Multi Sensor Loitering Detection Using Online Viterbi

In this paper the problem of loitering detection in image sequences involving situations with multiple objects is studied. A multi camera approach is used to incorporate data from several cameras viewing the same scene. A Hidden Markov Model describing the movements of a varying number of objects as well as their entries and exits is used. The maximum likelihood over all possible state sequences is found using online Viterbi optimisation. Previously similar models have been used for single camera setups. In this paper the technique is extended to allow several cameras. The model is also made less sensitive to uninteresting objects occluding the region of interest, by integration out their effect on the observation probabilities. Finally the online Viterbi optimisation technique is extended to be able to produce it’s output after a constant number of frames. In previous work this delay varied with the complexity of the data. The resulting system is tested on the PETS2007 dataset scenarios S0-S2, with promising result.