Recognizing Abnormal Behavior in Heterogeneous Crowd using Two Stream CNN

Recognizing anomalous behavior in heterogeneous crowd is paramount as it assists with crowd management, emergency response, crowd surveillance, public safety and security, and event monitoring. Crowd behavior is inherently complex and dynamic, especially in heterogeneous crowds comprising individuals with diverse characteristics, behaviors, and interactions. Traditional methods often struggle to accurately detect abnormal behavior in such environments. In the context of abnormal behavior recognition, Convolutional neural network, in spite of a number of issues such occlusion, clutter, uneven item distribution, and non- uniform object scale, is a promising method for recognizing human actions in huge data, such as CCTV. Hence, this study aims to propose a Two-Stream CNN based Abnormal Classifier model with optical flow for identifying the abnormal behavior of people such as racing, tossing objects and loitering in Heterogeneous crowd. The proposed Model has the ability to classify and obtain the final behavior classification results. The experimental depicts that proposed method is marginally outperforms the existing methods.