Thermal image super-resolution remains a challenging task due to the limited spatial detail captured by infrared sensors. Although RGB-guided methods and domain adaptation techniques have shown promise, they often introduce architectural complexity or require multimodal inputs at inference time. In this study, we investigate the integration of Fourier Domain Adaptation (FDA) as a lightweight preprocessing strategy to improve the performance of the state-of-the-art Dense-Residual-Connected Transformer (DRCT) model for image super-resolution. FDA transfers low-frequency information from grayscale versions of RGB images to thermal images during training, enabling the model to incorporate additional structural information without introducing any additional architectural complexity or adding overhead during inference. Experimental results on the Tufts Face Database demonstrate that the FDA-enhanced model improves PSNR by up to 0.17 dB and SSIM by up to 0.0004 compared to thermal-only training at a x2 scaling factor. These findings highlight the effectiveness of simple frequency-based adaptation techniques in improving the generalization of thermal SR models while preserving the simplicity of the inference pipeline.

The segmentation of breasts in mammography images can be used to remove useless information, such as annotations, from tissue regions that are important for further analysis. Since the nipple is comparatively small, models optimised for global metrics often fail to include it, even though overall metrics remain high. In this work, we investigate a complex method to improve the accuracy and efficiency of breast segmentation in mammography, focused on the nipple area. The proposed multi-phase method combines the deep learning approach with the morphological method of erosion and dilation. The proposed method is compared with two baselines: (i) pure deep learning, (ii) purely morphological. The results show that smoothing improves YOLO11's segmentation metrics and enhances the visual quality of UNet outputs. However, the best overall performance was achieved by the UNet model, with an IoU of 0.978 and Dice score of 0.989 for the entire breast, and 0.920 IoU and 0.957 Dice for the nipple area.

Camera-based vital sign estimation is a very active area of research for human-system interaction and medical diagnostics. The sequence of frames is processed to estimate respiration rate, blood volume pulse (BVP), emotions, etc. In remote photoplethysmography, the source of the blood pulse signal is the arteries in the skin. This study aims to investigate facial skin segmentation methods for rPPG signal estimation. Three methods were adapted and evaluated: a threshold-based method (YCrCb/HSV), deep semantic segmentation using the BiSeNet model, and an attention-based approach derived from Grad-CAM visualizations of the TS-CAN network. Each method was used to generate a facial mask applied to extract the green-channel intensity signal from video frames. Signal quality was evaluated using two metrics:$\mathbf{S N R}_{{raw }}$and$\mathbf{S N R}_{max}$. Additionally, IoU metrics were used to quantify spatial overlap between the masks. The results show that BiSeNet segmentation yielded the highest signal quality ($\mathbf{S N R}_{max }=\mathbf{7. 8 1 8} \mathbf{dB}$), while Grad-CAM masks, applied without additional processing, achieved the lowest performance ($\mathbf{S N R}_{max }=-\mathbf{1. 5 3 5} \mathbf{~ d B}$). This indicates that although attention maps can identify important facial regions, further refinement is required to improve their effectiveness in$r$PPG signal extraction. Spatial analysis revealed strong agreement between the threshold-based and BiSeNet masks, with significantly lower alignment between either and the Grad-CAM-based mask. An additional analysis of Grad-CAM attention distribution across facial regions showed that the model focused mainly on the cheeks, regions associated with more stable pulsatile signals. These findings emphasize the importance of accurate skin segmentation for enhancing rPPG signal quality and show that attention maps used for interpreting network behavior do not precisely reflect the regions relevant for accurate skin segmentation. This work supports the development of more explainable and reliable data-driven rPPG systems.

In this study, we investigate the robustness of various deep learning models to inpainted lung CT scans. We train several image classification models (EfficientNet, MobileNet, DenseNet, and a custom CNN) as well as one object detection model (YOLO) to determine whether a scan contains a tumor or not. We then apply a pretrained GAN to inpaint tumor regions in test images, effectively hiding cancer presence. The resulting set of modified images is then evaluated by the previously trained models. We observe that the classifiers test accuracy remains almost unchanged, whereas the YOLO model adjusts its predictions, failing to detect tumors that were previously visible. This suggests that classification models may have learned non-obvious patterns (or biases) associated with cancer presence, possibly integrating features from larger image regions rather than focusing solely on localized tumor areas. In contrast, the YOLO model depends more on local visual features and is more vulnerable to inpainting-based adversarial attacks. These findings highlight the need to consider model architecture and susceptibility to image manipulation when deploying AI in human-system interaction tasks.

