Search

Current automatic segment extraction techniques for identifying target characters in videos have several limitations, including low accuracy, slow processing speeds, and poor adaptability to diverse scenes. This paper introduces an optimized algorithm to address these issues that enhance the RetinaFace and FaceNet models. We selected RetinaFace for face detection, employing MobileNetV1-0.25 as the backbone network and simplifying its Feature Pyramid Network (FPN) structure to boost detection speeds. Analysis of 460 images with a 720P resolution demonstrated an average speed improvement of 20.6%. For face recognition, we utilized FaceNet with MobileNetV3 as the backbone, augmenting its feature extraction capability by integrating four Receptive Field Block (RFB) structures and replacing the Squeeze-and-Excitation (SE) module with the Convolutional Block Attention Module (CBAM). Experimental results indicate that our enhancements elevate the maximum accuracy to 97%, outperforming the original model. Additionally, we integrated these refined algorithms and conducted disintegration experiments on segment extraction in 10 videos, evaluating various metrics. The findings show improvements in both precision and recall. We also compared our algorithm against the Dlib model; our system achieved an overall interception accuracy of 79.94%, surpassing Dlib's 75.55%. This confirms the enhanced performance and feasibility of our proposed algorithm. [ABSTRACT FROM AUTHOR]

Intelligent farming has been an ever-increasing trend in agricultural production, with the development of artificial intelligence (AI) and Internet of Things (IoT). Rapid and accurate identification of cattle identity is of great significance to prevent the insurance fraud for the live cattle loans in the cattle industry. Among them, computer vision can be expected for the cattle face recognition in the modernization transformation of the livestock industry. Smart devices and systems can also be integrated to achieve the intelligent cattle management, feeding, and disease prevention. However, the traditional identification (such as ear tags and collars) has limited the large-scale production in recent years, due to the small differences in facial features among different cattle, the deep layers of the FaceNet network, slow inference speeds, and insufficient classification accuracy. In this study, a cattle face recognition was proposed using FaceNet, called VanillaFaceNet. Firstly, the backbone feature extraction network of FaceNet was replaced with the latest simplified network. VanillaNet-13. Dynamic activation and enhanced linear transformation of activation functions were proposed to improve the non-linearity of the network. Specifically, dynamic activation was fully utilized the expressive power of activation functions during training when dynamically adjusting, in order to flexibly adapt the variations in data distribution at different stages of training. Dynamic activation was used to merge the convolutional layers during inference phase. The computational load was reduced to improve the inference speed of networks. The performance and efficiency of model were then enhanced during training and inference. Activation functions with linear transformations were significantly enhanced the non-linearity through parallel stacking. Multiple activation functions were stacked in parallel, thus enabling each layer to capture more complex features. Additionally, spatial context information was embedded within the activation functions. The spatial relationships among features were better utilized to fit the complex feature distributions. Non-linearity and integration of spatial context information were achieved in a more accurate and efficient model when processing complex data. Secondly, DBCA (Dual-Branch Coordinate Attention) module was added into the global maximum pooling. Global average pooling was used to aggregate significant features of cattle faces, in order better represent the differences among cattle facial features. Therefore, the accuracy network was improved to recognize the cattle. Finally, a center loss was introduced to train the network with the center-triplet loss joint supervision, because the triplet loss was only reduced the inter-class differences among cattle. The intra-class separability of cattle was improved to compactly aggregate the same category of cattle. Thus, the accuracy of comparisons was improved among the same identities of cattle. Cattle face videos were collected at the Otai Ranch in Hohhot, Inner Mongolia Autonomous Region. An image dataset was constructed to train and test the model for the cattle face recognition. The experimental results show that VanillaFaceNet was achieved an accuracy of 88.21% in the cattle recognition, with a frame rate of 26.23 frames per second (FPS). Compared with FaceNet, MobileFaceNet, CenterFace, CosFace, and ArcFace, the model was improved the recognition accuracy by 2.99%, 9.58%, 6.26%, 3.85%, and 4.49%, respectively, and the inference speed by 2.67, 0.77, 0.10, 1.28, and 0.94 frames/s, respectively. The recognition accuracy and speed were greatly improved to fully meet the requirements of the ranch for the accuracy and real-time performance of cattle recognition. The excellent performance was achieved in the cattle recognition, suitable for the deployment on embedded devices, such as Jetson AGX Xavier. A better balance was also gained between accuracy and inference speed of cattle facial recognition. [ABSTRACT FROM AUTHOR]

