When to Use tablet thermal face recognition camera？

The technology for deep learning in the field of thermal infrared face recognition has recently become more available for use in research, therefore allowing for the many groups working on this subject to achieve many novel findings. Thermal infrared face recognition helps recognize faces that are not able to be recognized in visible light and can additionally recognize facial blood vessel structure. Previous research regarding temperature variations, mathematical formulas, wave types, and methods in thermal infrared face recognition is reviewed.

Sihovision Product Page

The human body gives off heat, which makes it possible to have a usable contrast between the body and its environment. The range of body temperature varies from 96-100 degrees Fahrenheit (F) while skin temperature in ambient room temperature varies from 79-83 F. The periorbital regions are some of the hottest areas in the human face, which allows them to be used as a feature for face tracking. TIR face recognition does, however, have difficulties in determining the eyes&#; position.

Thermal images show great usability for facial recognition since their sensitivity to illumination changes is low. Thermal infrared (TIR) imagery of faces is nearly invariant to changes in ambient illumination [ 1 ]. Infrared image facial recognition and the required equipment has not gathered so much interest in the past, due to their cost being higher than visible video equipment, their lower image resolution, lack of datasets and high image noise. However, the advancement of infrared (IR) technology has decreased these problems. The cost of thermal cameras at the retail level has decreased considerably, making them more widely available to consumers.

Review

Guzman et al. proposed a TIR framework for face recognition that extracts unique features and finds similarity in thermal images [1]. The framework&#;s protocol consisted of taking pictures of the person at four different times to allow for vascular changes over time that could affect their matching. Pavlidis et al. outlined a novel approach to the problem of face recognition in TIR [3]. The main ideas of their approach consisted of a Bayesian face detector method, followed by a physiological feature extractor. The face detector mainly uses the bimodal temperature distribution of human skin and typical indoor backgrounds. Mallat et al. discussed a novel solution based on cascaded refinement networks, which was able to generate high-quality color visible images, trained on a limited size database [4]. Their network is based on the use of contextual loss functions, enabling it to be inherently scale and rotation invariant. Gyaourova found that IR and visible imagery fusion in the wavelet domain demonstrated better recognition performance overall [5].

Figure 2 shows samples that were taken with thermal sensors in total darkness. Poor or absent illumination (shown in the right column) posed no impact on the images generated. Synthesizing images with informative facial attributes that are not in the visible spectrum was achieved [4].

A common triangulation of both reference shape and detected facial landmarks makes it possible to compute a piecewise affine transformation to each triangle and apply a set of transformations to transform a face from an arbitrary position in the image into a well-defined coordinate system. This makes it possible to use fixed regions of interest (ROIs) for image analysis, even for moving faces, so additionally increasing the number of algorithms that can be applied to images with unconstrained movement. See Figure 3.

Figure 3. Face frontalization.

Note how the regions of interest (ROIs) move with the face in the original video feed but remain in a fixed position in the frontalized view. Severe out-of-plane rotation may distort the frontalized image, however, as the image shows, the amount of acceptable rotation still covers most usual head poses.

Source: [6]

Carrapico et al. created results that showed face recognition, which could reach accuracy levels of 91% with localized binary pattern (LBP) [7]. Gabor and LBP are two well-known methods in face image analysis. The Color and Edge Directivity Descriptor (CEDD) and the Fuzzy Color and Texture Histogram (FCTH) are alternative methods that have shown good results in medical imaging. Two-dimensional Gabor filters are Gaussians weighted by an exponentially decaying sinusoid that can be applied at a given orientation and scale [8]. These filters are particularly interesting because of their behavior, resembling cells in the primary visual cortex. Selinger and Socolinsky focused their attention on longwave infrared (LWIR) imagery, in the spectral range of 8µ-12µ [9]. They studied a variety of methods in compensating for variation in illumination in order to boost recognition performance, including histogram equalization, Laplacian transforms, Gabor transforms, logarithmic transforms and 3-D shape-based methods. It is well-known that under the assumption of Lambertian reflection, the set of images of a given face acquired under all possible illumination conditions is a subspace of the vector space of images of fixed dimensions [10]. However, the set of LWIR images of a face under all possible imaging conditions is contained in a bounded set. Selinger and Socolinsky used a newly developed sensor capable of capturing simultaneous coregistered video sequences with a visible light charge-coupled device (CCD) array and LWIR microbolometer [9]. Radiometric calibration was found to provide non-uniformity correction. Socolinsky et al. saw that statistically significant evidence was presented, indicating that appearance-based face recognition algorithms applied to TIR, particularly LWIR imaging, have consistently better performance than when applied to visible imagery [11]. They found that the emissivity of the imaged object provides the further advantage of data where environmental factors contribute to a much lesser degree to within-class variability.

