南方建筑Issue(1)：61-73,13.DOI:10.3969/j.issn.1000-0232.2026.01.007
眼动追踪视角下城市街道建筑外立面视觉注意力影响机制
Influencing Mechanisms of Visual Attention to Urban Street Building Facades from an Eye-Tracking Perspective
摘要
Abstract
In the context of China's transition from incremental urban construction to stock-oriented renewal,this study systematically reveals the influence of the morphological and colour characteristics of urban street building facades on the time-series evolution of public visual attention.Results can provide quantitative evidence for decisions about facade micro-renewal and optimisation. This paper carried out a case study based on six central districts of Tianjin.Sampling points were generated randomly at an interval of 50 m based on the OpenStreetMap road network,and multi-temporal street view images in 2013,2015,2016,2017,2019,and 2020 were collected via the Baidu Maps Open Platform.In a Python environment,a Mask2Former semantic segmentation model trained on the Mapillary Vistas dataset was built and optimised to automatically extract building facades from street view images,and morphological indicators were computed,including area ratio(AR),edge contour roughness(ECR),and edge contours total perimeter(ECTP).Moreover,k-means clustering was performed on facade pixels in the CIELAB colour space by combining the Webcolors colour dictionary and a KD-tree structure to identify dominant colours and quantify colour features such as colour brightness(CB),colour contrast(CC),colour richness(CR),and colour saturation(CS).Using the Credamo online eye-tracking platform,a facade-viewing experiment was conducted with 76 participants to obtain eye-movement metrics,including first fixation duration(FFD),mean fixation duration(MFD),mean saccade amplitude(MSA),and saccades number(SN).Additionally,weights of indicators were determined using the analytic hierarchy process-entropy weight combination method,and a visual attention index(VAI)was built.On this basis,an XGBoost model was introduced to depict the nonlinear relationships between seven categories of facade features and VAI.SHAP and partial dependence plots(PDP)were used to examine the contribution,action direction,and threshold effects of each driving factor.The results show that:(1)From 2013 to 2020,the overall visual attention characteristics towards building facades in the six central districts remained relatively stable:the means of FFD and MFD stayed around 115 ms and within 113～118 ms,respectively.MSA fluctuated between 221 and 253 px;and SN remained at approximately 5,indicating that the overall attention load of the street interface did not change dramatically during the study period.(2)Morphological and colour features exhibited differentiated spatial evolution,and most indicators followed a spatial pattern of"high in the centre,but low in the periphery".Heping District consistently served as the core area of high-value clustering,although its high-value levels fluctuated to some extent over time.(3)The driving mechanism of VAI evolved from early dominance by colour attributes(CB and CR)to a later stage jointly dominated by composite morphological and colour features(AR and CC).(4)Clear threshold effects were observed between the driving factors and VAI,typically presenting N-shaped or U-shaped relations.After the morphological or colour indicators exceeded certain critical ranges,their positive effects on VAI weakened and could even become negative.These findings indicate that the formation of VAI in urban street facades is not a static outcome,but a dynamic evolution process from the immediate perception driven by direct colour differences to a more complex cognition based on morphology-colour combinations during urban construction transition and continuous appearance renovation.The facade micro-renewal should achieve an appropriate balance between factors such as AR,CC,and CS to avoid adverse effects caused by overemphasising a single attribute.Moreover,a quantitative evaluation framework was established by integrating multi-temporal street view images and eye-tracking data,which was conducive to identifying the time-series dynamic differences of facade-elated AVI.The results can provide scientific support for refined control and optimisation of building facades in the context of stock-oriented urban renewal.关键词
建筑外立面/视觉注意力/视觉感知/眼动追踪/深度学习Key words
building facade/visual attention/visual perception/eye-tracking/deep learning分类
建筑与水利引用本文复制引用
许沉风,朱梓博,李敏,程岩,胡一可..眼动追踪视角下城市街道建筑外立面视觉注意力影响机制[J].南方建筑,2026,(1):61-73,13.基金项目
国家自然科学基金重点项目(52038007):基于中华语境"建筑-人-环境"融贯机制的当代营建体系重构研究. （52038007）
