Restoration of Campus Pedestrian Space Based on Visual Perception and Machine Learning

The vision-based restoration of pedestrian space on campus is of great significance to the mental health of college students. In order to make up for the lack of objectivity and research depth of the existing research on the restoration of campus space, the outdoor pedestrian space of students' main activities is taken as the research object in this study. An eye tracker was used to explore gaze patterns in the landscape space, and SD questionnaire was used to screen the environmental factors related to the restorative state. The feasibility of the application of virtual reality technology was verified. The experimental results showed that, height different of building (HDB), height of tree (HT), area of elevation in the field of view (AS), hue of ground (HG) and texture of ground (TG) had a strong correlation with restorative state (?S). A virtual pedestrian experiment was conducted by setting up VR scenes with different parameters to reveal the influence mechanism of five environmental factors on ?S. After training and comparing three prediction models for the restorative state, t GA was applied for restorative design optimization. The optimal thresholds were as follows: HDB: 20m~28m, HT: 8.5m~15.5 m, AS: 12%~15%, TG: 0.20m2~0.50 m2, and HG: 140° ~245 °.