A data fusion approach for estimating traffic distribution characteristics of expressway: A case study of guangdong province, china

Abstract

Currently, people pay more and more attention to road maintenance, and the traffic characteristics of vehicles play an important role in road quality evolution and maintenance decision, which commonly depends on the collection and analysis of traffic data. Nevertheless, the rationality of traffic data analysis and the scientificity of maintenance decision are deficient. This study carries out a research on the data fusion of multisource traffic data including toll data and video surveillance data. First, the information of vehicle type and axle load is acquired from the toll data, and the lane, speed and temporal information are obtained from the video surveillance data. A Bayesian method is used to train toll data and video surveillance data to recover missing data. The vehicle type distribution probabilities of traffic volume during different periods and speeds in different lanes are investigated. Next, the number of equivalent standard axle load (ESAL) at different lanes, time periods, and speeds are estimated based on the axle load conversion relationship between different vehicle types. Then the axle load spectrum and distribution characteristics of traffic in different sections, lanes, speeds, and time periods are analyzed. Finally, the comparison of rutting depth from the multisource data fusion and specification is carried out, and it shows an apparent difference (e.g., beyond 20%) when the lateral distribution in lanes is taken into account. Although the difference is less than 10% by considering vehicle speed and time periods, the time to reach the same value of rutting depth maybe more than 1 year. Therefore, it greatly affects accurate determination of preventive maintenance timing. As a whole, this study provides beneficial information for accurately understanding the preventive maintenance opportunities and making reasonable maintenance decisions.

A Bayesian decision support system for optimizing pavement management programs

“…Bayesian techniques have been widely used in pavement studies for various applications such as estimating the correlation of pavement friction with the risk of traffic crashes at intersections [ 19 ], recovering missing traffic data from toll data and video surveillance data [ 20 ], predicting thermal conductivity in cement composites in pavements [ 21 ] and many others. The uncertainties associated with factors that have a detrimental impact on the long-term performance of road networks and the existence of significant heterogeneity among various locations and climate conditions have motivated researchers to deeply investigate the strength of BBN methods for better pavement management solutions [ 22 ].…”