The algorithms were parametrized using different methods to analyze the impact of algorithm modifications on tracking performance. A high-quality Active Appearance Model (AAM) parametrized for maximal precision and a high-speed AAM using a diagonal of 70 pixels were used. A Deep Alignment Network (DAN) was trained by following the results. With the trained algorithm, two different frame update strategies were implemented, which were an instance that is updated with the bounding box of the detected face (bounds-DAN) and a version that uses the detected landmark points directly for the shape update (shape-DAN). A ShapeNet was also used. The network with the database was trained, and two update strategies were evaluated, which were an instance that updates both bounding box size and position with each frame (dynamic ShapeNet) and a version that keeps a constant bounding box size and updates the face position only (fixed ShapeNet). See Figure 4.

Figure 4. Qualitative overview of fitting performance.

From left to right: Original image, fast AAM version, dynamic ShapeNet, fixed ShapeNet, high-quality AAM, boundary DAN, shape-DAN, and manual ground truth. Best viewed electronically

Source: [6]

AAM: Active Appearance Model; DAN: Deep Alignment Network

Socolinsky et al. also found that several approaches extract thermal contours and match the shapes for identification [11]. These techniques use shape matching and the Eigenface method, which shows better results with thermal images than with visible spectrum images. They performed a comparison of recognition performance between visible and LWIR imagery based on two standard appearance-based algorithms: Eigenfaces, which are sets of eigenvectors, derived from the covariance matrix of the probability distribution over the high-dimensional vector space of face images, and Architecture for the Recognition of threats to mobile assets using Networks of multiple Affordable sensors (ARENA). Constructive Auto-associative Neural Network (CANet), which is a novel neural network inspired by neural biology, has been proposed. It is a constructive auto-associative neural network that outputs an approximation of input images using a dynamic architecture. This neural network uses receptive fields for implicit feature extraction, lateral inhibition, and auto-associative memory for image reconstruction.

Haar-based features, Constructive Auto-associative Neural Network (CANet)) and Texture Histogram offer the best results for some facial expressions [12]. Facial expression recognition using IR images has also been explored, and Trujillo proposed a facial expression recognition feature extraction model for images [13]. Principal Component Analysis (PCA) techniques, also known as Karhunen-Loeve methods, choose a linear projection that reduces the dimensionality while maximizing the scatter of all projected samples [7]. Local Feature Analysis constructs a family of feature detectors based on PCA decomposition, which is locally correlated. Haar-like transforms reduce time requirements in real-time object detection [14]. The use of Analytic Wavelet Transform (AWT) is more beneficial over Mallat wavelet transform, as AWT is translation invariant in nature [13].

Kopaczka et al. developed a system that uses a ZeroMQ-based client-server system [15]. Image acquisition and loading, face detection, facial landmark detection, frontalization, and analysis are used as distinct ZeroMQ nodes that receive their data from the server, then send it back after processing. The server forwards the data to a graphical user interface, which allows for the selecting of different modules, which can then display their results. Because of the inherent robustness of the system, all modules are designed to work independently of the others. This makes it so that a crashing module does not affect the entire system, and modules can be interchanged at run-time. See Figure 5.

Figure 5. The user interface of the tracking system.

1: Live video view with overlaid landmarks from the automated ShapeNet landmark detection (red) and their respective bounding box corners (blue). 2: Frontalized view of the face for improved face analysis. 3: Output of the current analysis module, in this case, a breathing rate analysis. 4: Options panel with module selection

Source: [6]

If looked at closely, the fused face images show spatially varying texture patterns. Instead of globally evaluating the face's surface texture pattern, it would be beneficial to consider a face image as a collection of a finite number of overlapping local regions, then classifying the face image on the basis of texture information of each individual local region. Combining all the regional classification results will make the final classification. To deal with facial occlusions on 3-D images, local analysis of facial surface was proposed, which has been used for handling facial expression variations. Thirty-four overlapping local region templates of the face were considered in Figure 6. The white area represents the considered local face region and the black area represents the excluded face region [16].

Facial skin temperature is closely related to underlying blood vessels, and the pattern of the blood vessels below the skin can be extracted by obtaining a thermal map of the human face. By performing morphological operations, like opening and top-hat segmentation to give thermal signatures, thermal feature extraction from facial images can be attained.

