Assessing Visitors’ Interaction Using Virtual Reality in Dynamic Daylit Exhibition Spaces

Abstract

In this chapter, finding and predicting the underlying relationship between daylighting compositions and visual interactivity was sought through machine learning techniques. Subjective responses to daylighting in exhibitions virtual environments were extracted based on the luminance maps of HDR images to find how these ratings varied across daylighting compositions. Data were collected from 41 participants to rate a selection of 4 images via semantic differential scale. Data gathered about their impressions towards the space regarding: pleasantness and excitement showed the relationship between increasing spatial contrast with perceptual ratings. Also, their PAI was evaluated showing the effect of contrast on increasing it. Simultaneously, head tracking data were extracted to provide the frequency levels of their visual attention and AoI across the space. Insights about variations in these maps in relation to luminance data were represented. Machine learning techniques were explored to infer a predictive model for interactivity level through finding the underlying relation between visual attention and daylight luminance features. The 6 classifications of salient regions were labelled in the visual attention map to train ML; however, the highest accuracy reached was 59%. To simplify the process, the regions from 2 to 4 which were found to have very small areas were combined with region no 1 to have only two binary classes. This binary classification was inferred to facilitate the learning process on ML for obtaining higher prediction accuracy results. The accuracy has increased to 85% using bagged trees algorithms. Implications were demonstrated to develop a predictive model for expecting the salient areas in daylit exhibitions where visitors can be attracted to.