Face Orientation Research Articles

Facial Feature Localization and Analysis Using Machine Learning

The facial feature localization system with machine learning (ML) is an innovative solution designed to accurately detect and locate key facial landmarks. By integrating advanced ML algorithms, the system processes images or video streams in real time to identify the positions of facial features such as the eyes, nose, mouth, and jawline. The ML component analyses the data to precisely locate these features, enabling applications in areas like facial expression analysis and biometric verification. This approach enhances accuracy by adapting to various lighting conditions, face orientations, and individual facial variations. The system’s scalability and flexibility further extend its utility, supporting diverse applications across industries like healthcare, security, and human-computer interaction. Overall, this project represents a significant advancement in facial feature localization, using IoT and ML to enable more effective and accurate facial analysis. By adapting to various lighting conditions, face orientations, and individual facial variations" to "by adapting to diverse lighting conditions, different face orientations, and individual facial variations, ensuring reliable performance in various environments.

Deteksi Wajah dalam Foto Menggunakan Teknologi Visi Komputer

This study discusses the implementation of computer vision technology for face detection in photos using two sample images with variations in lighting and face pose. The developed system combines the Viola-Jones algorithm and Convolutional Neural Networks (CNN) to enhance resilience against lighting and face orientation variations. Experimental results show high accuracy even with only two sample images. This research also develops preprocessing techniques to handle extreme lighting conditions and demonstrates efficient implementation using Python and OpenCV.

Usability of an AI-based independent pedestrian navigation device designed for blind people

Abstract Background 2.2 billion people live with visual impairment (VI) worldwide and face orientation and mobility challenges in various aspects of their lives. Despite technological advancements, the most used mobility solutions for VI are still the walking cane and the guide dog. The objective of this research is to assess the system usability of an AI-based independent pedestrian navigation device (.lumen glasses), designed for blind people. Methods During 31st of October and 2nd of November 2023, 27 visually impaired people living in Bucharest, Romania, tested the .lumen glasses for 1-3 hours during an outdoor navigation task. The glasses use AI to recognize the user’s environment and haptic and audio feedback to guide them. The System Usability Scale (SUS), a 10-item instrument, was used to evaluate the level of usability the participants ascribe to the .lumen glasses. Qualitative feedback was collected as part of the post-test survey and analyzed using thematic analysis. Data were triangulated to identify usability issues. Results The mean age of the participants was 30.3 years old (SD = 12.07, min=21 max =56, range: 48), 55.6% (N = 15) of the participants were male, 96,3% (N = 26) were completely blind, and 81.5% were using a white cane for orientation and mobility. Our findings reveal a mean SUS score of 83.65 (SD = 15.23, range 47.5, min=52.5, max=100) indicating high usability. The thematic analysis revealed 5 themes: (1) auditive and (2) haptic feedback in navigation, (3) hardware (adjustment system & buttons), (4) tutorial & presentation of how the device works and (5) other aspects (usage in combination with other assistive technologies etc.). The participants declared that the glasses are intuitive to use and promote the feeling of safety and independence. Conclusions Data suggest that the .lumen glasses area promising assistive technology for the mobility of VI. Findings are currently used to steer the development and implementation of the AI-based system. Key messages • According to participants’ feedback, .lumen glasses are a scalable AI-based mobility solution that can make navigation and orientation easier for visually impaired individuals. • .lumen glasses designed for the visually impaired individuals are one of the current and limited assistive technology improving the mobility and orientation.

Spotting missing or wanted people: racial biases in prospective person memory

Searching for missing or wanted people is a crucial task in our society. Previous work on prospective person memory (PPM) has demonstrated that performance on this type of search task is worse relative to standard prospective memory tasks. Importantly, this process may be further affected by the race of the missing person, yet this has never been tested in laboratory settings. To test the effects of race on PPM, a convenience sample consisting primarily of self-identified Caucasian participants was asked to search for either a Caucasian or an Indian target person while judging the orientation of different Caucasian and Indian faces. Although the tasks were otherwise identical, 89% of Caucasian PPM targets were found while only 53% of Indian targets were found. Furthermore, relative to a control group with no PPM requirements, participants were slower and more error-prone when judging Indian faces relative to White faces, particularly if they were searching for an Indian face. We interpret these results as revealing other-race effects in prospective person memory, highlighting race as a critical factor for finding missing people. Importantly, this also emphasizes the need for real-world search efforts to factor in difficulty differences when people monitor for missing/wanted people from their own or different racial backgrounds. For example, media coverage of missing persons cases could perhaps be distributed more equitably by considering whether the missing person is from a racial minority in that region.

