Urban landscape element identification method and system based on VR deep visual perception

Abstract

The invention relates to an urban landscape element identification method and system based on VR deep visual perception, and the method comprises the steps: building an urban VR scene, enabling a userto experience the urban VR scene through wearing a VR helmet, and evaluating the urban VR scene; obtaining head posture change information when the user experiences the urban VR scene in real time through the VR helmet, and determining the focus of the user in the urban VR scene according to the information; meanwhile, dividing the user groups into high-evaluation groups and low-evaluation groupsaccording to the subjective evaluation data of the users; classifying the attention points according to the visual attention points of the high-evaluation crowd and the low-evaluation crowd, obtaining the thermodynamic values and the thermodynamic grades of the attention points, classifying the attention points belonging to the same style and feature element into one class, obtaining the attention grades, and establishing an analysis table. According to the method, subjective perception and objective deep visual perception of people are combined together to recognize urban landscape factors,and a reliable basis is provided for humanization and quantitative analysis of urban landscapes.

Description

technical field [0001] The present invention relates to a research method of city style, in particular to a method and system for identifying elements of city style based on VR depth visual perception. Background technique [0002] Urban style is a manifestation of urban value, and has always been an important content of architecture and urban planning research. Traditional urban landscape research relies too much on the subjective judgment of researchers, which makes the phenomenon of simplification and homogeneity of urban landscape more obvious. At present, there are four problems in the study of urban landscape, which are as follows: [0003] First, the traditional urban landscape research is mainly qualitative research and experience summarization, lacking quantitative research. Although as early as 1977 Rudolf Arnheim introduced the Gestalt psychology method in the book "Architectural Form Dynamics", at the same time, Yoshinobu Ashihara started from the perspective o...

AI Technical Summary

Problems solved by technology

Although many studies have focused on line of sight and field of view analysis, and introduced new digital technologies, for example, in 2016, Pang Feng’s team at Qingdao University of Technology proposed: a calculation method for setting the three-dimensional scale of visual corridors under urban planning and design (Patent No. : CN201610307603), in 2018, the Wang Jianguo team of Southeast University systematically proposed a new method of urban landscape design based on digital technology. However, it only discusses the problem of whether the corridor of sight is blocked, studies the simple problem of visible and invisible, and lacks the participation of people in the research

There are also some studies that only analyze the outline of the building interface, without in-depth research on the elements of architectural features, which makes these studies lack of connection with urban culture and people's actual feelings

[0005] Third, research costs are high, and there are great difficulties in data collection and analysis

Although artificial intelligence and machine learning can be used to perceive and identify urban landscape elements to a certain extent, for example, the "SegNet" tool developed by researchers at the University of Cambridge in 2015 (Badrinarayanan, Handa et al.2017) can identify There are a total of 12 elements such as sky, sidewalk, driveway, building, and greening. Another example is that in 2018, researchers at Tsinghua University used street view photos to identify and explore the informality of Beijing's built environment (Gan Xinyue, She Tianwei et al. 2018), However, this method requires a large amount of image data as the basis of the recognition model. In theory, more than 10 million labeled sample data are needed to obtain human perception performance, which makes this method still need a lot of room for improvement in terms of technology and algorithms. (He Wanyu, Li Chun et al.2019)

At the same time, although there are also researchers who wear eye trackers in real urban environments to analyze the perception of urban landscapes, for example, in 2019, Sussman used eye trackers to study the Piazza del Campo in Italy and the City Hall Square in Boston, USA. In a comparative evaluation study, using eye movement data to discover the different spatial perception of two squares (Ann Sussman JM W.2019), but using eye tracking to visually analyze a sample, the data collection will take half a day, and the experimental environment is affected by the environment. The human and vehicle activities in the experiment have a great influence, the experimental accuracy is very limited, and there is still a certain distance from the real application.

[0006] Fourth, due to the limitation of time and space, it is difficult to sample experimentally

In terms of space, although some existing studies use simple questionnaires or photos to feedback the perception of urban landscapes, ordinary experiencers cannot experience the real environment, but rely on sporadic memories for perceptual recall, or rely on photos. The limited information to infer from experience, which will cause great errors in the perception feedback of urban landscape

In terms of time, there are also differences in on-site perception at different time periods. If the experiencer is not familiar with the site, or has not been to the experimental site, or the experiencer has a wrong understanding of the content of the on-site perception, the experimental sampling work will be difficult.

Images