The world is producing a massive amount of single-use waste, especially plastic waste made from polymers. Such waste is usually distributed in large areas within cities, near roads, parks, forests, etc. It is a challenge to collect them efficiently. In this work, we propose a Cleaning Robot as an autonomous vehicle for waste collection, utilizing the Nvidia Jetson Nano platform for precise arm movements guided by computer vision capabilities. Integrated with the Raspberry Pi platform for mobility control, the robot employs the YOLO (You Only Look Once) framework for efficient waste detection and classification. The model was trained, implemented in the robot prototype, and tested on preprocessed waste images, resulting in mean average precision (mAP) above 80 percent. Our design emphasizes singular-object focus, enabling real-time detection of waste with accurate distance (83.7-95.6%) and direction (84.7-97.3 %) information. The robot autonomously navigates towards detected waste, halting at a predefined distance for collection and disposal into a designated bin. This work contributes to advancements in waste management systems using small robots.

Nowadays, ensuring road safety is a crucial issue that demands continuous development and measures to minimize the risk of accidents. This paper presents the development of a driver fatigue detection method based on the analysis of facial images. To monitor the driver's condition in real-time, a video camera was used. The method of detection is based on analyzing facial features related to the mouth area and eyes, such as the frequency of blinking and yawning, mouth aspect ratio (MAR), and the duration of eye closure. The method was implemented in Python using a convolutional neural network (CNN). To validate the method, a dataset was created containing eye images that were subjected to various modifications, including the use of corrective glasses. The model's results confirm the method's effectiveness in detecting fatigue, achieving an average accuracy of 92% for eye detection and 82% for yawning detection under well-lit conditions.

We are increasingly striving to introduce modern artificial intelligence techniques in medicine and elevate medical care, catering to both patients and specialists. An essential aspect that warrants concurrent development is the protection of personal data, especially with technology's advancement, along with addressing data disparities to ensure model efficacy. This study assesses various domain adaptation techniques and federated learning to determine optimal integration strategies for enhanced security and the challenges posed by diverse datasets. Experiments utilized deep learning models, three domain adaptation methods, and a federated learning framework, focusing on mammography imaging for breast cancer detection. Results indicate a notable improvement of up to 20% with domain adaptation and an additional 10% with federated learning integration.

Recent inpainting methods have demonstrated im-pressive outcomes in filling missing parts of images, especially for reconstructing facial areas obscured by occlusions. However, studies show that these models are not adequately effective in real-world applications, primarily due to data bias and the distribution of faces in images. This research focuses on domain adaptation of the commonly used Labeled Faces in the Wild (LFW) dataset, employing the Mask-Aware Transformer (MAT) inpainting method for reconstructing occluded facial regions and examining its impact on facial recognition accuracy. Three types of generated masks were applied to specific facial areas, covering key points on the face, using three datasets: CelebA-HQ, LFW, and a specially adapted LFW. The analysis employed various metrics to assess the quality of the reconstruction. The results indicate that applying a simple adaptation method to the LFW dataset significantly boosts facial recognition capabilities, with improvements reaching up to 16.43% compared to the original LFW. Subsequently, the experiments demonstrate that using inpainting methods enhances face recognition considerably when compared to images with applied masks without reconstruction. Notably, improvements in positively verified images were ob-served up to 89.30% for CelebA-HQ, 21.37% for LFW, and 29.69% for the adapted LFW.

Over one million people die in car accidents worldwide each year. A solution that will be able to reduce situations in which pedestrian safety is at risk has been sought for a long time. One of the techniques for detecting pedestrians on the road is the use of artificial intelligence in connection with thermal imaging. The purpose of this work was to design a system to assist the safety of people and car intelligence with the use of automatic detection of pedestrians in low-resolution thermal image sequences. The data acquisition system was designed and used to collect thermal images for the needs of training of machine learning methods. The created new dataset consists of 9178 annotated, low-resolution images of pedestrians in different traffic conditions. Several deep, object detection models were adapted and trained using the new dataset together with public datasets. The best model turned out to be the adapted Faster R-CNN ResNet50 FPN (Faster Region-based Convolutional, Neural Networks Residual network50, Feature Pyramid Network) model with mean Average Precision (mAP) equal to 94.00%. It was also shown that the use of transfer learning based on the features learned from the RGB images results in mAP greater than 85.00% for all investigated algorithms. The designed system finds practical application in increasing road safety through the potential use of autonomous cars and city monitoring.