Minimal exposure durations reveal visual processing priorities for different stimulus attributes

Human vision can detect a single photon, but the minimal exposure required to extract meaning from stimulation remains unknown. This requirement cannot be characterised by stimulus energy, because the system is differentially sensitive to attributes defined by configuration rather than physical amplitude. Determining minimal exposure durations required for processing various stimulus attributes can thus reveal the system’s priorities. Using a tachistoscope enabling arbitrarily brief displays, we establish minimal durations for processing human faces, a stimulus category whose perception is associated with several well-characterised behavioural and neural markers. Neural and psychophysical measures show a sequence of distinct minimal exposures for stimulation detection, object-level detection, face-specific processing, and emotion-specific processing. Resolving ongoing debates, face orientation affects minimal exposure but emotional expression does not. Awareness emerges with detection, showing no evidence of subliminal perception. These findings inform theories of visual processing and awareness, elucidating the information to which the visual system is attuned.

Self-Organizing a Latent Hierarchy of Sketch Patterns for Controllable Sketch Synthesis.

Encoding sketches as Gaussian mixture model (GMM)-distributed latent codes is an effective way to control sketch synthesis. Each Gaussian component represents a specific sketch pattern, and a code randomly sampled from the Gaussian can be decoded to synthesize a sketch with the target pattern. However, existing methods treat the Gaussians as individual clusters, which neglects the relationships between them. For example, the giraffe and horse sketches heading left are related to each other by their face orientation. The relationships between sketch patterns are important messages to reveal cognitive knowledge in sketch data. Thus, it is promising to learn accurate sketch representations by modeling the pattern relationships into a latent structure. In this article, we construct a tree-structured taxonomic hierarchy over the clusters of sketch codes. The clusters with the more specific descriptions of sketch patterns are placed at the lower levels, while the ones with the more general patterns are ranked at the higher levels. The clusters at the same rank relate to each other through the inheritance of features from common ancestors. We propose a hierarchical expectation-maximization (EM)-like algorithm to explicitly learn the hierarchy, jointly with the training of encoder-decoder network. Moreover, the learned latent hierarchy is utilized to regularize sketch codes with structural constraints. Experimental results show that our method significantly improves controllable synthesis performance and obtains effective sketch analogy results.

Effect of riding postures on kinematic responses and head injuries of ETW child passengers

Electric two-wheelers (ETWs) are extensively used for the transportation of vulnerable child passengers; however, diverse riding postures adopted by children may engender potential injury risks in the event of a collision. The objective of this study was to investigate the kinematics and head injuries of child passengers with different riding postures in ‘ETW-sedan’ collisional accidents. Typical riding postures were classified and different collision scenarios were designed using the orthogonal matrix analysis method. Multibody models of humans and vehicles were developed and validated using empirical accident data. The effects of riding postures of child passengers on head injury mechanisms and kinematics were studied. The result showed that it was the most dangerous posture when the child sat in the back seat and had a reverse facial orientation with the rider, due to the highest incidence of fatal head injury. A parametric study was further conducted on the effect of different impact velocities, impact angles, and offset distances. It was found that the velocity of the sedan was the most influencing factor, and severe damage can occur when the impact angle was ranging from 30 to 90°. The study could provide guidance for the safety protection of child passengers on ETWs.

