Spatial-Temporal Based Traffic Speed Imputation for GPS Probe Vehicles

Abstract

Due to the growth of vehicular network and big data analytics, missing data of traffic detector devices become a serious problem in analytics and applications of intelligent transportation systems. The purpose of data imputation is to complete the shortage of traffic data. In this paper, a spatial-temporal based data imputation for GPS probe vehicle in intelligent transportation systems is proposed. In the proposed system, GPS data with speed of vehicles are located into the map within corresponding road segments by GPS coordinates using R+-tree and Dijkstra's algorithm. Then, spatial features are extracted from the current road segment and its two neighboring segments' speeds, and temporal features are extracted from the current time sector, weekday, and speeds of the current road segment in 5 and 10 minutes ago, respectively. After that, each model of road segment is trained by support vector regression with spatial-temporal features for data imputation. Experimental results show that the proposed scheme is better than Gaussian processing with time series feature at different missing rates.