City-wide Traffic Volume Inference with Loop Detector Data and Taxi Trajectories

Chuishi Meng, Xiuwen Yi, Lu Su, Jing Gao, Yu Zheng

ACM SIGSPATIAL 2017 |

View Publication

The traffic volume on road segments is a vital property of the transportation efficiency. City-wide traffic volume information can benefit people with their everyday life, and help the government on better city planning. However, there are no existing methods that can monitor the traffic volume of every road, because they are either too expensive or inaccurate. Fortunately, nowadays we can collect a large amount of urban data which provides us the opportunity to tackle this problem. In this paper, we propose a novel framework to infer the city-wide traffic volume information with data collected by loop detectors and taxi trajectories. Although these two data sets are incomplete, sparse and from quite different domains, the proposed spatio-temporal semi-supervised learning model can take the full advantages of both data and accurately infer the volume
of each road. In order to provide a better interpretation on the inference results, we also derive the confidence of the inference based on spatio-temporal properties of traffic volume. Real-world data was collected from 155 loop detectors and 6,918 taxis over a period of 17 days in Guiyang China. The experiments performed on this large urban data set demonstrate the advantages of the proposed framework on correctly inferring the traffic volume in a city-wide scale.