基于自采集街景与深度学习的慢行环境提质研究OACHSSCD

In the process of urbanization,the contradiction between the sharp surge in motor vehicle ownership and the growing demand for cycling has given rise to intensified conflicts between motorized and non-motorized traffic.Research on slow-mobility systems encompassing cycling and walking has garnered increasing academic attention.As an emerging technical tool,street-view imagery has been extensively applied to evaluations of the built environment,fostering vigorous development in research on the non-motorized travel environment.However,existing studies exhibit inadequate refinement in analytical depth,failure to fully integrate the perspective of non-motorized lane users,and a lack of scientific real-time data support.In this study,the imbalance between objective indicators of the non-motorized travel environment and users'subjective perceptions was investigated through a case study based on major non-motorized lanes in Changsha,aiming to provide a scientific basis and design paradigms for slow-mobility environmental quality improvement.A total of 38643 images were acquired from the non-motorized lane perspective through a self-collected street-view approach with enhanced timeliness and precision.Ten street-view visual features were quantified via deep-learning-based semantic segmentation(DeepLabV3+model).Concurrently,correlations between objective data and subjective perceptions were analyzed by integrating 511 valid online questionnaires and 27 offline interviews.The results reveal that:1)the Green View Index(GVI)and Sky View Factor(SVI)are the primary determinants influencing people's perception of the non-motorized travel environment,2)spatial enclosure and interface continuity serve as the visual underpinnings of non-motorized safety perception,3)there's a severe imbalance between traffic elements and non-motorized vitality,and 4)there's a significant disparity between subjective experiences and objective data in the non-motorized travel environment.Further exploration identifies three types of imbalance phenomena and their fundamental causes:the misalignment between GVI and sensory comfort,the disconnect between facility configuration and usage demands,and the transformation failure from SVI to enclosure and safety perception.The root causes lie in the over-reliance of planning on engineering-oriented quantitative indicators,while overlooking the actual experiences of users.Two core principles,safety priority and quality enhancement,as well as six design models for non-motorized lanes,were proposed,filling the research gap of the evaluation system,which integrates user perspectives and objective quantification of real-time data.It verifies that the combination of self-collected street-view imagery and deep learning is applicable to urban refined governance.The research conclusions provide data support and implementation pathways for the construction of urban non-motorized travel environments.However,this study excludes community branch roads and waterfront slow-mobility roads.The single sample restricts the applicability of the research results.Moreover,insufficiently rigorous questionnaire sampling may induce sample bias,potentially affecting the reliability of the research results.Future research could further leverage high-precision timestamp information,introduce a temporal dimension to analyze the dynamic variation characteristics of street-view visual elements during cycling,and couple cycling trajectories with street-view visual indicators at a finer spatial scale.