Consistent features can be defined as features that are present in three or more thermal signatures from images taken at different times. Thermal signatures can then be extracted from each image, with the thermal signature extraction process divided into four main parts. These parts comprise face segmentation, noise removal, image morphology, and postprocessing. An individual&#;s thermal signatures vary from day to day because of exercise, environmental temperature, individual health, weight, imaging room temperature, and other factors. Due to these factors that can affect the thermal signature, it is recommended to establish a thermal signature template that keeps the characteristics in an individual&#;s thermal signature which are consistent over time. Thermal signature template generation requires the taking of extracted thermal signatures for each individual and adding them together. This creates an image, which is a composite of multiple, different signature extractions. Keeping the features that are present in all images as the dominant features that can best define the individual signature will be achieved. To fuse the predominant features, an anisotropic diffusion filter can be applied to the result of the added thermal signatures, shown in Figure 7 [1].

Figure 7. Generation of the thermal signature template.

(a) Resultant image of the addition of four thermal signatures. (b) Results of applying anisotropic diffusion on the summed image. (c) Thermal signature template of the subject.

Source: [1]

Are you interested in learning more about tablet thermal face recognition camera? Contact us today to secure an expert consultation!

To capture new anatomical and physiological face information, IR images can be used. This information can consist of the structure of blood vessels, facial vascular networks, and facial tissue, as well as thermal face signatures, which can be used as unique biometric features, as shown in Figure 8 [9].

Figure 8. Thermal and processed images.

 Face grayscale thermal image (a); blood perfusion obtained transforming a medial axis on bit-plane (b); morphological grey level erosion and medial axis transform (c); the result of Sobel operator (d)

Source: [9]

The process developed by Pavlidis et al. is shown, along with the facial vascular network in Figure 9 [3].

Figure 9. Segmented facial images annotated with the facial vascular network (white lines).

Source: [3]

In today&#;s world, where many health risks are at play, technology can do a lot to keep us safe. Many smart devices have proven to be effective in maintaining public security. One such device is a facial recognition security camera, which is used to recognize an individual&#;s face. It uses biometrics to identify facial features from an image or video. The system then analyzes and compares the information with a data bank of other faces to detect a match.

Typically, a facial recognition security system is seen as access control and is associated with other biometrics like eye iris or fingerprints recognition systems. Obviously, the accuracy of iris recognition or fingerprint recognition as biometric technology is higher than a facial recognition accuracy. However, face recognition products are widely adopted thanks to their non-invasive, contactless, quick, and accurate screening.

On top of that, today's market is offering facial scanners enabled with thermal scans. They are invaluable devices during this pandemic because they monitor temperatures along with allowing/denying entry based on a person&#;s health condition. All in all, face recognition technology is widely used for various purposes. In this post, our security experts will tell when and why we need security cameras with facial recognition. Keep reading to get to know their basic operation principles, the benefits of using these current devices, and more useful facts.

How a facial recognition security camera works

People are good at recognizing faces; we all easily identify the faces of our friends, family members, and acquaintances. That's because we are familiar with their face landmarks - eyes, nose, lips, etc. A face recognition system works the same way, but on a large, algorithmic scale - where you see a face, the device sees data. In other words, facial recognition hardware is a biometric artificial intelligence that can identify an individual by analyzing facial texture and shape. All the data is stored in the data bank and can only be accessed by authorized personnel. Technologies may vary, but below you'll find basic principles of operation.

• An image of an individual's face is captured from the image or video; the face may appear alone or in the crowd.
• A facial recognition scanner reads the face geometry; key factors are the distance from forehead to chin, the distance between eyes, bridge of the nose, the contour of the chin, lips, ears, etc. Plus, it recognizes facial landmarks. In the end, the software acquires a facial signature.
• A determination is made. The facial signature becomes a mathematical formula; it's stored in the data bank and can be compared with other faces.

Note that face recognition products vary in their ability to distinguish individuals under challenging conditions, e.g., suboptimal angle of view, low-quality image resolution, poor lighting, etc. So, it's always better to choose high-quality devices that promise to work effectively and provide excellent results.

Characteristics

Nowadays, facial recognition technology is advancing and becoming more widespread. It has a long history and was previously only used for security purposes. But today facial scanners are suitable even for medical use. Why are these devices so popular? What are their characteristics? Let's find out!

• A face recognition device can identify a person from about 8m distance.
• Thanks to the unique algorithm that combines a machine learning method and deep face learning, the system will be exceptionally accurate and recognize a person even if he/she wears a surgical mask or glasses.
• A high-quality device will feature an analytics section (gender and age statistics, people counting, etc.).
• A facial scanner can be integrated into the company's existing security system. Plus, it's apps-compatible.
• Premium-quality equipment can automatically optimize the settings to enhance video images detectability.
• Face images that the scanner has detected are retained in the database, and they can be searched later.
• Particular faces can be designated beforehand to send instant notification as soon as they are detected.
• Facial recognition access control will prohibit unverified individuals from entering the restricted area.
• The best equipment is enabled with a thermal scan, which will check visitors' temperature. As we all know, continuous temperature monitoring is one of the present-day pandemic precautions.

