An Advanced Autoencoder-Based Approach to Anomaly Detection for Video Surveillance Systems
Abhiram G, Madineni Nitheesha, Ambati Samatha, S Giriprasad · 2024
This paper attempts to give a state-of-the-art machine learning anomaly detection in video surveillance using the application of autoencoders. The use of video surveillance ensures safety and security in places like public spaces, office buildings, and homes. However, with the multiple cameras present, monitoring various video feeds is quite demanding and, thus, predisposed to human error or inefficiency when dealing with large environments. The single critical feature that will serve to automate this process pertains to identifying unusual events or behavior. The proposed system relies on an autoencoder neural network model for learning patterns of normal behavior in video streams. The UCSD, a diverse dataset containing surveillance videos, was used to train and test the system. Preprocessing of video data includes frame extraction, resizing, and normalization, followed by feeding into the model. It mainly consists of two parts, namely an encoder that compresses the input data and a decoder that reconstructs the input from the compressed representation. The reconstruction error is then calculated by comparing the original and reconstructed frames. The frames with high reconstruction errors are flagged as containing anomalies. This system shows a high level of accuracy measured in terms of precision and recall. In addition to real-time anomaly detection, the system offers visual feedback to the user regarding which frames are anomalous and can be further inspected. The experiment results demonstrate how the model is robust against vast and complex environments. The proposed system extends a reliable, highly scalable, and flexible solution for smart surveillance, reducing constant human supervision, and improving the speed of response to security threats. More work will go into enhancing the system so that it can better deal with subtler anomalies and generalize to larger datasets for ultimate deployment.