A Parallel Fusion Method for Heterogeneous Multi-sensor Transportation Data

Information fusion technology has been introduced for data analysis in intelligent transportation systems (ITS) in order to generate a more accurate evaluation of the traffic state. The data collected from multiple heterogeneous traffic sensors are converted into common traffic state features, such as mean speed and volume. Afterwards, we design a hierarchical evidential fusion model (HEFM) based on D-S Evidence Theory to implement the feature-level fusion. When the data quantity reaches a large amount, HEFM can be parallelized in data-centric mode, which mainly consists of region-based data decomposition by quadtree and fusion task scheduling. The experiments are conducted to testify the scalability of this parallel fusion model on accuracy and efficiency as the numbers of decomposed sub-regions and cyberinfrastructure computing nodes increase. The results show that significant speedups can be achieved without loss in accuracy.