Human body features recognition based adaptive user interface for extra-large touch screens

With the widespread use of extra-large touch screens (eLTS) in various settings such as work and education, interaction efficiency and user experience have garnered increased attention. The current user interface (UI) layouts of eLTS are primarily categorized into two modes: fixed position and manual adjustment. The fixed UI layout fails to accommodate users of different heights and sizes, while the manual adjustment mode involves cumbersome steps and lacks sufficient flexibility. This study proposes an adaptive UI for eLTS. The optimal operational area on the eLTS is determined based on users’ height, eye level, arm length, face orientation, and distance from the screen. The eLTS menu is then positioned and displayed within this optimal area. Simulations involving users of various heights (P1 female, P50 male and female, and P99 male) were conducted to evaluate fatigue using the rapid upper limb assessment (RULA) method. The results indicate that the proposed adaptive UI significantly reduces user fatigue.

A computational account of conflict processing during mental imagery.

Previous studies examining conflict processing within the context of a color-word Stroop task have focused on both stimulus and response conflicts. However, it has been unclear whether conflict can emerge independently of stimulus conflict. In this study, a novel arrow-gaze mental-rotation Stroop task was introduced to explore the interplay between conflict processing and mental rotation. A modelling approach was utilized to provide a process-level account of the findings. The results of our Stroop task indicate that conflict can emerge from mental rotation in the absence of stimulus conflict. The strength of this imagery conflict effect decreases and even reverses as mental rotation angles increase. Additionally, it was observed that participants responded more quickly and with greater accuracy to small rather than large face orientations. A comparison of three conflict diffusion models-the diffusion model for conflict tasks (DMC), the dual-stage two-phase model (DSTP), and the shrinking spotlight model (SSP)-yielded consistent support for the DSTP over the DMC and SSP in the majority of instances. The DSTP account of the experimental results revealed an increased nondecision time with increasing mental rotation, a reduction in interference from incompatible stimuli, and an improved drift rate in response selection phase, which suggests enhanced cognitive control. The findings from the model-based analysis provide evidence for a novel interaction between cognitive control and mental rotation.

Including Dynamic Viewing Behavior in Determining Vertical Eye-Level Light: Does it Matter for Office Design?

ABSTRACT When considering non-image-forming (NIF) light effects on people, knowing the light vertically at eye-level is necessary. However, people are dynamic in their behavior and constantly change their posture and viewing direction. This means that light measured vertically toward a constant direction might differ from the actual light that reaches people’s eyes. If the difference is large, occupant behavior might need to be included in lighting design measurements and simulations predicting the potential of NIF light effects. This paper presents an experiment on the difference between the actual dynamic eye-level light of office workers while seated at a desk (dynamic condition) and light measured statically toward a computer screen (static condition). Eye-level light conditions in relation to NIF effects were quantified using the melanopic Equivalent Daylight Illuminance (mEDI) and the relative alerting response (r D ). Measured dynamic mEDI was 4% to 10% larger than the static one. Analysis of r D was not possible due to saturation of the response. The dominant face orientation was toward the computer screen, with a deviation in the horizontal and vertical direction within 15 degrees from the center. The results show that measuring and simulating personal light conditions of office workers statically toward a computer screen might be a small underestimation of actual eye-level light, but the relevance of this underestimation in lighting design practice is not evident.

A smartphone application for personalized facial aesthetic monitoring.

Methods available at home for capturing facial images to track changes in skin quality and evaluate skincare treatments are limited. In this study, we developed a smartphone camera application (app) for personalized facial aesthetic monitoring. A face alignment indicators (FAIN) system utilizing facial landmark detection, an artificial intelligence technique, to estimate key facial parts, was implemented into the app to maintain a consistent facial appearance during image capture. The FAIN system is composed of a fixed target indicator and an alignment indicator that dynamically changes its shape according to the user's face position, size, and orientation. Users align their faces to match the alignment indicator with the fixed target indicator, and the image is automatically captured when alignment is achieved. We investigated the app's effectiveness in ensuring a consistent facial appearance by analyzing both geometric and colorimetric data. Geometric information from captured faces and colorimetric data from stickers applied to the faces were utilized. The coefficients of variation (CVs) for the L*, a*, and b* values of the stickers were higher compared to those measured by a colorimeter, with CVs of 14.9 times, 8.14 times, and 4.41 times for L*, a*, and b*, respectively. To assess the feasibility of the app for facial aesthetic monitoring, we tracked changes in pseudo-skin color on the cheek of a participant using skin-colored stickers. As a result, we observed the smallest color difference ∆Eab of 1.901, which can be considered as the experimentally validated detection limit using images acquired by the app. While the current monitoring method is a relative quantification approach, it contributes to evidence-based evaluations of skincare treatments.

