Intelligent Assessment for Visual Quality of Streets: Exploration Based on Machine Learning and Large-Scale Street View Data

Abstract

At present, the collection and analysis of large amounts of key data for the visual quality assessment of streets are performed manually. The assessment efficiency is not high, and the effective information is not fully explored. This study aims to establish an intelligent method for assessing the visual quality of streets. Taking the Hexi District of Tianjin as an example and using street view images as the assessment medium, an assessment model of objective physical indicators is established based on PaddleSeg, an assessment model of subjective perceptual indicators is established based on neural image assessment, and a visual quality assessment model of streets is established based on a random forest. The above models can intelligently evaluate the visual quality of streets and key indicators affecting visual quality. The influence of each key indicator on the visual quality of streets and the relationship between objective physical indicators and subjective perceptual indicators are analyzed. Through a combination of subjective and objective as well as qualitative and quantitative methods, the results show satisfactory assessment accuracy. In short, this study uses machine-learning techniques to improve the scientific rigor and efficiency of visual quality assessment and expand the scale of visual quality assessment data.