Video Anomaly Detection Based on HSOE-FAST Modified Deep Neural Network

In recent times, video surveillance has become indispensable for public security, leveraging computer vision advancements to analyze and comprehend lengthy video feeds. Anomaly detection and classification stand out as crucial components of this technology. Anomaly detection's primary objective is to swiftly identify irregularities within a given timeframe. It is promising to use Deep Neural Network (DNN) approaches for anomaly detection because they combine the ideas of deep learning and reinforcement learning, enabling artificial agents to learn from and gain insights from real-world data. A modified DNN (Deep Neural Network) technique known as HSOE-FAST (Histo sigmoid of Orientation and Enthalpy with Fast Accelerated Segment Test) was proposed in this study. From the dataset, the input is obtained. Initially, the input is pre-processed using a Gaussian filter followed by the feature extraction using the HSOE-FAST method and finally classification is done using the modified DNN approach. Compared to other approaches, our suggested solution achieves an accuracy of almost 99% while overcoming the shortcomings of the existing methodologies.

.