Recognizing mobile organizations from column formations using hierarchical hidden Markov models: a simulation experiment

In a paper at Fusion 2000, Bjornfot and Svensson discussed the use of HMMs to recognize mobile military organizations moving in column formation along a road from observations of the types and relative positions of its component vehicles. Straightforward use of standard HMMs in the sequence recognition task restricts the application of this technique to the rather unlikely situation where all vehicle types normally follow each other according to a fixed sequential pattern. In this paper, this restriction is relaxed by representing a multi-level organization by a Hierarchical Hidden Markov Model, in which each possible component sequence for each organizational subunit type is modeled by a separate "sub-HMM". Experiments are carried out using simulated data which illustrate the increased flexibility and applicability of this approach in column recognition tasks.