Estimating traffic volumes in an urban network based on taxi GPS and limited LPR data using machine learning techniques

In urban road networks, there are many segments without any detectors installed. The presence of these segments with unknown volumes makes it challenging to deduce city-wide network traffic conditions. However, for many of these networks, the speed data is available from GPS traces. This paper proposes a method to estimate city-wide traffic volume on urban road networks based on taxi GPS data and limited-scale License Plate Recognition (LPR) data. In this process, first, a similarity analysis is conducted by Jensen-Shannon (JS) divergence to distinguish whether the ‘unknown-volume segment’ is correlated with any set of ‘known-volume segments’ of the network. Secondly, the ensemble support vector regression (ESVR) method is deployed to estimate the traffic volumes based on features of taxi GPS data extracted from the ‘similar segment’ set. Finally, two case studies are conducted to demonstrate the performance improvements. Results show that the proposed framework can estimate the traffic volumes on two different types of traffic networks with reasonable accuracy.