Detection of violence using mosaicking and DFE- WLSRF: Deep feature extraction with weighted least square with random forest

The violence-related instances had surged recently in areas including footpaths, sports stadiums, remote roads, liquor stores and elevators that are tragically discovered only after some time. In exploring this issue, the complete video analysis model's potential to determine any violent acts from the sequence of video clips is evolved. However, the recent studies that work on the violent detection approach majorly focus on traditional hand-crafted features, less performance accuracy in violence detection and do not make entire utilization of deep learning research outcomes in computer vision. The proposed system is put forth a violence detection framework based on (CNN) Convolutional neural network with (LSTM) Long short-term memory feature extraction process and fine-tuned the image frame hyperparameter from extracted features using Random forest classifier updated with weight score through (WLS) Weight least square algorithm. The Model in prior subjected to the feature extraction phase and the image frames are segmented through the mosaicking pre-processing step, with a 30:20 enlargement ratio to image mosaics, aiding to generate time-consistent outcomes and algorithm's performance improvisation through minimizing search space. The integration of CNN and LSTM framework is applied to reduce the complexity of the extraction learning process, and the LSTM network in correlating feature value with past information, and retaining memory space. The dynamic weighing scheme is proposed with the WLS method and this weighted score is assigned to the most probable class in the decision tree. Such more similar parameters as hyperparameters were tuned through a random forest classifier, and it categorizes the outcomes as non-fight or fights clips dynamically. The comparative performance evaluation of the proposed framework (DFE-WLSRF), Deep feature extraction – Weighted least square random-forest classifier delineated the outperforming high accuracy results in comparison to other traditional violence detection approaches.