Abnormality Detection In Video Using Gaussian Mixture Model And Recurrent Conditional Random Field

Abstract

Surveillance system is helping to identify various problems in the video. Detection of anomalous events with long video sequence is challenging due to the ambiguity.  The explored work aims to identify anomalies with different dataset such as pedestrian (UCSD, UMN, Avenue), subway and traffic (QMUL). Existing model Gaussian Kernel Integration Model (GKIM) used for feature extraction done by combination of spatial-temporal information integrated into the Gaussian equation. Proposed works Gaussian Mixture Model (GMM) has been used for extracting features. Here spatial and temporal information extracted and feature selection is done in GMM itself. Recurrent Neural Network (RNN) with Conditional Random Field (CRF) has been used for classification with the selected by GMM features. Performance Analysis of GKIM and GMM has been considered with respect to Error and Detection Rate.