In sum, face recognition products offer plenty of beneficial features. Everyone (homeowner, enterprisers, authorities) who invests in such a device will benefit due to its reliability and accuracy. The technology may be as simple as a face recognition door lock system or sophisticated enough to serve national security purposes. You can choose any of them based on your demands and budget.

Benefits of facial recognition systems enabled with a thermal scan

There are many types of facial scanners available on the market, the principles of their operation are similar. However, some of the systems offer unique features. An excellent example of such systems is thermal cameras, or facial recognition systems enabled with a thermal scan. Thermal facial scanners capture heat emitted by the human body, detect the head's shape, and ignore accessories like makeup, glasses, or hats. Unlike conventional facial scanners, thermal cameras are capable of identifying even in nighttime or low-light conditions. There is no need to use a flash or expose the camera's location so that it can be used discreetly.

The accuracy of such a device is higher than 95%, depending on the brand and manufacturer. As we touched on above, the facial recognition system enabled with a thermal scan can monitor an individual&#;s temperature. This is one of the features that distinguishes a thermal camera from a conventional one. It performs not only as a fast access facial recognition tool but also as a temperature monitoring tool. This device will add huge value to the entrances, checkpoints, and other places that see the most foot traffic. It won't be an exaggeration to say that thermal scanners help to stop the spread of COVID-19 since the first sign of this deadly virus is elevated temperature.

Today&#;s market offers lots of thermal scanner options. Public health experts and numerous business owners highly recommend smart kiosks and smart sanitizing hubs that provide fast access facial recognition and temperature monitoring.

Smart Kiosk

This facial recognition system is ideal to place at waiting areas, entrances, and checkpoints. The system can be customized to a company&#;s needs, and its height is easily adjustable (2, 3.5, or 4.5 feet tall). The thermometer is designed to distinguish individuals, check temperatures, capture images, and send alerts to personnel if a high temperature is detected. The smart kiosk can perform as a quick, accurate employee-tracking tool, streamlining workers&#; flow during shift changes.

Smart Sanitizing Hub

This stationary equipment is perfect for various organizations and offices. It's not only capable of recognizing faces but also catches the body temperature and notifies the thermographer if an individual has fever-like symptoms. This station features a built-in hand-sanitizing tool so that employees or visitors can clean their hands while being monitored for higher temperatures. Using this device will minimize the chances of infection to zero.

Use Cases

Facial recognition systems may be used for a multitude of applications! Here are several examples of their typical use:

Security: a biometric face recognition system helps to secure employees and visitors in companies and public places. It's capable of detecting criminals, missing people, exploited children, etc. Such equipment always supports and accelerates investigations.

• Governments: facial recognition systems help to identify individuals who overstay their visas.
• Face ID technology: it allows users to unlock their phones with the help of their face signature.
• Social media: facial recognition tools help to tag people in photos.
• Marketing: such systems help to identify age, gender, ethnicity, and target specific audiences.
• Biometric face recognition systems improve human-computer interaction.
• Healthcare: thanks to face analysis provided by facial recognition devices, now it's possible to detect some genetic diseases, accurately track patient's use of medicines, etc.
• Company security: facial recognition systems allow/deny entry to buildings, offices, or restricted areas.

As for facial recognition systems enabled with thermal scan, they have proven to be an effective method of combating COVID-19. Business owners can install them at checkpoints, entrances, etc., as an employee-tracking tool. Such a device will only permit entry to verified individuals. It promises to keep potentially infected individuals at home. How? The system checks the person&#;s temperature, and if they are feverish, they will be denied entry.

Smart kiosks and smart sanitizing hubs feature instant notifications&#;the systems alert personnel as soon as an individual with fever-like symptoms is detected. These devices are perfect for hospitals, schools, waiting areas, lobbies, public transport terminals, and any other high traffic area.

Facial scanners with temperature controls have been vital during this public health crisis. Although biometric facial recognition systems can be pricy, they&#;re worth every cent to keep employees and customers safe as we adjust to this new normal. These systems offer accurate monitoring and minimize chances of cross-contamination in public spaces.

Cost

Facial recognition system costs vary greatly and can cost as much as $30,000. The good new is that it&#;s possible to find less expensive equipment without compromising the quality. For example, a high-quality smart sanitizing hub with facial recognition features costs about $3,200, while the smart kiosk is about $2,400. While it&#;s still a substantial investment, especially for small businesses, it&#;ll definitely worthwhile so you can create a safe environment for your employees, partners, and guests with the best, reliable technology.

Want more information on industrial android computer? Feel free to contact us.