Anomaly Detection Network for Surveillance Video

Abstract

Anomaly Activity is the prediction of a suspicious activity from a picture or video. This project would include using neural networks to detect suspicious human activity from real-time CCTV video. Human Anomaly Activity is a central issue in computer vision that has been researched for over 15 years. It is significant due to the large number of applications that can benefit from Activity detection. Human pose estimation, for example, is used in applications such as video monitoring, animal tracking and behavior recognition, sign language identification, advanced human-computer interaction, and marker less motion recording. Low-cost depth sensors have disadvantages such as being restricted to indoor use, and their low resolution and noisy depth information make estimating human poses from depth images difficult. As a result, we want to use neural networks to solve these issues. The detection of suspicious human activity in surveillance video is a hot topic in image processing and computer vision science. Human activities in areas such as bus station, highways, and school so on can be monitored using visual surveillance to detect terrorism, robbery, accidents and illegal parking, vandalism, fighting, chain snatching, violence, and other suspicious activities. It is extremely difficult to continuously track public places; thus, intelligent surveillance footage is needed that is capable of real-time monitoring of human activities, classify them as normal or unusual, and generate an alarm. Photographs, rather than recordings, are used for the majority of the analysis. Furthermore, none of the papers reported attempt to use CNNs to detect suspicious activity.