| Title |
Application and Validation of a Deep Learning Model to Predict the Walking Satisfaction on Street Level |
| Authors |
박근덕(Park, Keundeok) ; 이수기(Lee, Sugie) |
| Keywords |
도시설계 ; 보행가로 ; 보행친화도 ; 딥러닝 ; Google Street View Urban Design ; Walking Street ; Walkability ; Deep Learning ; Google Street View |
| Abstract |
This study examines the prediction model of walking satisfaction level of streetscape by deep learning technique. The model focuses on the streetscape imagery and walking satisfaction level of pedestrians. We trained and tested the prediction model using the Google Street View 360° Panorama images of the survey locations for walking satisfaction level. First, the correlation coefficient between machine rating and human rating regarding walking satisfaction level ranged from 0.16 to 0.84 by the transfer learning method. This finding indicates that deep learning models should be tested and validated for the purpose of specific research topic. Second, among four test models, VGG16 is relatively suitable for the prediction model of walking satisfaction level on streetscape comparing to the Inception structure in this study. Third, the result shows that transfer learning with pre-trained model could be the best one to predict average walking satisfaction level using Google Street View images. Lastly, This study shows that deep learning skills could play an important role in analyzing and predicting urban physical environments with open source imagery data. |