การตรวจจับพฤติกรรมผิดปกติและระบุตำแหน่งของกลุ่มบุคคลโดยใช้เทคนิคการวิเคราะห์ภาพวิดีโอ

Abstract

Abnormal detection in crowded scene is an important issue in computer vision and video surveillance systems. Many researches have studied and tried to define the phenomena of crowd behavior. In complex situations, however, ambiguities usually arise due to chaotic movements in the scene, especially when it is solved based upon motion trajectory analysis involving object segmentation and tracking. In this research we present a Momentum Force Model (MFM) to define the interaction among people. This model consists of two models, which are Low-Level Momentum Force Model (MF,) and High-Level Momentum Force Model (MF). MF, is the model of low-level feature based on dense optical flow and its interactions are defined by a force inspiring the energy propagation phenomena that depend on directions and velocities. MF, is modelled based on MF, and include three principal forces: group motion force (GMF), interaction motion force (IMF), and reciprocal motion force (RMF). MFM model can detect three kinds of group abnormal behaviors. First, abnormal events in crowd are detected using a thresholding method. Our method is evaluated with the well-known UMN dataset. Second, we can detect fighting by applying the MFM with fighting factor to define the fighting force. The fighting event is detected by using the threshold method. Furthermore, MFM can recognize six group activities and we classify the six activities by using the neuron network method. We test the algorithm for fighting detection and group activity recognition with NUS-HGA and BEHAVE datasets. The best results of our algorithm when tested on UMN dataset and NUS-HGA dataset, show that the accuracy is 97.5%, while other results give an accuracy of more than 93%. The results of our approach technique reveal good efficiency with high accuracy without the tracking method, and are competitive with other state-of-the-art methods.