Hierarchical Trajectory Matching via Multilayered Map Model

As the traditional indoor localization methods, such as pedestrian dead reckoning (PDR) and fingerprinting-based ones, may fail due to their instinct problems, the spatial information of the indoor environment is proposed to refine their trajectories by matching the trajectory to the map model. However, the time cost of the traditional map-matching methods increases dramatically with the complex model generated in a large environment. In this article, we propose a hierarchical map-matching technology to reduce the time cost while maintaining reliable localization accuracy by the recursive trajectory inference based on the multilayered map model. As the trajectory is mapped to the model from a coarse to a fine level, the time cost is reduced significantly by segmenting the trajectory into parts for inference. Our method is composed of two stages. First, the indoor environment is divided into large subregions according to their connections and locations to build a coarse model with global spatial information. Each subregion is then divided into smaller and local ones recursively to generate a region tree, whose node is represented as the subset of the total reference points and their transition matrix. We treat the trajectory matching as a hidden Markov model (HMM) and map the points of the trajectory to the coarse model initially. Then, the trajectory is segmented by the inference with the subregions of the coarse model and the process is conducted recursively in the region tree until the final result is achieved at the leaf nodes. The results at the nodes are combined according to their sequential information as the output of our method. In the experiment, we prove that the proposed method achieves at least
$20\times $
acceleration than the traditional methods and the localization accuracy degrades little which is comparable to the result of mapping to the whole fine map model.