A Real-time Framework for Abnormal Activity Detection in Crowd Videos using Comprehensive Feature Sets and Bag of Feature Classifiers

Crowd behavior analysis (CBA) is a crucial task in video surveillance due to the rising frequency of abnormal events in public places. This paper presents a real-time framework for detecting abnormal activities in crowd videos by leveraging optical flow-based motion heat maps, feature extraction, and classification models. The proposed method combines Support Vector Machines (SVM) and Bag of Features (BoF) classifiers to achieve high accuracy. Experiments conducted on the UMN dataset and real-time indoor datasets demonstrate that the model achieves an Area Under Curve (AUC) of 99.6% on standard datasets and 97% on real-time scenarios, outperforming several state-of-the-art methods.