Multi-Person Key Points Detection for Abnormal Human Behavior Analysis Using The ConvLSTM-AE Method

Abstract

The detection of abnormal human behavior is an interesting issue to consider in computer vision. The problem of detecting abnormal activities carried out in this study is a problem formulated in the process of monitoring human activities. In understanding the nature of human activities, a system is needed that is applied to data training which is specifically proposed based on ConvLSTM-AE for the detection of moving objects in motion-based events. Detection of anomalies and localization that arise due to camera jitter and movement of objects in the area. With the unusual motion of the video frame, connected component analysis is exploited to provide bounding boxing to find out whether human movements whose activities are based on human poses will be classified as normal or abnormal in a set of activity data. The learning and training process so that the proposed model can capture different duration of time and more optimal processes.