Visual Perception Optimization of Residential Landscape Spaces in Cold Regions Using Virtual Reality and Machine Learning

Abstract

The visual perception of landscape spaces between residences in cold regions is important for public health. To compensate for the existing research ignoring the cold snow season’s influence, this study selected two types of outdoor landscape space environments in non-snow and snow seasons as research objects. An eye tracker combined with a semantic differential (SD) questionnaire was used to verify the feasibility of the application of virtual reality technology, screen out the gaze characteristics in the landscape space, and reveal the design factors related to landscape visual perception. In the snow season, the spatial aspect ratio (SAR), building elevation saturation (BS), and grass proportion in the field of view (GP) showed strong correlations with the landscape visual perception scores (W). In the non-snow season, in addition to the above three factors, the roof height difference (RHD), tall-tree height (TTH), and hue contrast (HC) also markedly influenced W. The effects of factors on W were revealed in immersive virtual environment (IVE) orthogonal experiments, and the genetic algorithm (GA) and k-nearest neighbor algorithm (KNN) were combined to optimize the environmental factors. The optimized threshold ranges in the non-snow season environment were SAR: 1.82–2.15, RHD: 10.81–20.09 m, BS: 48.53–61.01, TTH: 14.18–18.29 m, GP: 0.12–0.15, and HC: 18.64–26.83. In the snow season environment, the optimized threshold ranges were SAR: 2.22–2.54, BS: 68.47–82.34, and GP: 0.1–0.14.