Video surveillance system ensures the security and safety of the public in dynamic environments. Detection of abnormal events and classifying their severity in crowded scenes gain more importance to increase public safety. Various existing approaches are designed to detect abnormal events, like snatching, road accidents, real-time explosions, fighting, and car fires, but increasing the classification accuracy still faces complex issues, such as computational complexity and flexibility issues. To overcome these issues and to provide higher classification accuracy for detecting the abnormal event and to classify the severity level, this research designs a proposed method named Multihead Criss Sparse Attention-enabled Mixed learning-based Convolutional Memory Network (MCSA-MCMN). The proposed model shows higher performance in the surveillance system through the extraction of optimal features. Meanwhile, the capability of the designed model is upgraded by the integration of the MCSA module, which strengthens the generalizability function gradually. Moreover, the model ensures the semantic similarities of objects for providing robust detection and classification results with minimal processing time and memory constraints. Under these implications, the model attained better potency with specific estimation measures, and that obtains a better value as 98.17% sensitivity, 97.9% specificity, and 98.10% accuracy for the Shanghai Tech dataset.
Ugile et al. (Fri,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: