To calculate the areas-of-interest we used an OpenPose base head detection based on a previously published study (Callemein et al., 2018; Cao, Hidalgo, Simon, Wei, & Sheikh, 2018). The OpenPose framework detects 18 anatomical key-points in the images that together represent the full human pose skeleton. In the present study, only the five key-points located in the head region, comprising the location of the nose and both ears and eyes, were used (see Fig. 3 for an illustration). Whenever two or more of these key-points are visible the head region is identified, when only one key-point is visible, the head remains undetected. Defining a bounding box around these points creates a rectangular area-of-interest. This bounding box enables a dynamic and autonomous definition of the head area-of-interest around each person’s face visible in the image. In most cases, a simple bounding box around these points would not suffice, since the full head region is not covered by these five points. This issue is solved using the relative distance and orientation of these points to first determine the face orientation (Callemein et al., 2018). We defined a frontal face when all five face points were available, or a profile face when fewer points were visible due to for example turning of the face. Using the largest distance between the available points as the area-of-interest width in pixels, we calculated the area-of-interest height by multiplying the width with the aspect ratio parameter. To ensure the complete coverage of the face, we also used an additional scale (margin) parameter. Different parameters are needed depending on whether the image shows a frontal or a profile face and this accounts for variation in centre location for the area-of-interest. The areas-of-interest are thus calculated based on these parameters.