Several countries have implemented designated school zones and installed traffic calming measures to enhance the safety of vulnerable pedestrians near schools. While macro-level built environment attributes (e.g., land use) have been widely acknowledged in relation to the role they play in urban traffic safety, the effects of micro-level streetscape characteristics on crash frequency have not been investigated to any significant extent. Moreover, the associations between these environmental features and crashes in school zones remains largely unknown. To address this issue, we first identified school zone-related crash hotspot using spatiotemporal hotspot mining on a comprehensive dataset of 20,484 pedestrian-vehicle crashes between 2017 and 2021 in Seoul, South Korea. Streetscape characteristics were analysed using street view imagery and advanced computer vision techniques to extract and classify pixel-wise visual elements. Preliminary findings reveal spatiotemporal variations in fatal and severe injury (FSI) crashes, with school zones in central commercial and industrial areas emerging as persistent crash hotspots that have remained statistically significant hotspots for 90 % of the study period. Further impact analysis using interpretable machine learning helped to uncover the non-linear relationships between both micro and macro environmental features and FSI frequency. Lower levels of street enclosure and walkability were associated with a higher frequency of FSI crashes, while increased openness and imageability were also correlated with more FSI incidents. Additionally, street greenery was found to reduce FSI crashes once it reached a certain threshold. Our findings extend existing knowledge of how the built environment and streetscape design influence pedestrian safety in school zones, paving the way for more targeted interventions to plan safer pedestrian environments around schools.

中文翻译：

---

探索学区碰撞热点的致命/严重行人伤害碰撞频率：使用可解释机器学习评估微观层面的街道环境


一些国家已经实施了指定的学区并安装了交通平静措施，以提高学校附近弱势行人的安全。虽然宏观层面的建筑环境属性（例如土地利用）在城市交通安全中的作用已得到广泛认可，但微观层面的街景特征对碰撞频率的影响尚未得到任何重要研究。此外，这些环境特征与学区车祸之间的关联在很大程度上仍然未知。为了解决这个问题，我们首先在 2017 年至 2021 年韩国首尔 20,484 起行人车祸的综合数据集上使用时空热点挖掘确定了与学区相关的车祸热点。使用街景图像和先进的计算机视觉技术分析街景特征，以提取和分类像素级视觉元素。初步研究结果揭示了致命和重伤 （FSI） 车祸的时空差异，中央商业和工业区的学区成为持续的车祸热点，在 90% 的研究期间仍然是具有统计学意义的热点。使用可解释机器学习的进一步影响分析有助于揭示微观和宏观环境特征与 FSI 频率之间的非线性关系。较低的街道围护和步行性水平与 FSI 事故频率较高相关，而开放性和可成像性的增加也与更多的 FSI 事故相关。此外，一旦达到一定阈值，街道绿化就会减少 FSI 车祸。 我们的研究结果扩展了关于建筑环境和街景设计如何影响学区行人安全的现有知识，为更有针对性的干预措施铺平了道路，以规划学校周围更安全的行人环境。