Abnormal Event Behavior Detection For Home Surveillance Cameras System

Home surveillance cameras have a crucial role in ensuring the security and safety of residential premises. However, analyzing the vast amount of video data generated by these cameras manually is a time-consuming and challenging task. This study designed a camera work system consisting of three main modules namely object detection, face recognition, and anomaly behavior to detecting abnormal event behavior. Object detection used the Histogram Oriented Gradient and Support Vector Machine (HOG-SVM) combination. Facial recognition used Convolutional Neural Network (CNN) with the Visual Geometry Group-16 (VGG16) architecture that utilized transfer learning. Anomaly detection used spatiotemporal autoencoder based on threshold. Modules are tested using the standard metrics such as accuracy, precision, recall, and f1score. The three modules are integrated using a state machine into one system that can detect the abnormal event behavior. The overall accuracy of the system reached 84%, the precision value was 84.615%, the recall value was 94.286%, and the f1score was 89.189%.