YOLOv8 for Anomaly Detection in Surveillance Videos
Rejuwan Shamim, Badria Sulaiman Alfurhood, Trapty Agarwal, Biswadip Basu Mallik · 2024
Security and safety rely heavily on anomaly detection in surveillance footage. In this study, we propose a modified YOLOv8 model for anomaly detection to pinpoint unusual occurrences in video surveillance footage. When tested on the UCF-Crime dataset, a popular benchmark for anomaly detection, the modified YOLOv8 model shows an impressive performance in terms of accuracy (89.6%), recall (88.2%), precision (90.4%), and F1 score (88.9%). We demonstrate the superiority of the modified YOLOv8 model in reliably detecting and localizing anomalies through comprehensive trials and comparisons with state-of-the-art approaches. We also investigate how dataset size and training duration affect the model's final output. Cases in which the model successfully detected anomalies are depicted visually to demonstrate the model's efficacy. Security and surveillance systems, public safety, industrial monitoring, traffic management, and retail loss prevention are some areas that could benefit from the suggested method's use of the modified YOLOv8 model.