Effective and Efficient Predictive Density Queries for Indoor Moving Objects

Abstract

Density queries are defined as querying the dense regions that include more than a certain number of moving objects. Previous research studies mainly focus on how to answer the snap-shot density queries over historical trajectories. However, the real applications usually tend to predict whether a region is a dense region. Especially in indoor environments, such predictive density queries are valuable for high-level analysis but face tremendous challenges. In this paper, by leveraging the Markov correlations, we effectively predict the future locations of moving objects and conduct the density queries accordingly. In particular, we present an optimized framework which contains three phases to tackle this problem. First, we design an index structure based on the transition matrix to facilitate the search process. Second, we propose the space and probability pruning techniques to improve the query efficiency significantly. Finally, we apply an accurate method and an approximate sampling method to verify whether each unpruned region is a dense region. Extensive experiments on real datasets demonstrate that the proposed solutions can outperform the baseline algorithm by up to $2$ orders of magnitudes in running time.