A Mixed Markov Model Approach To Predict Future Points Of Interest In Indoor Space

Open Access
Sun, Ying
Graduate Program:
Electrical Engineering
Master of Science
Document Type:
Master Thesis
Date of Defense:
March 29, 2013
Committee Members:
  • David Jonathan Miller, Thesis Advisor
  • prediction; trajectory pattern; Mixed Markov Model
With the advances in GPS and other location acquisition technologies, an increasing amount of trajectory data is being captured and recorded. This trajectory data has attracted the attention of researchers from many domains due to the potential benefits of discovering the underlying behavior patterns and predicting future trajectories of the users. In this thesis, we focus on indoor trajectory data in an attempt to detect heterogeneous trajectory patterns across users within the raw x-y coordinated data records in an indoor environment and further predict individuals’ future trajectories. Due to the heterogeneity in human behaviors, individuals do not exhibit the same pattern in their trajectories. Consequently, one single trajectory model is not capable of capturing these behavior patterns. To tackle this problem, we propose a Mixed Markov Model (MMM) approach that models the latent trajectory patterns from all input trajectories without the need of user identification. This study may show high impact in domains, such as health care, transportation systems, public security, etc., where privacy issue and heterogeneity in behavior patterns are concerned. A case study using real indoor trajectory data of workers in an engineering design space is presented for validating our prediction model.