Deep Learning Based Intelligent Surveillance in Shopping Malls and Retail Stores

Deep learning algorithms for recognising human activities have lately gained popularity. These algorithms can analyse large volumes of data to detect trends over time and anomalies that may indicate suspicious activities. Various methods have been developed to identify such activities, but observing persons in real time and detecting abnormalities remains an ongoing problem. Two approaches are proposed to solve the problem: spatio-temporal models employing models ConvLSTM, LRCN, and InceptionV3-GRU for recognising suspicious behaviours in a frame, and a spatial model YOLOv8 for tracking individual activities over subsequent frames. The datasets are gathered from the UCF crime dataset, Kaggle and Roboflow. A specific dataset consisting of three major activities: normal, shoplifting, and weapons is created. Of the models that have been trained and tested, YOLOv8 model produced the best results with a maximum accuracy of 86.7%.