Abstract
Aiming at the problem of low performance of crowd abnormal behavior detection caused by complex backgrounds and occlusions, this paper proposes a single-image crowd counting and abnormal behavior detection via multiscale GAN network. The proposed method firstly designed an embedded GAN module with a multibranch generator and a regional discriminator to initially generate crowd-density maps; and then our proposed multiscale GAN module is added to further strengthen the generalization ability of the model, which can effectively improve the accuracy and robustness of the prediction detection and counting. On the basis of single-image crowd counting, synthetic optical-flow feature descriptor is adopted to obtain the crowd motion trajectory, and the classification of abnormal behavior is finally implemented. The simulation results show that the proposed algorithm can significantly improve the accuracy and robustness of crowd counting and abnormal behavior detection in real complex scenarios compared with the existing mainstream algorithms, which is suitable for engineering applications.
Highlights
With the expansion of urban scale and the increase of crowd, the probability of traffic accidents, congestion, stampede, and other emergencies in public place increases [1, 2]
While surveillance systems are cheap and common, the cost of hiring the right people to observe and analyze recorded videos is still very high [4]. erefore, realtime analysis of abnormal behavior in public places is important for intelligent surveillance system because abnormal behavior can be prevented and stopped only when the surveillance system has the ability of understanding human behavior [2, 5]
In order to improve the accuracy and robustness of crowd abnormal behavior detection algorithms, this paper proposes a crowd counting model based on multiscale network. e model can be regarded as an embedded GAN structure. e embedded GAN module learns crowd features and optimizes the local correlation of images. e scale module further extracts local multiscale features and generates the final crowd density image. e structure of the proposed model is shown in Figure 1, which consists of three parts: generation network, discrimination network, and scale module. e generation network and discrimination network are embedded in the whole model to construct an embedded GAN module, where the generation network is composed of partial structure of VGG-16
Summary
With the expansion of urban scale and the increase of crowd, the probability of traffic accidents, congestion, stampede, and other emergencies in public place increases [1, 2]. E clustering model (CM) has strong representation ability for highdensity crowd, but as the number of pedestrians in the group decreases, the accuracy of the behavioral consistency estimation decreases, which leads to the significant decrease of the representation ability To solve this problem, according to the consistent characteristics of particle behavior, literature [16] proposed a global direction descriptor to extract the overall motion of the group, and the local and global descriptors were fused to establish a direction-cluster model, which enhances the model’s ability to characterize crowd characteristics. In order to solve the above problems, some scholars [18] proposed a crowd abnormal behavior detection method based on synthetic optical-flow feature descriptor and trajectory in single image. As the performance of the deep learning model improves, the accuracy of pedestrian detection in the crowd is greatly improved. erefore, we can consider improving the abnormal behavior detection performance based on pedestrian detection and counting. e proposed method firstly designed an embedded GAN module with a multibranch generator and a regional discriminator to initially generate crowd density maps; and our proposed multiscale module is added to further strengthen the generalization ability of the model; synthetic optical-flow feature descriptor is adopted to obtain the crowd motion trajectory, and the classification of abnormal behavior is implemented
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