The increased criminal activity associated with unauthorized entry into facilities has become a global concern. Traditional mechanical locks suffer from drawbacks such as key loss, theft, duplication risks, and time-consuming operation. Therefore, biometrics has been explored as a key to accessing a restricted area. However, some challenges still need to be solved in developing such systems, including user registration, response speed, maintainability, and the ability to distinguish between real and fake individuals. This paper proposes and develops a biometric lock system (BLS) whose opening is performed by recognizing a person’s face. It solves the challenges of re-training, antispoofing, real-time response, and works inside an embedding system. The proposed BLS overcomes these challenges using a pre-trained network called FaceNet for feature extraction and coding into 128-dimensional vectors. We use the characteristic vector provided by FaceNet and a cosine distance to recognize the persons. It also incorporates ResNet18 + remote photoplethysmography (rPPG) to avoid spoofing. The architecture was implemented in a BLS, demonstrating an impressive false acceptance rate of 0% under varying lighting conditions, with an average response time of 1.68 seconds from facial detection to door opening. The BLS has easy maintainable devices, providing enhanced security by accurately identifying individuals and preventing unauthorized access. The system can distinguish between real and fake people without using specialized hardware. Making it a versatile solution suitable for homes, offices, and commercial spaces. The results underscore the potential efficacy of our proposed BLS in mitigating security challenges related to unwarranted access to restricted facilities. [ABSTRACT FROM AUTHOR]

IoT applications revolutionize industries by enhancing operations, enabling data-driven decisions, and fostering innovation. This study explores the growing potential of IoT-based facial recognition for mobile devices, a technology rapidly advancing within the interconnected IoT landscape. The investigation proposes a framework called IoT-MFaceNet (Internet-of-Things-based face recognition using MobileNetV2 and FaceNet deep-learning) utilizing pre-existing deep-learning methods, employing the MobileNetV2 and FaceNet algorithms on both ImageNet and FaceNet databases. Additionally, an in-house database is compiled, capturing data from 50 individuals via a web camera and 10 subjects through a smartphone camera. Pre-processing of the in-house database involves face detection using OpenCV's Haar Cascade, Dlib's CNN Face Detector, and Mediapipe's Face. The resulting system demonstrates high accuracy in real-time and operates efficiently on low-powered devices like the Raspberry Pi 400. The evaluation involves the use of the multilayer perceptron (MLP) and support vector machine (SVM) classifiers. The system primarily functions as a closed set identification system within a computer engineering department at the College of Engineering, Mustansiriyah University, Iraq, allowing access exclusively to department staff for the department rapporteur room. The proposed system undergoes successful testing, achieving a maximum accuracy rate of 99.976%. [ABSTRACT FROM AUTHOR]

These days, a significant portion of the solutions for vehicle Internet of things applications come from wireless sensor networks. This article uses cameras, radio-frequency identification, and ultrasonic sensors to address typical issues with vehicle technology, such as unlawful vehicle use inside a community, vehicle thefts, and vehicle accidents. It also addresses the issue of identifying vehicle pollution parameter values like carbon monoxide (CO) and carbon dioxide ( CO 2 ), providing information about the driver's alcohol consumption, and verifying the driver's eligibility (driving license). The driving license will be used to identify the driver. Deep learning algorithms, such as Multi-Task Cascaded Convolutional Neural Networks and facenet algorithms, can identify driving licenses. The proposed algorithm has an 92% accuracy rate in detecting the driver's face. The proposed system is installed and demonstrated using Micro-controller, Micro-processor and other sensors in real time environment. The River Formation Dynamics based Multi-hop Routing Protocol for Vehicles (RFDMRPV) is used for communication between vehicles. Data collected from the sensors mounted in vehicles are communicated to server utilizing RFDMRPV for storing. Alert the driver, owner of the vehicle and other authorities depending on the acquired sensor results. [ABSTRACT FROM AUTHOR]

