Anomaly Detection in Crowded Scenes With Machine Learning Algorithms

Abstract

Camera systems have a very important place as they are widely used in providing security in crowded environments. The video images recorded by cameras are examined to check whether there is dangerous or unusual behavior. It is tried to develop appropriate measures according to the result of this control. Modeling human behaviors for the definition and detection of abnormal behavior has become a popular research area in recent years. This study was carried out by applying supervised learning algorithms, one of the machine learning methods, on five different scenes that in two open data sets. Normal and abnormal motion scenes were detected on the videos in the data sets. In these two data sets, abnormal motion was detected in a total of five different locations. Random Forest, Support Vector Machines and k Nearest Neighbor algorithms, which are among the supervised learning algorithms, were used in this process. The algorithms used were compared with performance criteria such as accuracy, sensitivity, precision and F1 score.