Fog-Assisted Abnormal Motion Detection System: A Semantic Ontology Approach

The growing concern over high-profile violence has led to a widespread adoption of intelligent smart video surveillance systems in educational institutions. However, accurately identifying abnormal events in the video stream remains a complex task due to the absence of a definitive generic definition for abnormal actions, heavily reliant on contextual factors. In this paper, we propose a fog-assisted abnormal motion recognition system utilizing an ontology-based semantic algorithm to bolster security within university campuses under real-time surveillance. Our ontology-based semantic analysis is meticulously designed to characterize intricate spatio-temporal interactions and contextual relationships in the video scene, thereby significantly improving the sensitivity in distinguishing abnormal events based on their severity level. To ensure swift responses to specific abnormal events, the video streams captured through surveillance cameras are processed at the edge of the network. Prior to semantic analysis, two crucial steps—foreground object segmentation and object tracking—are executed to streamline the detection process. These steps involve segmenting the target foreground using a connected components labeling algorithm and tracking motion patterns based on velocity, direction, and acceleration, employing the Kalman filter. The rule-based reasoning of the ontology accurately defines abnormal conditions in the video scene, providing a clearer understanding of the decision-making process. Furthermore, we incorporate a context-aware refinement step aimed at enhancing detection accuracy. This step distinguishes abnormal events based on their severity and generates corresponding alerts. We conducted extensive experiments and evaluated the proposed system using various measures, demonstrating its potential in real-time video analysis. The results showcase an impressive prediction success ratio of 0.989, affirming the reliability and robustness of our system in detecting abnormal events.