Inferring border crossing intentions with hidden Markov models

Law enforcement officials are confronted with the difficult task of monitoring large stretches of international borders in order to prevent illegal border crossings. Sensor technologies are currently in use to assist this task, and the availability of additional human intelligence reports can improve monitoring performance. This paper attempts to use human observations of subjects' behaviors (prior to border crossing) in order to make tentative inferences regarding their intentions to cross. We develop a Hidden Markov Model (HMM) approach to model border crossing intentions and their relation to physical observations, and show that HMM parameters can be learnt using location data obtained from samples of simulated physical paths of subjects. We use a probabilistic approach to fuse "soft" data (human observation reports) with "hard" (sensor) Additionally, HMM simulations are used to predict the probability with which crossings by these subjects might occur at different locations.