Efficient pedestrian detection is a very important task in ensuring safety within road conditions, especially after sunset. One way to achieve this goal is to use thermal imaging in conjunction with deep learning methods and an annotated dataset for models training. In this work, such a dataset has been created by capturing thermal images of pedestrians in different weather and traffic conditions. All images were manually annotated with bounding boxes. As a result, the created dataset consists of 9178 annotated, thermal images that can be used in many applications including nighttime pedestrian detection.

The rate of progress in the field of Artificial Intelligence (AI) and Machine Learning (ML) has significantly increased over the past ten years and continues to accelerate. Since then, AI has made the leap from research case studies to real production ready applications. The significance of this growth cannot be undermined as it catalyzed the very nature of computing. Conventional platforms struggle to achieve greater performance and efficiency, what causes a surging demand for innovative AI accelerators, specialized platforms and purpose-built computes. At the same time, it is required to provide solutions for assessment of ML platform performance in a reproducible and unbiased manner to be able to provide a fair comparison of different products. This is especially valid for Human System Interaction (HSI) systems that require specific data handling for low latency responses in emergency situations or to improve user experience, as well as for preserving data privacy and security by processing it locally. Taking it into account, this work presents a comprehensive guideline on preferred benchmarking criteria for evaluation of ML platforms that include both lower level analysis of ML models and system-level evaluation of the entire pipeline. In addition, we propose a Systematic Taxonomy of Embedded Platforms (STEP) that can be used by the community and customers for better selection of specific ML hardware consistent with their needs for better design of ML-based HSI solutions.

In this paper an investigation to determine the optimal placement of IMU sensors for the purpose of children characteristic activity detection is presented. The article compares four different placement of two IMU sensors on human body. Ten healthy volunteers participated within the study. Data were collected firstly from two wireless 9-axial IMU sensors placed at the left and right wrists, then sensors were placed at lower back and hip (dominant hand side). Activities included jumping, rotating, walking, walking on tiptoe, running, clapping hands, standing still, sitting still and dancing. Several parameters such as mean, standard deviation, skewness, kurtosis, energy, correlations, Hjorth parameters (activity, mobility and complexity) and spectra purity index, were calculated from measured data. Data from all locations provided similar levels of accuracy in differentiate analyzed activities.

The industrial advancement of human society has been fundamentally driven by diverse ‘systems’ that facilitate ‘human interaction’ within physical, digital, virtual, social and artificial environments, and upon the hyper-connected layers of system-system interactions across these environments. The research and practice of Human System Interaction (HSI) has undergone exponential development due to the enhanced capabilities, increased efficiencies and decreased costs of digitalization. Primarily driven by its unique capacity for information persistence, digitalization is now leading us into a nexus of transition where HSI is being transformed by Artificial Intelligence (AI). AI has leveraged the data and information amassed by digitalization to learn, reason, predict, optimize and thereby augment both human-system interaction and system-system interaction, within and across all hyper-connected environments noted above. In this paper, we review this evolution of HSI and contribute towards its future directions by articulating the AI transformation strategy for this nexus of transition into a Human-AI-System Interaction. The paper begins with a review of HSI that focuses on developments in the past 15 years, followed by the AI transformation strategy which comprises of the primary configurations for Human-AI-System Interaction, the current capabilities of AI, a lifecycle approach for the design, development and deployment of an AI solution and the ethical implications of AI in HSI.

In this paper, we summarize the results obtained from fire experiments. The aim of the work was to develop new methods of fire detection using IR thermal imaging cameras and dedicated image processing. We conducted 4 experiments in different configurations and with the use of different objects. The conducted experiments have shown the great usefulness of infrared cameras for detecting the seeds of a fire. Even cheap low-resolution bolometric detector modules can detect hot spots.

Increased interest in non-contact evaluation of the health state has led to higher expectations for delivering automated and reliable solutions that can be conveniently used during daily activities. Although some solutions for cough detection exist, they suffer from a series of limitations. Some of them rely on gesture or body pose recognition, which might not be possible in cases of occlusions, closer camera distances or impediments that prevent users from performing such movements at all. Others focus on analyzing breath using audio recordings, which cannot be easily applied in crowded or loud spaces. Many of them utilize visible light data which is prone to changing lighting conditions and can lead to various privacy concerns. Taking these into account, we propose to make use of the temporal pixel value changes occurring within specific facial areas. Due to the use of a combination of object detection and signal classification models, our system allows for fully automated classification of breathing anomalies. The benchmark evaluation performed on the newly created thermal cough data set proved the reliability of the introduced solution (precision of cough detection equals 94%). Due to the use of a lightweight deep learning model, the proposed system also has huge practical value, as it can potentially be deployed on edge devices frequently sought out in markets such as autonomous vehicles, drones, smart home or military applications.

