Detection of anomaly in surveillance videos is a critical task for better public safety. This study used a Deep Learning (DL) approach to classify and detect anomalous events in real-time video feeds. MobileViT model was used for classification purposes which is the combination of the CNN and Transformer model showed 93% accuracy for 7 classes and 98% accuracy for 6 classes on the UCF-Crime dataset. The UNI-Crime dataset was solving a binary classification problem and showed 100% accuracy. These results are much higher than stateof- the-art work. After classification, the classified images are passed to the YOLO-v11 model and detect the part of anomalous activity on the selected Hyperparameters. The YOLO-v11 model was applied to the classification model output for accurate localization of anomalies. The result shows that the proposed models are effective in identifying anomalous events. This study has opened new opportunities for developing real-world applications that improve surveillance systems in various environments. The promising outcomes of this research provide a robust basis for future work in anomaly detection, with the potential to enhance both security and effectiveness in video-based surveillance.