Hidden Markov Model-based Occupancy Grid Maps of Dynamic Environments

For a reliable localization in dynamic environments, a robust representation is of vital importance to obtain an accurate position estimation. This paper introduces a hidden Markov model-based approach to obtain robust representations of dynamic environments. The model uses occupancy grid maps created at different times as observations. The approach involves a grid map registration process for pre-processing to align new observations impacting the robust representation. It uses a combined feature-based registration method and NDT-based refinement. For the robust representation, a hidden Markov model is used to estimate the probabilities of static and dynamic states of cells based on observations. The robust representation is updated with each new observation using an iterative propagation algorithm. Experiments on real world radar data demonstrate that a localization algorithm based on this method provides a more reliable localization performance than a standard approach.