In this paper the simple architecture of neural network for hand gesture classification was presented. The network classifies the previously calculated parameters of EMG signals. The main goal of this project was to develop simple solution that is not computationally complex and can be implemented on microprocessors in low-cost 3D printed prosthetic arms. As the part of conducted research the data set EMG signals corresponding to 5 different gestures was created. The accuracy of elaborated solution was 90% when applied real time on data sampled with 1kHz frequency and 75% when applied real time on data acquired and process directly on microprocessor with lower,100Hz sampling frequency.

In this paper the main aspects of mechanical design behind the low-cost prosthetic arm are presented. The fundamentals of a proper design has been defined to obtain functional 3D printed 5 degree of freedom (DOF) prosthesis. The designed prosthetic arm is a part of the hybrid interface with eye tracking movement control. The main focus was to create affordable but usable prosthesis which corresponds in size and weights to the human arm. The iterative process (starting from the final segment of the arm) was used to design fully functioning arm. All the elements were evaluated regarding total weight and the maximum load that can be carried by the arm. The result of this work is a prototype that weighs below 6kg and has a range of motion comparable to the human's arm. Final product is able to freely move an object of a total weight of 1 kg. All the mechanical parts of the designed arm were 3D printed which therefore presented construction can be adopted by people with different disabilities and (when connected to interfaces like EEG, EMG or eye tracking) provide support in everyday life activities.

In the era of collecting large amounts of tissue materials, assisting the work of histopathologists with various electronic and information IT tools is an undeniable fact. The traditional interaction between a human pathologist and the glass slide is changing to interaction between an AI pathologist with a whole slide images. One of the important tasks is the segmentation of objects (e.g. cells) in such images. In this study, we apply U-net and V-net convolutional neural network models to perform image segmentation. In particular, we analyze the role of the contour thickness in the reference (labels, masks) images on the results of image segmentation, also for the degraded images. We show the role of the proper mask definition and the results obtained for the ensemble models that use the same architecture but are trained using two sets of inverted masks.

This paper describes the possibility of monitoring the progress of knowledge and skills acquisition by the students of radiology. It is achieved by an analysis of a visual attention distribution patterns during image-based tasks solving. The concept is to use the eye-tracking data to recognize the way how the radiographic images are read by recognized experts, radiography residents involved in the training program, and untrained users who graduated from biomedical engineering. The results of research presented in this paper support the usefulness of earlier elaborated eye metrics, Visual Attention Time Product VATP, for skills measurement when using Electronic Medical Record for CT images evaluation. Moreover, it appears that it is possible to differentiate the level of residents skills in radiology structures recognition. Presented methodology can also be utilized in other areas where operator performs image related tasks.

Automation or even computer assistance of the popular infertility treatment method: ICSI (Intracytoplasmic Sperm Injection) would speed up the whole process and improve the control of the results. This paper introduces a preliminary research for automatic spermatozoon injection into the oocyte cytoplasm. Here, the method for detection a correct orientation of the polar body of the oocyte is presented. Proposed method uses deep learning U-Net architecture for object segmentation. This solution proved to be universal and had no demand for numerous dataset or high-quality Images.

Human-system interactions frequently require a retrieval of the key context information about the user and the environment. Image processing techniques have been widely applied in this area, providing details about recognized objects, people and actions. Considering remote diagnostics solutions, e.g. non-contact vital signs estimation and smart home monitoring systems that utilize person's identity, security is a very important factor. Thus, thermal imaging has become more and more popular, as it does not reveal features that are often used for person recognition, i.e. sharp edges, clear changes of pixel values between areas, etc. On the other hand, there are much more visible light data available for deep model training. Taking it into account, person recognition from thermography is much more challenging due to specific characteristics (blurring and smooth representation of features) and small amount of training data. Moreover, when low resolution data is used, features become even less visible, so this problem may become more difficult. This study focuses on verifying whether model trained to extract important facial embedding from RGB images can perform equally well if applied to thermal domain, without additional re-training. We also perform a set of experiments aim at evaluating the influence of resolution degradation by down-scaling images on the recognition accuracy. In addition, we present deep super-resolution (SR) model that by enhancing donw-scaled images can improve results for data acquired in scenarios that simulate real-life environment, i.e. mimicking facial expressions and performing head motions. Preliminary results proved that in such cases SR helps to increase accuracy by 6.5% for data 8 times smaller than original images. It has also been shown that it is possible to accurately recognize even 40 volunteers using only 4 images per person as a reference embedding. Thus, the initial profiles can be easily created in a real time, what is an additional advantage considering a solution setup in a new environment.

