Behavior Analysis in Video Surveillance Applications

Somboon Hongeng


Abstract

The goal of video surveillance applications is to monitor a scene and trigger an alarm to alert humans if there is an abnormal activity. To achieve this goal, we must be able to describe activities that occur in an image sequence. In the first half of this talk, I briefly describe the algorithm of the current behavior analysis module that we use for USC VSAM project. More precisely, we explain how we use context to compute the properties of moving objects and how we recognize a scenario describing human activities as a combination of sub-scenarios and properties on the mobile objects involved in the scenario. Depending on the type of combination (temporal or non temporal) we check whether some specific constraints are verified, or we use automata as methods to recognize scenarios. We then illustrate how the scenario recognition module works through an example of utilization.

In the second half of the talk, I underline the major problems that we have to face with during the development of the current system: the difficulty of the parameter initialization process , the tuning of parameters as the scene environment evolves, and the need of probabilistic automaton. Then, I describe the Bayesian framework that we propose to use to automatically learn and tune the parameters and explain a technique that we will adopt for probabilistic automaton. Finally, I illustrate how we can learn the parameters through some new sequences that we took and how we will use these parameters to analyse behavior in other sequences.

On-line references

Video Surveillance And Monitoring


Maintained by Alexandre R.J. FRANÇOIS