A Layered Hidden Markov Model for Predicting Human Trajectories in a Multi-floor Building

Tracking and modeling the movement of large number of users in a multi-floor building using wireless devices is a challenging task. This is due to the complexity of crowd movement and the accuracy of signal sensing data. In this paper, we use Layered Hidden Markov Model (LHMM) to fit the spatial-temporal trajectories (with large number of missing values). We decompose the problem into distinct layers that Hidden Markov Models (HMMs) are operated at different spatial granularities separately. Baum-Welch algorithm and Viterbi algorithm are used for finding the probable location sequences at each layer. By measuring the predicted result of trajectories, we compared the predicted results of both standard HMM and LHMM though 2D/3D path plotting, execution time and trajectory distance. The results indicate that LHMMs are better than HMMs for modeling and predicting incomplete, long-distance temporal-spatial trajectories data.