This paper focuses on convolution neural network quantization problem. The quantization has a distinct stage of data conversion from floating-point into integer-point numbers. In general, the process of quantization is associated with the reduction of the matrix dimension via limited precision of the numbers. However, the training and inference stages of deep learning neural network are limited by the space of the memory and a variety of factors including programming complexity and even reliability of the system. On the whole the process of quantization becomes more and more popular due to significant impact on performance and minimal accuracy loss. Various techniques for networks quantization have been already proposed, including quantization aware training and integer arithmetic-only inference. Yet, a detailed comparison of various quantization configurations, combining all proposed methods haven't been presented yet. This comparison is important to understand selection of quantization hyperparameters during training to optimize networks for inference while preserving their robustness. In this work, we perform in-depth analysis of parameters in the quantization aware training, the process of simulating precision loss in the forward pass by quantizing and dequantizing tensors. Specifically, we modify rounding modes, input preprocessing, output data signedness, bitwidth of the quantization and locations of precision loss simulation to evaluate how they affect accuracy of deep neural network aimed at performing efficient calculations on resource-constrained devices.

In this paper the implementation of recurrent neural network models for hand gesture recognition on edge devices was performed. The models were trained with 27 hand gestures recorded with the use of a linear optical sensor consisting of 8 photodiodes and 4 LEDs. Different models, trained off-line, were tested in terms of inference time on several mobile devices. The impact of different network topologies (different number of neurons and layers) and different effective sampling frequency of a recorded gesture on the inference time were evaluated. Inference time tests performed on the most effective of available platforms, Samsung Galaxy S8 smartphone, present that recurrent neural networks trained on unprocessed (raw) data from the gesture sensor (1-layer model) execute faster than models trained on differently processed features (2 and 3-layer models): 374ms vs 765ms and 1048ms for data recorded with 100Hz. Also, the reduction of effective sampling frequency allows to shorten the inference time, e.g., for raw data from 374ms at 100Hz to 115ms at 25Hz. Presented results confirm the usability of low complexity interfaces like linear gesture sensor in mobile devices with limited computational capabilities, which allows to consider such sensors in wearable electronic devices e.g. smart glasses.

In this paper we have investigated the capabilities of a quadrant photodiode based gesture sensor in the recognition of digits drawn in the air. The sensor consisting of 4 active elements, 4 LEDs and a pinhole was considered as input interface for both discrete and continuous gestures. Index finger and a round pointer were used as navigating mediums for the sensor. Experiments performed with 5 volunteers allowed to record 300 examples of each digit from 0 to 9, which were drawn in the air. Digits were converted from a list of recorded coordinates into images processed as in the MNIST database. Three approaches for recognition of digits recorded by quadrant photodiode were considered: convolutional neural network trained only on examples from the MNIST database, network trained on mixed data of MNIST with examples recorded using quadrant photodiode (4/1 proportions) and trained on the MNIST with examples recorded using the elaborated sensor but after the arbitral rejection of 20% of worst quality data (4/1 proportions preserved). The application of the third approach in comparison to the first one allowed to increase the overall accuracy of digits classification from 34.4% to 86% for testing data recorded with the use of the pointer and from 32% to 81.2% for data recorded with the use of a finger (for 50Hz sampling frequency).

The purpose of this paper is to perform an analysis of the accuracy of the pulse estimation by comparing readings from a smartwatch with readings from medical devices. The study required writing applications that allow continuous pulse measurement. As a result, two applications were created for the smartwatch. The first one is dedicated to Android Wear devices, while the other one is compatible with Tizen watches. The next step was to create another application using a phone with a camera and a flashlight. The current heart rate value could be obtained by putting a finger to the lens of the camera and its flashlight, followed by a suitable pulse calculation algorithm. At the end of the experiment, a short and non-obligatory form for the gender, age and cardiovascular diseases of the person could be filled and the results are recorded in the database in a way that they could be compared with the measurements from watches. After collecting the intended amount of measurements, it was necessary to send a measurement file to a computer that could calculate the heart rate estimations. Obtaining these values has allowed to draw conclusions about the accuracy and reliability of pulse measurement using a mobile devices.

