Crowd Behavior Monitoring using Self-Adaptive Social Force Model
Crowd can be defined as a large number of people gathered closely together. The larger size of crowd results number of behavior either in group or individually. To prevent or minimize the effects of the abnormal behavior, it has to be monitored continuously. Crowd behavior monitoring is an important task in public places to ensure public safety and avoid any unwanted incidents. It has become a popular research among computer vision communities nowadays due to its needs by the authorities. The current method used is Social Force Model (SFM), which can describe the behavior of a crowd based on the interaction forces between individuals. However, some limitations in the previous works caused by its parameters make it fail to correctly classify the crowd behavior into normal or abnormal. Hence, some modification has been introduced to SFM theory in order to provide significant interaction force; which absolutely portrayed the behavior of the crowd. This work aims to develop a crowd behavior monitoring system using Self-Adaptive SFM. This algorithm is jointly used with Horn-Schunck optical flow as a motion detector for the input video. Instead of using any segmentation methods, the motion of particles in each frame is captured by particle advection method. This is done by advected the particles using the underlying flow vectors of each particle. The obtained new locations for all the particles are necessary in estimating the interaction force of each particle. The combination of psychological and physical parameters in Self-Adaptive SFM makes it more realistic and mimicked the dynamic motion of people in a crowd. The estimated interaction forces of each particle represent the behavior of the crowd, whether it is normal or abnormal. The experimental evaluations on challenging datasets shows that the proposed method achieves the better detection result and outperforms the other methods, optical flow and SFM; with the average accuracy of about 94%.
Copyright (c) 2019 UMP Press
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.