Real-Time Surveillance Through Face Recognition Using HOG and Feedforward Neural Networks

Abstract

Real-time security requirements continue to increase due to the occurrence of various suspicious activities in open and closed environments. Day-today security threats may seriously affect everyone lives. Many techniques have been introduced in this regard, but still some issues remain unaddressed. The work presented in this paper provides video surveillance with improved accuracy and less computational complexity. The most significant part of the system consists of face localization, detection and recognition. The system obtains underlined facial data through a video dataset or from the real-time environment. Subsequently, face/foreground and background keyframes are extracted from at hand captured video data. Finally, extracted facial image data is compared with the facial images in the database. In case no match is found with the existing data, a security alarm or signal is generated, alerting security personnel to take action. The proposed system is more accurate, has better performance, and is low cost compared with existing systems. INDEX TERMS Face recognition, HOG features, feedforward backpropagation neural network, surveillance video, principal component analysis.

DRONET: Multi-Tasking Framework for Real-Time Industrial Facility Aerial Surveillance and Safety

“…Similarly, the ML approach fuses these traditional computer vision methods to achieve better object classification and detection results [32]. Examples of ML Approaches include Haar Wavelet features [33]; Haar-like features and motion information, Implicit Shape Models, Histogram of Oriented Gradients (HOG) [34], Covariance descriptor, and Extended Histogram of Gradients (ExHOG) [35]. The drawbacks of the ML approach include time-cost in image extraction, limited range of image detection, low resolution of extracted image feature, and scalar/static image detection, which makes it unfit for real-time drone detection.…”

Fast and accurate face recognition system using MORSCMs-LBP on embedded circuits

“… Awais et al (2019) also produced Face Recognition for Real-Time Surveillance. The images were taken from the video stream and then they were compared with the images stored in the database.…”

A high-performance, hardware-based deep learning system for disease diagnosis

“…It is important to note that deep learning and neural networks are not two separate ideas or techniques; any neural network that has two or more layers is considered ‘deep’. Neural networks find applications in stock market prediction, agriculture, medical sciences, document recognition, and facial recognition, among others ( Awais et al, 2019 ; Nti, Adekoya & Weyori, 2021 ; Zhou et al, 2019 ; Guan, 2019 ; Chen et al, 2020 ; Kim et al, 2020 ; Lammie et al, 2019 ). The process of learning is usually carried out using ‘backpropagation’, a supervised learning technique in which the parameters of a neural network are adjusted according to a predefined error function.…”