Spatiotemporal Modeling of Street Vitality Using Color Semantic Segmentation and Pedestrian Trajectory Prediction

Urban street vitality is a critical yet under-quantified metric for evaluating public space performance, especially in dynamic and heterogeneous urban settings.This paper proposes a spatiotemporal modeling methodology to assess street vitality by integrating color semantic segmentation and pedestrian trajectory prediction.To address the fragmentation of existing vitality assessment approaches, this study aims to systematically characterize urban street vitality and uncover spatiotemporal evolution patterns using multi-source data.Deep convolutional neural networks (DCNs) are employed to analyze street scene images, extracting color distribution and semantic features to reflect functional and aesthetic aspects of streets.Meanwhile, a long short-term memory (LSTM) network is used to capture spatiotemporal movement patterns of pedestrians, highlighting dynamic street usage.A weighted fusion framework is developed to comprehensively integrate static visual elements with dynamic behavioral data.Experimental results demonstrate that this approach effectively extracts visual richness and street activity, outperforming existing uni-modal evaluation baselines in both accuracy and interpretability.Compared to color-only and trajectory-only models, the proposed method improves vitality prediction accuracy to 91.7%, with a significant reduction in error metrics.This study establishes an original theoretical framework and technical roadmap for a multidimensional assessment of urban street vitality.The findings provide insights for data-driven urban design and governance, contributing to more adaptive and sustainable public spaces.