Real-Time Abnormal Event Detection in the Compressed Domain of CCTV Systems by LDA Model

Abstract

For storage and displaying, huge amounts of data are produced by the CCTV (Close Circuit TeleVision) systems. The automatic detection of abnormal events in the compressed domain of these systems makes it possible to extract these events to alert and possibly store the sequences in a compressed format to optimize the capacity of storage and transfer of data. This paper describes a solution for a real-time abnormal event detection. The proposed method is based on the LDA model for classifying events in the compressed domain in CCTV systems. Experimental results, demonstrating reliable real-time extractions and storage, shows that the classification of events with the LDA model allows the extraction of abnormal events in the compressed domain at very high compression rate with an accuracy of 95% for two standardized datasets considered.