Trajectory extraction for abnormal behavior detection in public area

Abstract

Surveillance system to improve safety and security is a major demand for the management and control of public area. Crowd management and control system requires a situation recognition technique which can predict accidents and provide alarms to the monitoring personnel. In this paper, we propose an abnormal behavior detection technique by using trajectory extraction of moving objects in video. Abnormal behavior includes running persons. The proposed abnormal behavior detection system separates background and foreground using Gaussian mixture model. And then, foreground image is used to generate the trajectories of moving objects using a Kanade-Lucas-Tomasi algorithm of the optical flow method. In addition, noise removal step is added to improve the accuracy of the created trajectory. From the trajectory of moving objects information, such as length, pixel, coordinate and moving degree is extracted. As the result of the estimation of abnormal behavior, objects' behavior is configured and analyzed based on a priori specified scenarios, such as running persons. In the results, proposed system is able to detect the abnormal behavior in public area.