The Smart Optimization of Anomaly Detection in Video Image Processing using Swin-Transformer
Anil Kumar Gupta, Rupak Sharma, Rudra Pratap Ojha · 2025
The video snapshot processing undertakes the critical assignment of anomaly detection that enables the identification of any unexpected or unusual events that can take place in a current video flow. Traditional systems for anomaly detection in this space face borders that encompass a sluggish processing pace and coffee accuracy. We recommend using this transformer-based neural network architecture, SwinTransformer, to overcome the above challenges. This method has sustained promising gains throughout numerous computer idea tasks, such as object detection and image classification. Our method leverages the self-attention mechanism of the SwinTransformer to explore the spatiotemporal relationship in video sequences, providing a promise that anomaly detection performance can also be significantly improved. Further, we include a clever optimization strategy that uses a pre-trained model and fine-tuning to speed up the training time and boost performance. Experiments on benchmark datasets demonstrate the efficacy of our method, outperforming the state-of-the-art approaches.