Audio analysis for surveillance applications

We proposed a time series analysis based approach for systematic choice of audio classes for detection of crimes in elevators in R. Radhakrishnan et al. (2005). Since all the different sounds in a surveillance environment cannot be anticipated, a surveillance system for event detection cannot completely rely on a supervised audio classification framework. In this paper, we propose a hybrid solution that consists two parts; one that performs unsupervised audio analysis and another that performs analysis using an audio classification framework obtained from off-line analysis and training. The proposed system is capable of detecting new kinds of suspicious audio events that occur as outliers against a background of usual activity. It adaptively learns a Gaussian mixture model (GMM) to model the background sounds and updates the model incrementally as new audio data arrives. New types of suspicious events can be detected as deviants from this usual background model. The results on elevator audio data are promising.