People Flow Trend Estimation Approach and Quantitative Explanation Based on the Scene Level Deep Learning of Street View Images

People flow trend estimation is crucial to traffic and urban safety planning management. However, owing privacy concerns, the collection of individual location data for people statistical analysis difficult; thus, an alternative approach urgently needed. Furthermore, in reflected streetscape factors, yet relationship between them remains unclear existing literature. To address this, we propose end-to-end deep-learning that combines street view images human subjective score each view. For a more detailed study, were implemented using different time movement patterns. Consequently, achieved 78% accuracy on test set. We also gradient-weighted class activation mapping deep learning visualization L1 based methods proposed quantitative understand land scape elements feeling identify effective gradient impact method. In summary, this study provides novel sheds light streetscape, feeling, trend, thereby making important contribution evaluation development.