Estimating Total Traffic Volume with Statistical Modeling Approach

The measurement of total traffic volumes plays a central role in modern intelligent transportation systems. In this study, we regard this issue as a total traffic volume estimation task based on synchronous partial observations of probe vehicles. Conventionally, measurements by loop detectors have been used to collect total traffic volume data. However, the number of loop detectors is severely limited owing to the high maintenance costs. Therefore, improving the feasibility of extended estimations by combining loop detector and probe vehicle data has gained attention. We propose an accurate and lightweight algorithm to estimate the total traffic volume with a mixed effect model by leveraging the statistical modeling approach. This algorithm proposes to incorporate various factors, such as road type, day and hour traffic frequency, and road network topology, into a model. We further introduce a model fusion algorithm to reduce the total number of model sizes built in various areas, such as cities, states, and prefectures. The effectiveness of this algorithm was demonstrated by experiments conducted in 18 cities of Japan.