Abnormal Event Detection in Videos using LSTM Convolutional Autoencoder

Abstract

This research investigates the application of a novel LSTM Convolutional Autoencoder (LSTM-CAE) framework for video anomaly detection, utilizing the UCSD Anomaly Detection Dataset. Traditional methods in video anomaly detection often face challenges in capturing both temporal and spatial dependencies within dynamic scenes. The proposed LSTM-CAE integrates Long Short-Term Memory networks and Convolutional Autoencoders to overcome these challenges, demonstrating superior performance compared to conventional techniques. Despite challenges in model interpretability and hyperparameter tuning, the findings emphasize the efficacy of the proposed approach in enhancing precision and adaptability in video anomaly detection. The implications of this research extend to diverse applications in security and surveillance. Future work may explore ensemble methods, attention mechanisms, and transfer learning for further refinement and generalizability of the LSTM-CAE framework. Overall, this study contributes to the advancement of video anomaly detection methodologies, providing a foundation for continued exploration and application of deep learning approaches in this domain.