Title:CiNCT: Compression and retrieval for massive vehicular trajectories via relative movement labeling

Abstract:How can we compress a data structure containing millions of vehicular trajectories enough to be processed by a small memory device, such as an in-vehicle navigation system or IoT device? Can we retrieve the trajectories along a given path from such a compressed representation? To address these questions, this paper considers a spatial index compression problem and the related retrieval problem for a huge number of vehicular trajectories that are represented as sequences of road segment IDs. We focus on FM-index, a fast and compact data structure for pattern matching. Our method, CiNCT, achieves a high compression ratio by using a relative movement labeling (RML) scheme based on the sparsity of a road network. We also describe a novel algorithm that enables fast search under the encoding. In addition, we prove the optimality of RML and give a theoretical explanation why CiNCT outperforms existing methods in terms of index size and query processing time. Experiments with real trajectory datasets demonstrate that CiNCT significantly outperforms existing methods, including several variants of FM-indexes.