In this study, we propose a vision system for personnel recognition and tracking in a multiple-persons space. The personnel tracking function can be implemented in service vehicles or medical systems to identify specific users, thus facilitating service tasks such as follow-up care or support. User identification and tracking processes become highly challenging in environments that contain multiple people who are not wearing any external sensors. The proposed system involves three steps. First, the You Only Look Once version 4 (YOLOv4)-tiny model of object detection is used to extract a personnel-bounding box from an image. Second, the image coordinates and camera coordinates are converted into three-dimensional (3D) space coordinates on the basis of a depth map derived from a binocular camera to obtain 3D spatial information on the personnel. Finally, the unscented Kalman filter (UKF), FaceNet, or a combination of the UKF and FaceNet (UKF--FaceNet) produced through decision tree fusion, is used to identify or track the target personnel. Experimental results indicated that the proposed system can successfully track personnel. The UKF--FaceNet method can effectively improve the drawbacks of UKF- and FaceNet-based tracking. The UKF--FaceNet method can be adopted to achieve accurate and stable personnel identification in a multiple-persons environment. This method can reidentify and retrack a target user accurately when the user is obscured. [ABSTRACT FROM AUTHOR]

Smart classroom enables better learning experience to the students and aid towards efficient campus' management. Many studies have shown positive correlation between attendance and student's performance, where the higher the attendance, the better the student's performance. Therefore, many higher learning institutions make class attendance compulsory and students' attendances are recorded. Technological solutions for an advanced attendance system such as face recognition are highly desirable. Face recognition offers an automated, smart, and authentic solution in recording student attendance. In this work, an artificial intelligence-based face recognition system is used for attendance recording system. The recognized face is used to confirm the presence of a student to the class. Six pretrained face recognition models are evaluated for the adoption in the system developed. FaceNet is adopted in this work with accuracy of more than 95%. The automation system is supported by IoT. [ABSTRACT FROM AUTHOR]

Identifying faces in non-frontal poses presents a significant challenge for face recognition (FR) systems. In this study, we delved into the impact of yaw pose variations on these systems and devised a robust method for detecting faces across a wide range of angles from 0° to ±90°. We initially selected the most suitable feature vector size by integrating the Dlib, FaceNet (Inception-v2), and "Support Vector Machines (SVM)" + "K-nearest neighbors (KNN)" algorithms. To train and evaluate this feature vector, we used two datasets: the "Labeled Faces in the Wild (LFW)" benchmark data and the "Robust Shape-Based FR System (RSBFRS)" real-time data, which contained face images with varying yaw poses. After selecting the best feature vector, we developed a real-time FR system to handle yaw poses. The proposed FaceNet architecture achieved recognition accuracies of 99.7% and 99.8% for the LFW and RSBFRS datasets, respectively, with 128 feature vector dimensions and minimum Euclidean distance thresholds of 0.06 and 0.12. The FaceNet + SVM and FaceNet + KNN classifiers achieved classification accuracies of 99.26% and 99.44%, respectively. The 128-dimensional embedding vector showed the highest recognition rate among all dimensions. These results demonstrate the effectiveness of our proposed approach in enhancing FR accuracy, particularly in real-world scenarios with varying yaw poses. [ABSTRACT FROM AUTHOR]

Recently, the increase in the emergence of fake videos that have a high degree of accuracy makes it difficult to distinguish from real ones. This is due to the rapid development of deep-learning techniques, especially Generative Adversarial Networks (GAN). The harmful nature of deepfakes urges immediate action to improve the detection of such videos. In this work, we proposed a new model to detect deepfakes based on a hybrid approach for feature extraction by using 128-identity features obtained from facenet_CNN combined with most powerful 10-PCA features. All these features are extracted from cropped faces of 10 frames for each video. FaceForensics++ (FF++) dataset was used to train and test the model, which gave a maximum test accuracy of 0.83, precision of 0.824 and recall value of 0.849.