A Sheep Identification Method Based on Three-Dimensional Sheep Face Reconstruction and Feature Point Matching.

As the sheep industry rapidly moves towards modernization, digitization, and intelligence, there is a need to build breeding farms integrated with big data. By collecting individual information on sheep, precision breeding can be conducted to improve breeding efficiency, reduce costs, and promote healthy breeding practices. In this context, the accurate identification of individual sheep is essential for establishing digitized sheep farms and precision animal husbandry. Currently, scholars utilize deep learning technology to construct recognition models, learning the biological features of sheep faces to achieve accurate identification. However, existing research methods are limited to pattern recognition at the image level, leading to a lack of diversity in recognition methods. Therefore, this study focuses on the small-tailed Han sheep and develops a sheep face recognition method based on three-dimensional reconstruction technology and feature point matching, aiming to enrich the theoretical research of sheep face recognition technology. The specific recognition approach is as follows: full-angle sheep face images of experimental sheep are collected, and corresponding three-dimensional sheep face models are generated using three-dimensional reconstruction technology, further obtaining three-dimensional sheep face images from three different perspectives. Additionally, this study developed a sheep face orientation recognition algorithm called the sheep face orientation recognition algorithm (SFORA). The SFORA incorporates the ECA mechanism to further enhance recognition performance. Ultimately, the SFORA has a model size of only 5.3 MB, with accuracy and F1 score reaching 99.6% and 99.5%, respectively. During the recognition task, the SFORA is first used for sheep face orientation recognition, followed by matching the recognition image with the corresponding three-dimensional sheep face image based on the established SuperGlue feature-matching algorithm, ultimately outputting the recognition result. Experimental results indicate that when the confidence threshold is set to 0.4, SuperGlue achieves the best matching performance, with matching accuracies for the front, left, and right faces reaching 96.0%, 94.2%, and 96.3%, respectively. This study enriches the theoretical research on sheep face recognition technology and provides technical support.

Effect of an accelerating metro cabin on the diffusion of cough droplets

Coronaviruses being capable of spreading through droplet contamination have raised significant concerns regarding high-capacity public rail transport, such as the metro. Within a rapidly moving railcar cabin, the internal airflow lags behind the bulkhead, generating internally induced airflow that accelerates droplet dispersion within a non-inertial reference system. This study investigates the impact of acceleration on the diffusion of cough droplets of varying sizes using computational fluid dynamics. The modified k–ε equation in ANSYS® Fluent was utilized to simulate droplet diffusion under different body orientations by adjusting the inertial force correction source term. Results indicate that droplets in the middle size range (50–175 μm) are primarily influenced by inertial forces, whereas smaller droplets (3.5–20 μm) are predominantly controlled by air drag forces. Regardless of facial orientation, the outlet of high-capacity public rail transport poses the highest risk of infection.

Effect of constrained conditions, crystallographic and geometric factors on deformation patterns in FCC single crystals. Molecular dynamics study

Predicting of the mechanical properties of a polycrystal requires knowledge of the behavior of its individual grains during deformation and their mutual influence on each other. The self-similarity of deformation processes at different scales in crystalline solids allows using of molecular dynamics simulations to analyze the deformation of individual grains of a polycrystal, treating them as single crystals. We have studied the patterns of slip and twinning deformation caused by differences in stacking fault energy, crystal morphology, crystallographic orientation and shear conditions towards the lateral surfaces. It was found that in addition to the stacking fault energy and the crystallographic direction of compression, the crystallographic orientation of the lateral faces and the boundary conditions on them must be taken into account. They affect the accumulation of dislocations, the types of formed dislocations, low-mobile and immobile dislocation complexes, and the shear propagation in the crystal. The changes in shear conditions towards the lateral faces explain the size effect of the grain size/thickness ratio on mechanical properties under uniaxial strain.

