An Efficient Abnormal Event Detection System in Video Surveillance Using Deep Learning-Based Reconfigurable Autoencoder
Jyothi Honnegowda, Komala Mallikarjunaiah, Mallikarjunaswamy Srikantaswamy · Ingénierie des systèmes d information · 2024
Video surveillance captures a vast number of normal and abnormal events worldwide.During this process, some attenuation occurs.Due to this phenomenon, inaccurate labelling of abnormal events and normal events happens and is difficult to identify, and these large events are presented in a Gaussian distribution.In the Gaussian distribution, more events are normal, and very few are abnormal events.With this information, it is very difficult to identify the abnormal events.Hence, the proposed method utilises deep learning technology to construct the hidden layer representation of the normal sample in a Gaussian distribution using a reconfigurable encoder (RAE).For video surveillance, the proposed methods use a deep learning-based reconfigurable autoencoder that improves the accuracy of finding anomalies and the ability to adapt to different video inputs.It processes video streams in real-time, enabling swift identification and response to security threats, representing a significant improvement in surveillance capabilities.It effectively addresses this issue.Consulting datasets, such as the University of California, San Diego (UCSD) data sets and the Avenue dataset, achieve this activity.We have used the proposed method to detect normal and abnormal events based on the threshold.According to the simulation results, the proposed method shows better performance than the conventional method.Our proposed method achieved frame-level AUCs of 93.7% and 83.56% at an average speed of 575 frames per second, demonstrating its success and efficiency compared to conventional methods.