Implementation of Abnormal Event Detection using Automated Surveillance System
M. Vaishnavi, Jasmine Sowmya, Matta Yaswanth, P. R. Maruvarasi · 2023
In military zones, banks, streets, hospitals, borders, airports, and schools, surveillance systems are widely employed Security personnel examine the footage from check- in rooms to look for unusual humanoid activity. This research study presents a new surveillance system that uses machine learning to detect human activity from a large number of cameras. The Internet of Things (IoT) technology is used to capture images from cameras in both inside and outside surroundings. One of the main issues with object detection that has a significant impact on performance is unevenness issues. This study addresses an issue that results from fault detection. In order to tackle complex problems, a detection approach that identifies strange patterns are implemented Particularly efficient techniques in this area include those based on autoencoders and the Generative Adversarial Network (GAN). Firstly, this study suggests a GAN-based detection model that uses an auto-encoder as the originator and two isolated discriminators for each usual and irregular input; next, two new loss functions, Patch loss and Anomaly adversarial loss are considered, and additional compound to train the model together. Using this technique, the proposed model can be effectively optimized This study assesses the performance of the proposed model against conservative standard datasets, such as the MNIST, Fashion MNIST, and CIFAR 10/100 records, as well as against a real-world industrial dataset and the defects in smartphone cases. Finally, experimental results show that the proposed approach surpasses cutting-edge techniques in conditions of the average area under the ROC curve, demonstrating the effectiveness of the proposed approach (AUROC).