In recent years, malignant crowd incidents that endanger public safety occur frequently, which alarms humans of the security in public areas. Focus on the detection of abnormal crowd behavior has been heightened in the field of computer vision. To solve this problem, the predictive neural network is innovatively applied to crowd anomaly detection by enlarging the differences between predictive frames and real frames in moving object regions. Firstly, aiming at the drawback of slow convergence speed of running Gaussian average model for object detection, an improved attenuation strategy based on learning rate is proposed. Then the processed foreground data set is put into the predictive neural network, and we define the degree of anomaly as the differences between predictive frames and real frames. By adjusting the threshold adaptively, the abnormal behavior in the crowd can be judged. The experimental results show that, compared with optical flow method and social force model, this method is not only fast in calculation, but also can detect crowd anomalies accurately and effectively. The accuracy rate of this method in UMN data set is up to 97.7%.
Software And Hardware
• Hardware: Processor: i3 ,i5 RAM: 4GB Hard disk: 16 GB • Software: operating System : Windws2000/XP/7/8/10 Anaconda,jupyter,spyder,flask Frontend :-python Backend:- MYSQL