IoT applications revolutionize industries by enhancing operations, enabling data-driven decisions, and fostering innovation. This study explores the growing potential of IoT-based facial recognition for mobile devices, a technology rapidly advancing within the interconnected IoT landscape. The investigation proposes a framework called IoT-MFaceNet (Internet-of-Things-based face recognition using MobileNetV2 and FaceNet deep-learning) utilizing pre-existing deep-learning methods, employing the MobileNetV2 and FaceNet algorithms on both ImageNet and FaceNet databases. Additionally, an in-house database is compiled, capturing data from 50 individuals via a web camera and 10 subjects through a smartphone camera. Pre-processing of the in-house database involves face detection using OpenCV’s Haar Cascade, Dlib’s CNN Face Detector, and Mediapipe’s Face. The resulting system demonstrates high accuracy in real-time and operates efficiently on low-powered devices like the Raspberry Pi 400. The evaluation involves the use of the multilayer perceptron (MLP) and support vector machine (SVM) classifiers. The system primarily functions as a closed set identification system within a computer engineering department at the College of Engineering, Mustansiriyah University, Iraq, allowing access exclusively to department staff for the department rapporteur room. The proposed system undergoes successful testing, achieving a maximum accuracy rate of 99.976%.

Employees use their level of attendance or absence to demonstrate their presence at work or absence from it in an agency. This absence is connected to how discipline is applied, which is decided by each organization or institution. It can be inferred from this that student absenteeism in a setting where there is activity serves to increase discipline and demonstrate attendance. With IP Camera technology, it can be applied to the attendance system using the MTCNN method as face detection and FaceNet to extract high-quality features from the face. The system created can detect faces at a distance of 40 cm – 180 cm with an accuracy of 90.5% and can detect more than 1 object in 1 frame so that the IP Camera function is in accordance with the design in real-time. Tested the object of 3 pairs of twin faces produces a maximum accuracy of 90% where the level of match between the faces and the data is appropriate.

The conventional methodology for recording student attendance, which heavily relies on manual data transcription, is prone to inefficiencies and errors. Consequently, the development of an automated attendance management system has emerged as a critical need for efficient and accurate maintenance of attendance records. This study presents the design and implementation of an automated attendance management system, exploiting face recognition technology for identifying students within a class setting. A unique dataset was curated, consisting of 3900 facial images, captured in five varying positions and under diverse lighting conditions. In the initial phase of the system's operation, images of students are captured via a mobile camera. Subsequently, the Haar Cascaded classifier is utilized for the detection of faces within these captured images, and the FaceNet network is employed to recognize the detected faces. In the subsequent phase, the system registers attendance by cross-referencing the recognized faces with the primary student record. An attendance sheet copy is then dispatched to the teacher. Upon evaluating the system's effectiveness, it was ascertained that the system successfully identifies students and registers their attendance with an impressive accuracy of 97.5%. It outperforms traditional systems in terms of workload reduction, error avoidance, speed, and accuracy. The proposed system holds potential for widespread deployment in institutes and schools for recording attendance and could be extended for employee attendance recording. By reducing human errors and the time required for attendance registration, and by swiftly generating electronic attendance lists, this system signifies a substantial improvement over conventional systems. [ABSTRACT FROM AUTHOR]

In the realm of surveillance, closed-circuit television (CCTV) cameras serve as a vigilant watch over unfamiliar entities. However, the unpredictability of such entities necessitates continuous human monitoring, an endeavor prone to error and demanding of significant resources. The automation of this process through face recognition could alleviate these burdens, provided the system delivers high precision and rapid judgment capabilities. This study presents a novel solution to these challenges: an automated human recognition and verification surveillance system, founded on a max-voting ensemble method. This innovative approach amalgamates five influential feature extraction models: VGGFace, FaceNet, FaceNet-512, Dlib, and Arcface, with a support vector machine deployed for classification. The proposed system was subjected to rigorous testing on the AT&T, faces94, Grimace, Georgia Tech, and FaceScrub datasets, demonstrating an impressive accuracy of 100% on the AT&T, faces94, and Grimace datasets, and 99.3% and 98% on the Georgia Tech and FaceScrub datasets, respectively. The system's performance was further enhanced through a re-verification technique, which facilitated swift and precise prediction of unknown entities in real time. This study thus contributes a significant advancement to the field of automated surveillance, offering a potent tool for efficient, accurate human recognition. [ABSTRACT FROM AUTHOR]

