Improving Public Safety: Through Real-Time Anomaly Detection in Surveillance Videos Using 3D CNN and Spatiotemporal Autoencoder
2025
In today's fast-changing world, surveillance cameras are being installed for safety maintenance and crime prevention.However, detecting abnormal behaviors in real time still seems to be a difficult task.Previously, 2D Convolutional Neural Networks were used to analyze the frames of the videos.This methodology is effective in identifying spatial patterns but is unable to discriminate temporal patterns.Therefore, in realworld systems, it is difficult to identify anomalies using 2D CNNs due to their limitations.Our system, called EyeSpy, overcomes these issues by introducing anomaly detection in surveillance videos using a SpatioTemporal Autoencoder based on 3D CNNs.The performance of the proposed model is assessed on the dataset of UCF-ShanghaiTech.To identify anomalous or suspicious activity inside a video stream, the 3D CNN model is trained to collect both spatial and temporal data in video frames.With a focus on attaining precise anomaly identification in real-time applications, the system is built to handle large video datasets.The detection of abnormal activities shows how well the model detects unusual occurrences, providing a reliable way to improve security monitoring in a range of surveillance settings.