Anomaly Detection in Surveillance Video Using Pose Estimation

Abstract

In this paper, we focus on key points detection of a person. We judge the abnormal behavior of the person by detecting the motion of key points of that person. In the starting frame, we select a bounding box. If some numbers of key points are moved out of the bounding box, then the system gives an alert message. So here we will take you through a new approach for detecting an anomaly in the surveillance video. The surveillance in these days has only capable of saving the videos on a hard disk. These are not yet the best solution. The crime record shows that the crimes are recorded. How ever, the criminals escape from the punishment due to mishandling of the recorded videos. The project aims to store only the anomalous behavior, by storing the picture with date and time. This kind of alternative not only decreases the disk space but also provides intelligence to the system. The techniques used to achieve this project are comparable to state of art techniques in the computer vision domain.