Abstract
In this report, we propose a method for anomality detection by classifying people behavior patterns based on situation. In public spaces, people in most cases behave quite typically based on some regular patterns. The proposed method learns these patterns from the observed trajectories by composing Hidden Markov Model for each separate situations such as train arrival and departure. The anomalous behaviors are detected by thresholding the output probability. Over 2,500 trajectories observed in an actual station were used for evaluation, which resulted in a fine performance with the overall classification rate of 94.2%.
