An Efficient Deep Learning Based Approaches for Crime Activities classification in Surveillance Videos

Abstract

Video surveillance is mostly utilized in crowded areas to find and identify unusual activity in a complicated environment. Something out of the ordinary, or abnormal, is called an anomaly. Modeling and processing the results of the unusual situation might be a daunting and seemingly impossible task. Consequently, this study presents a deep-learning methodology for constructing a crime detection system. This advanced methodology encompasses many layers essential for feature extraction and classification, enabling the system to effectively and reliably identify criminal behaviors, the presented criminal activity classification models, where, two CNN methodologies employed (EfficientNet-B7 and ResNet50) trained and evaluated on the widely recognized UCF Crime datasets. The experiments uncover that the proposed system achieved outstanding results and outscored the other deep learning approaches with an accuracy of 99.48%, precision of 99.47%, sensitivity of 99.41%, and F1 score of 99.44% using the UCF crime dataset. Hence, this system is effectively capable of tracking criminals' trails and detecting crime events.