Forecasting Traffic Speed during Daytime from Google Street View Images using Deep Learning

We forecasted traffic speeds for each hour of the day for the city of Porto. Figure 6 shows predicted traffic speeds at the city scale (taken at 6 a.m., noon, 6 p.m., and midnight.). The map shows that, spatially, the traffic speed of highways is much higher than that of city roads. Traffic speed is lowest at the center of the city. Temporally, traffic speed is higher at night than at any other time, and traffic speed is lowest during commuting hours (6 p.m. in this map). The red lines in Figure 6 have the lowest traffic speed, revealing areas of traffic congestion.

Abstract

This study addresses the challenge of obtaining comprehensive historical data for city-wide traffic forecasting by utilizing SceneGCN, a deep learning approach. SceneGCN leverages Google Street View (GSV) images and pretrained Resnet18 models to extract scene features, followed by a graph convolutional neural network for predicting traffic speed across various times of day. Results demonstrate the model's effectiveness, achieving up to 86.5% accuracy, with Resnet18 pretrained by Places365 proving to be the optimal choice for feature extraction. This study highlights the potential of GSV images in capturing detailed human activity for city-scale traffic speed prediction.


Team

Junfeng Jiao, Huihai Wang

Author Contributions

The authors confirm contribution to the paper as follows: study conception and design: H. Wang; data collection: H. Wang; analysis and interpretation of results: H. Wang; draft manuscript preparation: H. Wang, J. Jiao. All authors reviewed the results and approved the final version of the manuscript.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article

The cover image is sourced from Pexels and is free of copyright issues.