Analysis of Deep Models for Abnormal Event Detection in Video Surveillance

The happening of abnormal events compared to normal events is hardly rare. When abnormal events occur in a place that is under a surveillance system, those surveillance systems are capable of detecting the type of abnormal events by differentiating them from normal events, which will enhance the security system and be useful in rapid incident response. Scene classification from surveillance video is intrinsically complex, because of different dynamics in nature and human intervention. In this paper to address these challenges, the proposed method combines advanced deep learning techniques to effectively process and analyse video data. Convolutional Neural Networks (CNNs) such as VGG19 and ResNet-101 are employed to capture spatial features, while Recurrent Neural Networks (RNNs) with Long Short-Term Memory (LSTM) units and Gated Recurrent Units (GRUs) are utilized to model temporal dependencies within the video sequences. Additionally, Support Vector Machines (SVMs) are explored for their classification capabilities in this context. Experiments were conducted on a comprehensive dataset comprising CCTV footage representative of each event category. The proposed system shows significant promise in distinguishing between different normal and different kinds of abnormal events.