Analyze the visual quality of roads utilizing deep learning algorithms and street view images

Abstract

Visual quality of roads is one of very important factor in road design. However, in current practice objective analysis and judgment is lacking and mostly utilized subjective judgment to evaluate the visual quality of road. Therefore, this study attempts to develop a data driven framework to quantify the visual quality of roads. For that purpose, the study utilized deep learning algorithms and street view images. The study comprised of four-staged. The study developed two main models to quantify the quality of streets and conducted a validity assessment consisting of both internal and external validation to test the effectiveness of the proposed framework. The proposed framework achieved 90.51% internal validation accuracy using the tenfold cross validation technique and 86.7% external validation accuracy. Further, the framework recorded an accepted level of kappa accuracy of 80%. Accordingly, the study concludes that proposed framework and models would be effective tools for transport planners and street designers to objectively measure and map the visual quality of roads and proposed street designs.