This research proposes a single network model architecture for mask face recognition using the FaceNet training method. Three pre-trained convolutional neural networks of different sizes are combined, namely InceptionResNetV2, InceptionV3, and MobileNetV2. The models are augmented by connecting an otherwise fully connected network with a SoftMax output layer. We combine triplet loss and categorical cross-entropy loss to optimize the training process. In addition, the learning rate of the optimizer is dynamically updated using the cosine annealing mechanism, which improves the convergence of the model during training. Mask face recognition (MFR) experimental results on a custom MASK600 dataset show that proposed InceptionResNetV2 and InceptionV3 use only 20 training epochs, and MobileNetV2 uses only 50 training epochs, but to achieve more than 93% accuracy than the previous works of MFR with annealing. In addition to reaching a practical level, it saves time for training models and effectively reduces energy costs. [ABSTRACT FROM AUTHOR]

In this study, we utilize blockchain to design an information transmission system for unmanned stores. It is aimed to create a network for exchanging information that does not rely on a third party, a so-called decentralized system. To achieve it, a peer-to-peer (P2P) network replaces the master-slave architecture and uses asymmetric encryption to identify the system users. To authenticate the identities of the stores' customers, we use the face recognition network FaceNet developed by Google. FaceNet has several advantages that make it suitable for performing identity authentication in an unmanned store, including its high accuracy and convenience. A database of face images is required to perform facial recognition, but to address privacy concerns, every pixel in all images in the database is encrypted using the Rivest-Sharmir-Adleman (RSA) algorithm. The system proposed in this paper has three identity endpoints: the host end, the recognition end, and the client end. Messages are transmitted through a P2P network, and a directed acyclic graph is used to achieve message broadcasting while avoiding infinite loops when sending and receiving messages. A sidechain is used to change the structure and consensus mechanism of a traditional blockchain so that they can apply to more scenarios, thereby increasing scalability. The simulation results are displayed via the user interface. [ABSTRACT FROM AUTHOR]

The paper presents an evaluation of a Pareto-optimized FaceNet model with data preprocessing techniques to improve the accuracy of face recognition in the era of mask-wearing. The COVID-19 pandemic has led to an increase in mask-wearing, which poses a challenge for face recognition systems. The proposed model uses Pareto optimization to balance accuracy and computation time, and data preprocessing techniques to address the issue of masked faces. The evaluation results demonstrate that the model achieves high accuracy on both masked and unmasked faces, outperforming existing models in the literature. The findings of this study have implications for improving the performance of face recognition systems in real-world scenarios where mask-wearing is prevalent. The results of this study show that the Pareto optimization allowed improving the overall accuracy over the 94% achieved by the original FaceNet variant, which also performed similarly to the ArcFace model during testing. Furthermore, a Pareto-optimized model no longer has a limitation of the model size and is much smaller and more efficient version than the original FaceNet and derivatives, helping to reduce its inference time and making it more practical for use in real-life applications. [ABSTRACT FROM AUTHOR]

Educators have problems conducting online learning, such as monitoring student attendance while presenting the material. This paper aims to predict student names who attend zoom video conferences with various lighting conditions and face angles by comparing two detection and two recognition methods. This paper proposes an intelligent system based on the use of a bot that will analyse a combination of face detection and recognition method for attend-ance systems using video conferencing applications to carry out online learning. The proposed system will use the best combination of two methods to recapitu-late student attendance. The face detection system uses Haar Cascade and Multi-Task Cascaded Convolutional Neural Network (MTCNN), and the face recogni-tion system uses ResNet and FaceNet. The tests were conducted on video zoom footage taken during online lectures. The results show that MTCNN and FaceNet get the highest accuracy, 93.23%. [ABSTRACT FROM AUTHOR]