The distinct effects of fearful and disgusting scenes on self-relevant face recognition

Self-face recognition denotes the process by which a person can recognize their own face by distinguishing it from another’s face. Although many research studies have explored the inhibition effect of negative information on self-relevant face processing, few researchers have examined whether negative scenes influence self-relevant face processing. Fearful and disgusting scenes are typical negative scenes, but little research to data has examined their discriminative effects on self-relevant face recognition. To investigate these issues, the current study explored the effect of negative scenes on self-relevant face recognition. In Study 1, 44 participants (20 men, 24 women) were asked to judge the orientation of a target face (self-face or friend-face) pictured in a negative or neutral scene, whereas 40 participants (19 men, 21 women) were asked to complete the same task in a fearful, disgusting, or neutral scene in Study 2. The results showed that negative scenes inhibited the speed of recognizing self-faces. Furthermore, the above effect of negative scenes on self-relevant face recognition occurred with fearful rather than disgusting scenes. Our findings suggest the distinct effects of fearful scenes and disgusting scenes on self-relevant face processing, which may be associated with the automatic attentional capture to negative scenes (especially fearful scenes) and the tendency to escape self-awareness.

Online Orientation Recognition of Single-Crystal Diamond Tools in the Process of Indexing Grinding Based on HMM and Multi-Information Fusion

Single-crystal diamond tools occupy an important position in the field of optical processing as the basis and key to advanced optical manufacturing technology, such as grating manufacturing and optical mirror-turning processing. Single-crystal diamond tools have become the cornerstone of the development of related industries. This paper takes a single-crystal diamond arc tool as the research object. Sound signal analysis technology and vibration signal analysis technology are comprehensively applied to the online orientation identification of a single-crystal diamond tool in the indexing grinding process. The online orientation method of the tool is explored, the sound signal and the vibration signal are taken as the characteristic signals, and a wavelet algorithm (WT) is used to reduce the noise of the vibration signal and sound signal. The kurtosis of the sound signal and the kurtosis and skewness of the vibration signal in the high-order statistics strongly related to the grinding direction of a single-crystal diamond are used as the characteristic parameters, and the online direction recognition model of the tool is established using the Hidden Markov Method (HMM). The above characteristic parameters are used as model input for multi-information fusion. The mapping relationship between the characteristic parameters of the characteristic signal and the crystal orientation of the single-crystal diamond crystal face is obtained, and then the online orientation method of the single-crystal diamond arc tool in the process of indexing grinding is formed. The effectiveness of the method is verified by experiments, and effective orientation information is provided for research on the positioning control strategy of the tool grinding process to ensure the efficiency of grinding and improve the manufacturing level of the tool.

The Timing of Gaze Direction Perception: ERP Decoding and Task Modulation

Distinguishing the direction of another person's eye gaze is extremely important in everyday social interaction, as it provides critical information about people's attention and, therefore, intentions. The temporal dynamics of gaze processing have been investigated using event-related potentials (ERPs) recorded with electroencephalography (EEG). However, the moment at which our brain distinguishes the gaze direction (GD), irrespectively of other facial cues, remains unclear. To solve this question, the present study aimed to investigate the time course of gaze direction processing, using an ERP decoding approach, based on the combination of a support vector machine and error-correcting output codes. We recorded EEG in young healthy subjects, 32 of them performing GD detection and 34 conducting face orientation tasks. Both tasks presented 3D realistic faces with five different head and gaze orientations each: 30°, 15° to the left or right, and 0°. While the classical ERP analyses did not show clear GD effects, ERP decoding analyses revealed that discrimination of GD, irrespective of head orientation, started at 140 ms in the GD task and at 120 ms in the face orientation task. GD decoding accuracy was higher in the GD task than in the face orientation task and was the highest for the direct gaze in both tasks. These findings suggest that the decoding of brain patterns is modified by task relevance, which changes the latency and the accuracy of GD decoding.

