Identifying Abnormal Human Activity Behavior In Diverse Crowds Using Transfer Learning

Deep learning approaches transform traditional human behavior recognition in video surveillance into more intelligent techniques. This shift offers numerous advanced features that can enhance safety measures in public gatherings. By enabling proactive surveillance, identification, and oversight of diverse crowds, this new paradigm can significantly improve various crowd management operations in terms of effectiveness, capacity, predictability, and safety. Convolutional neural networks (CNNs) present a promising method for recognizing human actions in large datasets, such as those generated by CCTV footage, despite challenges like occlusion, clutter, uneven item distribution, and non-uniform object scale. Therefore, this study proposes a CNN-based Abnormal Classifier model to identify abnormal behavior in public venues such as airports, malls, and theaters. The proposed model classifies behaviors but also delivers definitive results in behavior classification. Experimental results indicate that this method slightly outperforms existing approaches.