LDA Based on Real-time Classification of CCTV Systems Using Codeblocks Information

Today, CCTV systems play a preponderant role in public security. While we are witnessing many tools for processing these video sequences, the potential to identify abnormal events in such sequences in real time is a difficult problem in computer vision. Using the Latent Dirichlet Allocation (LDA) has brought significant results. In this article, we propose an optimization of the LDA method by relying on information codeblocks to provide a better classification of abnormal events in real-time. The implementation is tested on the normalized PETS datasets. The results obtained make it possible to identify the `Running' event considered to be abnormal with an accuracy ranging from 86.55%, informed by the assessment of the proposed method with the ROC curves.