The popularity of video anomaly detection may be attributed to its use in security cameras. A variety of plausible abnormalities may be captured by surveillance cameras. Convolutional neural networks (CNNs) have recently found great success in the field of video analysis, and as a result, more and more AD algorithms are incorporating CNNs into their own to improve processing speed, efficiency, and detection accuracy. In this research, we use a novel approach to anomaly identification in video surveillance by means of a unique algorithm. To improve performance and achieve high accuracy, we employed the Resnet18+DeeplabV3 anomaly detection method enhanced with PSO. We show how our methods can automatically indicate the difference between regular and suspicious actions using real-world video footage, which may help with security monitoring. The proposed method shows much better performance with 91.2% accuracy which is higher that present methodologies.

Video anomaly detection, Resnet18, DeeplabV3, PSO.