Massive Crowd Abnormal Behaviors Recognition Using C3D

Recognizing abnormal activity in a massive crowd can be difficult due to various camera angles, large crowds, and other challenges. Existing methods are frequently investigated using small crowd datasets, clear surveillance video, and simple anomalous behavior. Additionally, proposed solutions are based on complex frameworks that failed to deliver satisfactory results for the large-scale crowd. This research examined methods to enhance identifying abnormal behavior using the large crowd dataset from the Hajj. Large-scale crowds and various anomalous behavior examples can be found in this dataset. This study experiments with different deep learning models to categorize videos from the Hajj dataset into the corresponding class of abnormal behavior. The precision, accuracy, and F1-score of the C3D model on the test set are all outstanding, at 0.90, 0.89, and 0.88 respectively. The model enhances the recognition of abnormal behaviors as compared to previous studies that also employed the Hajj dataset.