Gesture-based intuitive interactions with electronic devices can be an important part of smart home systems. In this paper, we adapt the contactless linear gesture sensor for the navigation of smart lighting system. Set of handled gestures allow to propose two methods of active light source selection, continuous dimming, and turning on and off based on discrete gestures. The average gesture recognition accuracy was 97.58% in the first of proposed methods of lamp selection and 95.45% in the second method. Apart of the accuracy of the gesture sensor in the real case implementation the user convenience of navigation was monitored. 72.73% of users pointed the first method as more intuitive and convenient to use. 63.64% of users indicated that the dimming of light is more intuitive when moving a hand along the sensor than closing it towards the package. The impact of utilizing dirty or gloved hand on the pose recognition accuracy of the sensor was verified as well. The results indicate the capability of incorporating considered touchless gesture sensor to homes, factories or operating rooms. These are places where the need of the functionality of discrete and continuous gestures for handling some basic devices or functions can be limited due to dirty hand scenarios or sterility requirements.

The aim of this work was to evaluate the usability of low-cost and low-power components for road condition classification and pedestrian detection in challenging environments. Therefore, deep learning based system dedicated for resource constrained environments is proposed to classify possible road hazards. In the performed experiments we investigated the influence of various factors (lightning conditions, moisture of the road surface and ambient temperature) on the system ability to accurately detect a person on the road. The implemented system was tested on images acquired in different climate zones. Preliminary results show that CNN models with optimized number of parameters can classify thermal road conditions with precision and recall above 70% in most cases, with exception of differentiating lightning conditions. This, in turn, proved system insensitivity to the poor lightning on the road and allows for evaluating road hazards regardless of the time of the day. Pedestrian detection using the model optimized for mobile environments proved to be highly efficient (precision and recall above 95%) both when ambient temperature is lower and higher than human body temperature. System was deployed on Movidius Neural Compute Stick, that allowed for increasing processing throughput (whole pipeline took 0.287±0.028s to process) while conserving energy, what indicates suitability of the solution for implementation in real-time automotive systems.

Gesture sensors for mobile devices, which have a capability of distinguishing hand poses, require efficient and accurate classifiers in order to recognize gestures based on the sequences of primitives. Two methods of poses recognition for the optical linear sensor were proposed and validated. The Gaussian distribution fitting and Artificial Neural Network based methods represent two kinds of classification approaches. Three types of hand poses, differing in the number of fingers joined together, were investigated. The reflected light intensity pattern originated by hand located closely to the sensor was parameterized into 14 features. The change of reflection pattern originated by hand dislocation was reduced by application of two variable functions in the first of the methods. A one and two hidden layers topologies were considered in the neural network related approach. Both methods were designed with the use of a training set of samples and validated with another (testing) set. The results present the average poses recognition rate of 81.19% for Gaussian distribution fitting and 90.02% for ANN based method.

This paper presents 2D image processing approach to playback detection in automatic speaker verification (ASV) systems using spectrograms as speech signal representation. Three feature extraction and classification methods: histograms of oriented gradients (HOG) with support vector machines (SVM), HAAR wavelets with AdaBoost classifier and deep convolutional neural networks (CNN) were compared on different data partitions in respect of speakers or playback devices: for instance with different speakers in training and test subsets. The playback detection systems were trained and tested on two speech datasets S1and S2manufactured independently by two different institutions. The test error for both datasets oscillates about the level of 1% for HOG+SVM and even below it for CNN in bigger S1base. In cross validation scenario in which one base was used for training and second base for the test the results were very poor what suggests that the information relevant for playback detection appeared in each base in different way.

Welcome to HSI2017, the 10th International Conference on Human System Interactions in 2017 was held at the University of Ulsan in Ulsan, Republic of Korea. The University of Ulsan have organized the conference and the conference is technically co-sponsored by IEEE Industrial Electronics Society. HSI conference series has been one of the most important academic meetings in the field of interactions between human and systems. Until now the HSI conference series have been held in Krakow (Poland) 2008, Catania (Italy) 2009, Rzeszow (Poland) 2010, Yokohama (Japan) 2011, Perth (Australia) 2012, Gdansk (Poland) 2013, Lisbon (Portugal) 2014, Warsaw (Poland) 2015, and Portsmouth (United Kingdom) 2016.