The effect of processing partial information in dynamic face perception

Face perception is a major topic in vision research. Most previous research has concentrated on (holistic) spatial representations of faces, often with static faces as stimuli. However, faces are highly dynamic stimuli containing important temporal information. How sensitive humans are regarding temporal information in dynamic faces is not well understood. Studies investigating temporal information in dynamic faces usually focus on the processing of emotional expressions. However, faces also contain relevant temporal information without any strong emotional expression. To investigate cues that modulate human sensitivity to temporal order, we utilized muted dynamic neutral face videos in two experiments. We varied the orientation of the faces (upright and inverted) and the presence/absence of eye blinks as partial dynamic cues. Participants viewed short, muted, monochromic videos of models vocalizing a widely known text (National Anthem). Videos were played either forward (in the correct temporal order) or backward. Participants were asked to determine the direction of the temporal order for each video, and (at the end of the experiment) whether they had understood the speech. We found that face orientation, and the presence/absence of an eye blink affected sensitivity, criterion (bias) and reaction time: Overall, sensitivity was higher for upright compared to inverted faces, and in the condition where an eye blink was present compared to the condition without an eye blink. Reaction times were mostly faster in the conditions with higher sensitivity. A bias to report inverted faces as ‘backward’ observed in Experiment I, where upright and inverted faces were presented randomly interleaved within each block, was absent when presenting upright and inverted faces in different blocks in Experiment II. Language comprehension results revealed that there was higher sensitivity when understanding the speech compared to not understanding the speech in both experiments. Taken together, our results showed higher sensitivity with upright compared to inverted faces, suggesting that the perception of dynamic, task-relevant information was superior with the canonical orientation of the faces. Furthermore, partial information coming from eye blinks, in addition to mouth movements, seemed to play a significant role in dynamic face perception, both when faces were presented upright and inverted. We suggest that studying the perception of facial dynamics beyond emotional expressions will help us to better understand the mechanisms underlying the temporal integration of facial information from different -partial and holistic- sources, and that our results show how different strategies, depending on the available information, are employed by human observers when judging the temporal order of faces.

Infant Faces Induce Behavioral Approach Tendencies: Evidence From a Manikin Task1,2

AbstractInfant faces are generally perceived as being cute and motivate the observer to look at them longer. This phenomenon is thought to be based on elemental features called baby schema (e.g., a round face and a large forehead). This study investigated the behavioral tendencies of observers in approaching infant faces using a manikin task in which infant and adult face images were presented as stimuli. Participants (N = 40) were instructed to identify with the manikin, discriminate a face on the screen, and respond by moving the manikin closer to the face (approach trials) or farther away from it (avoidance trials). The faces were presented either upright or inverted. The results showed that the approach–avoidance indices (the mean reaction time and error rate of avoidance trials minus those of approach trials) were larger for infant faces than for adult faces regardless of face orientation, although the difference was greater for upright faces than for inverted faces. This approach tendency toward infant faces is possibly due to elemental features rather than face‐specific holistic features.

One-Shot Neural Face Reenactment via Finding Directions in GAN’s Latent Space

In this paper, we present our framework for neural face/head reenactment whose goal is to transfer the 3D head orientation and expression of a target face to a source face. Previous methods focus on learning embedding networks for identity and head pose/expression disentanglement which proves to be a rather hard task, degrading the quality of the generated images. We take a different approach, bypassing the training of such networks, by using (fine-tuned) pre-trained GANs which have been shown capable of producing high-quality facial images. Because GANs are characterized by weak controllability, the core of our approach is a method to discover which directions in latent GAN space are responsible for controlling head pose and expression variations. We present a simple pipeline to learn such directions with the aid of a 3D shape model which, by construction, inherently captures disentangled directions for head pose, identity, and expression. Moreover, we show that by embedding real images in the GAN latent space, our method can be successfully used for the reenactment of real-world faces. Our method features several favorable properties including using a single source image (one-shot) and enabling cross-person reenactment. Extensive qualitative and quantitative results show that our approach typically produces reenacted faces of notably higher quality than those produced by state-of-the-art methods for the standard benchmarks of VoxCeleb1 & 2.