MDSV: Mobs Detection by Enhanced Fused Feature Base Deep Neural Network from Surveillance Camera

Identifying mobs in images has become a significant field of study and implementation as digital photography and computer vision progress. The automated detection and evaluation of crowded material in visual data has significant potential for improving public safety, security, and content control. Detecting events among mobs is challenging because of disturbances in the optical flow, challenges arising from differences in camera aperture, and issues associated with lighting and feature tracking. The present study introduces a new TFFDNN-Net model for detecting mobs in photos. The recordings captured by the surveillance camera are first transformed into sequences of frames and undergo pre-processing. The YOLOV3 method is used to identify items inside frames that have been pre-processed. Concurrently, the applied method is used on the pre-processed photos to estimate motion, guaranteeing a thorough information analysis. Subsequently, the features are retrieved using the fixed feature extraction phase, namely the independent component analysis, and then chosen using the DINGO algorithm. Next, the enhanced fused (EF) features are used in the Deep Neural Network (DNN) method to classify situations as either violent or non-violent. The research results indicate that the TFFDNN-Net, as designed, achieves a crowd detection accuracy of 98.71% when applied to photos. The DNN improves the total F1-Score by 9.97%, 6.15%, and 3.94% compared to Alex Net, ResNet, and Dense Net, respectively. The TFFDNN-Net improves the accuracy by 5.75%, 0.28%, and 28.13% compared to Inception-Resnet-V2, CNN, LSTM, and ResNet50, respectively.