Exploring Variability in Human-Machine Interaction through Gesture Recognition based on Hand-Tracking within XR

This paper introduces an experimental procedure to estimate the performance variability of hand gesture recognition in enabling content selection in extended reality (XR) environments. Two metrics of interest were considered: i) the Euclidean Distance between the detected point of selection and reference points, also named Distance Error, and ii) the time needed by users to select the desired icons, also named Selection Time. Without loss of generality, the XR device Microsoft HoloLens 2 was used as a case study to carry out an experimental campaign involving four healthy subjects. The metrics of interest are evaluated in terms of both intra- and inter-individual variability. The obtained experimental results showed robust performance in terms of content selection, although both the intra- and inter-individual variability need further exploration to fully understand the true potential of this technology. The results obtained on four participants revealed a measured Distance Error in the interval [6.2 -;-7.4] mm and a measured Selection Time in the interval [2.2-;-2.8] s, indicating the robustness of performance in content selection.