Facial recognition is a highly developed method of determining a person's identity just by looking at an image of their face, and it has been used in a wide range of contexts. However, facial recognition models of previous researchers typically have trouble identifying faces behind masks, glasses, or other obstructions. Therefore, this paper aims to efficiently recognise faces obscured with masks and glasses. This research therefore proposes a method to solve the issue of partially obscured faces in facial recognition. The collected datasets for this study include CelebA, MFR2, WiderFace, LFW, and MegaFace Challenge datasets; all of these contain photos of occluded faces. This paper analyses masked facial images using multi-task cascaded convolutional neural networks (MTCNN). FaceNet adds more embeddings and verifications to face recognition. Support vector classification (SVC) labels the datasets to produce a reliable prediction probability. This study achieved around 99.50% accuracy for the training set and 95% for the testing set. This model recognizes partially obscured digital camera faces using the same datasets. We compare our results to comparable dataset studies to show how our method is more effective and accurate. [ABSTRACT FROM AUTHOR]

Facial recognition is a part of Computer Vision that is used to get facial coordinates from an image. Many algorithms have been developed to support Facial Detection such as Cascade Face Detection using Haar-Like features and AdaBoost to classify its Cascade and Convolutional Neural Network (CNN). Face recognition in this study uses the Deep Convolutional Neural Network (DCNN) method, and the output of this method is the measurement value of the face. In the model training process, Triplet Loss from Triplet Network Deep Metric Learning is used to get good face grouping results. The value of this face measurement will then be measured using the Euclidean distance calculation to determine the similarity of the input face from the dataset. This Research is using 6 images of Government officers in Indonesia to determine the accuracy of the model when there is a new picture of these officers inputted into the training machine. The result provides a 94% accuracy level with a variety of face positions and levels of brightness.

Epidemics are severe around the world, and people must wear face masks and authenticate in and out of public places to reduce the risk of virus transmission. However, wearing a face mask cannot detect facial features and affect the accuracy of the recognition system, so limitations need to be overcome. Considering that identity recognition does not rely solely on facial information, we use KINECT to get a wider range of skeleton information for judgment. At the same time, it is proposed to replace the diffractive optical elements of the KINECT sensor with a holographic optical element. The problem of uneven intensity and distortion of the light field can be improved to improve the accuracy of skeleton detection. Next, clear scattering information and images are provided to MTCNN and FaceNet models for whole-body imaging identification and pre-training. Besides, we only keep the skeleton features to represent the standing posture and ignore the difference in appearance, which can reduce the changing appearance features. The experimental results show that the proposed method and model with the improved KINECT can solve the problem that is difficult to identify with the mask, and its accuracy is 99.5%. [ABSTRACT FROM AUTHOR]

Traditional attendance recording methods can be replaced with artificial intelligence (AI) methods. This paper aims to develop an automated attendance system with a user-friendly graphical interface in Arabic to enhance user interaction. The Python programming language is used to build the system, which employs object detection and feature extraction algorithms. This system processes a live broadcast from an IP camera installed in the classroom. The system employs object detection algorithms to detect the faces in the video. The face locations are then sent to the feature extraction algorithms to extract and compare features with the features stored in the database. Our paper concludes that automating the attendance registration process using facial recognition technology can achieve up to 100% accuracy. Comparing our proposed system with previous studies, we find that it outperforms them in several aspects. Our method handles real-time tests with a larger number of students. It accommodates students positioned at different distances from the camera within the classroom, demonstrating our method's effectiveness in automating the attendance registration process.

Face recognition (FR) has matured with deep learning, but due to the COVID-19 epidemic, people need to wear masks outside to reduce the risk of infection, making FR a challenge. This study uses the FaceNet approach combined with transfer learning using three different sizes of validated CNN architectures: InceptionResNetV2, InceptionV3, and MobileNetV2. With the addition of the cosine annealing (CA) mechanism, the optimizer can automatically adjust the learning rate (LR) during the model training process to improve the efficiency of the model in finding the best solution in the global domain. The mask face recognition (MFR) method is accomplished without increasing the computational complexity using existing methods. Experimentally, the three models of different sizes using the CA mechanism have a better performance than the fixed LR, step and exponential methods. The accuracy of the three models of different sizes using the CA mechanism can reach a practical level at about 93%. [ABSTRACT FROM AUTHOR]