Semantic-aware trajectory compression with urban road network

Abstract

Vehicles are generating more and more trajectories, which are expensive to store and mange. Thus it calls for vehicular trajectory compression techniques. We propose a semantic-aware compression framework that includes two steps: Map Matching(MM) and Semantic Trajectory Compression(STC). On the one hand, due to measurement errors, trajectories cannot be precisely mapped to real roads. In the MM step, we utilize multidimensional information(including distance and direction) to find the most matching roads and generate aligned trajectories. On the other hand, some unnecessary points in trajectories can be reduced based on the roads. In the STC step, we extract a set of crucial points from the aligned trajectory, which capture the major driving semantics of the trajectory. Meanwhile, our decompression method is fairly lightweight and can efficiently support various applications. Empirical study shows that MM achieves high matching quality, STC achieves more than 8 times compression ratio, and decompression is efficient on real datasets.