Measuring Visual Quality of Street Space Based on Deep Learning and Street View Picture : Pilot in The Lilong Area in Shanghai
Qinglai Zhang,,Jiatian Bu,Yifan Yi,,,
DOI: https://doi.org/10.47472/kjck4765
2019-01-01
Abstract:Built environment indicates of street space quality have been carried out in a profound influence on the image of city,human behavior and public health. A street that is considered as a fundamental element in urban studies. Of the 5 elements of the image of a city, i.e. landmarks, paths, nodes, districts and edges, suggested that the paths are the most dominant elements, the research of which would provide a basis for the clustering and organization of the meanings and associations of the other four elements and the city as a whole. Additionally, taking a quantitative measurement of the visual appearance of street space has proven to be challenging because visual information is inherently ambiguous and semantically impoverished. Recent developed image semantic segmentation techniques and Street View Picture dataset make it possible to eliminate the previous restrictions, Furthermore, bringing forward a research paradigm shift. Lilong, which typically represent for historical street space in Shanghai, are selected for empirical study .This paper attempts to measure subjective qualities of the Lilong environment comprehensively and objectively. By employing Street View Picture, pictorial information is the proxy for street physical appearance, which utilizes the image semantic segmentation techniques (The model used in this study is Deep lab V3 which can achieve a 85.7% pixelwise accuracy when classifying 150 categories of objects) to parse an street scene into scene elements, such as buildings, roads, pavement, trees, cars, pedestrians, and bicycles. Then, the elements corresponding to each street point in the two directions of spatial coordinate are summarized and the average value is calculated. The potential factors were calculated based on 3-dimensional composition calculation of greenery, openness, enclosure and motorization may serve as indicators for inferring. The outcomes were used to evaluate the street types, functionality, quality, time, status, and human activities of a street. The result indicates that visual quality of Lilongs are barely satisfactory, while some regeneration projects in the historical protection block is better. A lot of Lilongs are in shortage of visual green, relative more continuous but with low vertical diversity. In the most recent 6 years, less than 2.74 million square meters Lilongs are not regenerated which are mainly slow building renovation. A series of quantitative analyses demonstrates the ability and great potential of auto-calculation method useful for auditing street environments.