Speaker
Description
As the most common public space, urban street is not only an important carrier of people's daily activities, but also a direct reflection of the quality of urban environment and social inclusion. In the face of the multiple challenges of global climate change, resource crisis and social inequality, optimising the spatial quality of urban streets has become a key issue to enhance environmental sustainability, social inclusiveness and livability. However, most existing studies focus on the single dimension of objective environmental characteristics or subjective perception, ignoring the systematic differences between the two and their far-reaching impacts on spatial design and planning decisions. How to identify the differences between subjective and objective qualities in a scientific and quantitative way and formulate targeted optimisation strategies in the context of rapid urbanisation is still a key issue to be solved.
In this study, a spatial quality assessment framework based on multi-label classification is innovatively proposed by combining deep learning techniques and semantic analysis methods. Based on 110,988 street view images from the MIT Place Pulse dataset, the Microsoft Trueskill algorithm is used to quantify six subjective perception scores of streets, including safety, comfort, and richness; meanwhile, based on the Cityscapes dataset, the Mask2Former algorithm is used to semantically segment the objective features of the 28,662 streets in the downtown area of Xi'an. streets based on the Cityscapes dataset; meanwhile, the Mask2Former algorithm was used to semantically segment the objective features of 28,662 streets in Xi'an city centre, and 19 landscape element indicators were extracted from the four dimensions of safety, comfort, richness and convenience.
It was found that there are significant differences between subjective and objective street qualities in the urban centre of Xi'an. Green visibility, proportion of walking space and transparency of street interfaces have a significant effect on subjective comfort and richness, while deficiencies in objective indicators such as road surveillance and pedestrian accessibility in some areas lead to lower perceptions of safety and convenience. In addition, tourist attractions and amenity streets had significantly higher subjective ratings than traffic streets, and densely built-up areas had insufficient green visibility and sky visibility limiting comprehensive perceptions despite high accessibility.
Based on this, the study proposes recommendations to optimise the quality of streets, including improving green coverage and street richness, enhancing the design of pedestrian spaces, and improving the transparency and functional diversity of buildings along streets. Through quantitative analysis, the study reveals the key deviations and optimisation paths in the subjective and objective perspectives of street space, providing practical basis and theoretical insights for street planning and design in the context of rapid urbanisation, and helping to build a more inclusive and livable urban public space.
| Keywords | Urban Street Quality, Perception Differences, Deep Learning |
|---|---|
| Best Congress Paper Award | Yes |
