ANOMALY DETECTION WITH MACHINE LEARNING ALGORITHMS IN CROWDED SCENES IN UMN ANOMALY DATASET

In recent years, keeping security under control in crowded environments has been a common problem. Camera systems are used to ensure security in crowded environments. When the video images recorded by the cameras are examined, it is checked whether there is any dangerous and unusual movement in the environment and appropriate measures are developed. Human behavior must be modelled to detect normal and abnormal behaviors in crowded scenes. In this study, crowded scenes in three different environments in the UMN Anomaly Data Set were examined. Random Forest, Support Vector Machines and k Nearest Neighbour algorithms, which are one of the machine learning methods in these three different environments, are applied. As a result of algorithms applied, the abnormal behaviour (like escape) of people in a crowded scene has been detected. Performance criteria such as accuracy, sensitivity, precision and F1 score of these applied algorithms were calculated and compared.