People from all walks of life have always been concerned about public safety issues. There has been a meteoric rise in the number of security cameras used for monitoring both public and private settings in recent years. The key to avoiding public safety hazards is now in spotting anomalous human behaviour on film, thanks to advances in video detection technologies. The ability to recognise anomalous human behaviour is crucial, especially in social settings like those found in student groups. Existing algorithms for detecting anomalous human behaviour tend to focus on outdoor activity detection, and their performance indoors is subpar at best. The majority of a student’s time is spent inside, and most modern classrooms have some form of surveillance technology. This investigation focuses on identifying anomalous actions taken by indoor humans, and it does so by employing a novel abnormal behaviour detection framework. In this project, we are detecting the human abnormal activity using tensorflow pose estimation model by estimating the location of key joints of the person. This model can estimate not only one person but also many persons in the videos. This method has better detection performance.