Human Recognition Techniques from Video Surveillance System

Abstract

Security-based technologies have advanced to the point that they are now extensively newlineemployed in a range of real-time applications. Using computer vision-based techniques, visual newlinesurveillance is a potential tool for detecting, monitoring, and identifying certain objects. In this newlinecontext, face recognition is a crucial component of a surveillance system. Several methods for newlineface recognition have been developed in the past, however contemporary methods are only newlineutilised on face data. newlineVideo face recognition methods, which provide more information to enhance the newlineprotection system, have recently been implemented. Thanks to deep learning that improves newlineface recognition accuracy. Current methods, on the other hand, need fixed-size images for newlineimage processing, and most methods are a single network for extracting features, limiting newlinemodel generalization. To overcome these issues, we develop an approach that begins with a newlinekalman filtering-based face detection and tracking method. We extract the combined features newlineof the input image after face recognition and save the learned data. newlineTo improve the learning process, the Bayesian learning approach is used. The next stage newlineis to demonstrate a visual face recognition method based on Convolutional Neural Networks. newlineWe use a context subtraction strategy in the proposed approach, which helps to minimize the newlinefeature extraction method by increasing scene complexity. To reduce the face recognition error, newlinewe use a bounding box regression model in this step. newlineFinally, an RCNN-based learning model is used to discriminate the groups of detected newlinefaces using Joint Bayesian learning. Background Subtracted Faster RCNN for video based face newlinerecognition in the proposed model based on these phases(BSF-RCNN-VFR). Finally, we newlinepresent a CNN-based scheme that improves CNN learning by combining extraction of features newlineand feature embedding modules with GoogleNet architecture. To increase the quality of video newlineframes, we used a histogram redistribution image enhancement technique newline