head posture: Summary statistics of orientation angles (in DEG).
Pitch, yaw, and roll angles of each trial video sequence were summarized by their respective mean and range. For each dataset (BioVid, UNBC, and BP4D) and statistic (columns) we report mean and standard deviation, M (SD), of pain and control trials as well as the p-value of the respective analysis of variance (rows). Significant differences are highlighted in bold.
Table 3.
Head movement: Summary statistics of angular velocities (in DEG/s).
Pitch, yaw, and roll velocity of each trial video sequence were summarized by their respective mean value and mean of magnitude. For each dataset (BioVid, UNBC, and BP4D) and statistic (columns) we report mean and standard deviation, M (SD), of pain and control trials as well as the p-value of the respective analysis of variance (rows). Significant differences are highlighted in bold.
Fig 3.
Specific head posture’s occurrence counts and significance test results.
Each dataset was subdivided in 27 postures. The 8 postures that occurred most frequently among pain trials were considered for comparing the frequency of occurrences in pain and control trials with binomial tests. The figure illustrates the 8 postures per dataset and lists the occurrence frequencies in the trial categories. Significant differences are marked by asterisks.
Fig 4.
Head movement clusters in the BioVid dataset with number of pain and control trials falling into the cluster.
Clusters are illustrated by their mean movement. Significant differences (according to the conduced binomial tests) are marked by asterisks.