In this paper we propose to use temporal muscle contraction to perform certain actions. Method: The set of muscle contractions corresponding to one of three actions including “single-click”, “double-click” “click-n-hold” and “non-action” were recorded. After recording certain amount of signals, the set of five parameters was calculated. These parameters served as an input matrix for the neural network. Two-layer feedforward neural network with one hidden layer of 200 neurons was applied to classify gestures based on the input matrix. Results: The network was trained using the dataset consisted of 43 samples and then tested on the 34 samples dataset. All gestures from the test set were correctly classified.

Nearly 10% of all upper limb amputations concern the whole arm. It affects the mobility and reduces the productivity of such a person. These two factors can be restored by using prosthetics. However, the complexity of human arm makes restoring its basic functions quite difficult. When the osseointegration and/or targeted muscle reinnervation (TMR) are not possible, different modalities can be used to control the prosthesis. In this paper the usability of electromyography (EMG) signals for such a control is evaluated. Method: first, the types of operations performed by the prosthetic arm that could be handled by EMG module were defined. The raw EMG signal, corresponding to the predefined gesture, was acquired from the surface of trapezius muscle. The pattern recognition neural network was trained to classify gestures based on recorded RAW data. Results: The neural network was trained using 56 signals corresponding to performed gestures. Optimal performance was achieved for 29 training cycles. The network was tested using data set of 56 gestures. The designed network was tested on gestures recorded from 10 volunteers. The gestures were correctly classified with nearly 84% accuracy. Conclusions: The EMG analysis is a reliable modality when it comes to hybrid interfaces for control over prosthetic arm.

Recently, capabilities of many computer vision tasks have significantly improved due to advances in Convolutional Neural Networks. In our research, we demonstrate that it can be also used for face detection from low resolution thermal images, acquired with a portable camera. The physical size of the camera used in our research allows for embedding it in a wearable device or indoor remote monitoring solution for elderly and disabled people. The benefits of the proposed architecture were experimentally verified on the thermal video sequences, acquired in various scenarios to address possible limitations of remote diagnostics: movements of the person performing a diagnose and movements of the examined person. The achieved short processing time (42.05±0.21ms) along with high model accuracy (false positives -0.43%; true positives for the patient focused on a certain task -89.2%) clearly indicates that the current state of the art in the area of image classification and face tracking in thermography was significantly outperformed.

This paper presents the concept of visualization of amber works in a virtual reality cave. It describes problems encountered during the acquisition, modeling and rendering of geometrical objects distinguished by heterogeneous transparency.

The paper presents an alternative approach to the monitoring of the cardiovascular system.The study depicts configurations of the utilized system and preliminary results of electrical and mechanical parameters of the cardiac system which can be measured using a head-worn device.

In this paper we describe possibility of use the RSSI signal (Radio Signal Strength Indication) from Texas Instruments SensorTag CC2650 for indoor positioning purposes.This idea is not a new but in our opinion it is possible to use SensorTags with Bluetooth LE wireless interface for positioning inside buildings in such applications as people findings in hospitals, senior come care, etc. RSSI is mostly selected as the sensor localization method in the indoor circumstances.In this paper, we aim to analyze accuracy, calibrate and map RSSI to distance by doing a series of the experiments.Obtained results are very promising and shows possibility of use this technique for position estimation.

A technique for the acquisition of an increased number of pupil positions, using a combined sensor consisting of a low-rate camera and a high-rate optical sensor, is presented in this paper.The additional data are provided by the optical movement-detection sensor mounted in close proximity to the eyeball.This proposed solution enables a significant increase in the number of registered fixation points and saccades and can be used in wearable electronics applications where low consumption of resources is required.The results of the experiments conducted here show that the proposed sensor system gives comparable results to those acquired from a high-speed camera and can also be used in the reduction of artefacts in the output signal.

Respiration rate is a very important vital sign.Different methods of respiration rate measurement or estimation have been developed.However, especially interesting are those that enable remote and unobtrusive monitoring.In this study, we investigated the use of smart glasses for the estimation of respiration rate especially useful for indoors applications.Two methods were analyzed.The first one is based on measurements of respiration-related body movements using an accelerometer.The second one uses the thermal camera to observe temperature changes in the nostril region.For both methods signals were extracted, filtered and processed using two different respiration rate estimators.Both methods were validated during experiments with the participation of volunteers using the respiration belt as a reference measurement method.Results proved that for both methods it is possible to reliable estimate the respiration rate with Root Mean Square Error lower than 2 breaths per minute, which is sufficient for medical screening.

