Design An Optimization Based Deep Learning’s Framework For Detecting Faces From Videos

Abstract

A surveillance system is a valuable tool for crime detection today. Additionally, Face detection is a programme that can track, identify, categorise, and authenticate human faces from footage that has been recorded using security cameras. But because the recorded surveillance footage is so big, it takes longer. The least accurate detection and high mistake rate tasks in face recognition are the most difficult.This research develops the Fusion based Convolution and Recurrent Network (FbCRN) framework for improving face recognition in order to address these problems. Multiple video frames are initially gathered and uploaded into the system. Then CNN is processed for preprocessing, face detection, and feature extraction. Golden Eagle Optimization (GEO) is updated to improve the efficiency of feature extraction. Following that, the RNN classification layer receives the extracted features in order to recognise the faces. In the RNN classification layer update Vulture Optimization (VO) to improve face recognition performance. Finally, improved face detection and recognition capabilities of the created model are verified against those of other currently used models in terms of detection accuracy, precision, sensitivity, specificity, F-measure, and error rate.