Development of An Event-Based Dataset For Abnormal Activity Detection

Event cameras capture data in a non-sequential manner, which differs from the frame-based approach of conventional video datasets. This work aims to address the limited availability of event data that has hindered research in this field. To overcome this challenge, we developed a framework that transforms conventional video datasets into asynchronous event datasets for abnormal activity detection. We then apply the developed algorithm to the UCSD Anomaly Detection Dataset. Additionally, we employ the Local Outlier Factor (LOF) clustering technique for abnormal event detection. To enhance the understanding and interpretation of the detected abnormal events, we employ a 3D visualization approach. The ability to gain insights into abnormal activities aids decision-making processes and opens up new avenues for modern AI research utilizing event-based technologies.