In this preliminary study we analyzed the possibility of the reliable measurement of biomedical signals with some potential hardware extensions of smart glasses. Using specially designed experimental prototypes four category of biomedical signals were measured: electrocardiograms, electromyograms, electroencephalograms and respiration waveforms. Experiments with volunteers proved that using even simple construction of sensors it is possible to reliable measure biomedical signals with the quality enough for screening purposes as for the needs of simple interaction between an user and smart glasses.

In this paper we deal with the problem of precise gesture recognition for the active optical proximity sensor with sparse 8 photodiodes. We particularly focus on developing the method of estimating the real, usually not observable, maximum signal value representing maximum intensity of light reflected from an obstacle present in the front of the sensor. Different configurations of the fingers were used as an obstacle. The Monte Carlo simulations were performed in order to cognize the accurate pattern created by fingers of considered configurations. The accurate description of curves representing finger profiles was obtained after applying the least squares method. The results of analyzes proved that finger arrangement configurations, particularly 1 finger and 2 jointed fingers, can be considered as sources of Gaussian like shapes on the face of the sensor. Therefore, data measured by 8 sparse photodiodes of the active optical gesture sensor were compared to normal distribution curves. The results of the comparison allow normalization of the measured data and the estimation of the real, usually not observable, maximum signal value. The results and discussion sections of the paper present the average detection accuracy, advantages and limitations of the proposed approach.

In this paper we investigate the effectiveness of the interaction using eye tracking and electromyography. Smart glasses requires reliable interfaces for controlling the graphic content displayed directly in front of the user's eye. Presented research is related with the eGlasses project, which is focused on the development of an open platform in the form of multisensory electronic glasses and related interaction methods. One of the implemented interaction methods is the one based on eye tracking module. The eye tracking provides quite accurate pointing but there is always a matter of how to select and confirm while using the eye tracking. In this paper we combine the eye tracking with the EMG signal acquired from the skeletal muscles. Both, the advantages and limitation of the method are discussed.

In this paper we present the possibility of using eGlasses eye tracking module as an extension for Scratch programming tool which is a visual programming language supporting computer skills learning. The main concept behind this project is to setup the interface for rapid interaction design. Eye tracking is a powerful tool for hands free communication but for that requires a dedicated software. This software is rarely tailored for a specific needs of a potential user. It means that people who wants to explore gaze driven computer interfaces often need to create their own eye tracking software or need to have certain computer skills to be able to go through the software development kit (SDK) provided by the eye tracker vendors. It makes designing interaction by gaze quite challenging. The eGlasses eye tracking module was used to extend sensor values available in Scratch by the information about pupil position, gaze position and eye opened/eye closed signal. In this paper we have tested if the gaze data acquired by the eGlasses are reliable and can be used as an input signal for rapid interaction design using Scratch. The whole interface was validated regarding the system usability score (SUS). The purpose of our work is to support disabled children and their therapists but also to give a chance to a wider community, without computational skills, to easily check they ideas for gaze based interaction. The results of conducted studies with the participation of volunteers give strong foundations to support this statement.

Recently, facial feature tracking systems have become more and more popular because of many possible use cases. Especially in medical applications location of the face and facial features are very useful. Many researches have presented methods to detect and track facial features in visible light. However, facial feature analysis in thermography may also be very advantageous. Some examples of using infrared imagery in medicine include the estimation of the respiration rate using an analysis of temperature changes in the area below nose region. Moreover, due to technological development small thermal cameras may be embedded into wearable devices, like smart glasses and used to support remote patient monitoring. Therefore, in this paper, we focused on face tracking in low quality thermal images. Especially, we compared four interest points detectors for facial feature tracking in thermal images. All methods were tested for processing time, displacement of detected areas and errors of calculated mean value of pixel intensities in the detected nose region. Finally, we presented a fully automatic system for facial features tracking, which allows to process one frame in about 27.7ms (Harris), 23.9ms (ORB), 19.7ms (SIFT), 27.6ms (SURF) with acceptable accuracy (Harris - 7.2±4.3%, ORB 9.9±2.2%, SIFT 7.0±1.9%, SURF 8.9±2.7%).

The goal of this paper is to analyze the accuracy of pulse rate estimation from the sequence of face images. Simulated and real signals were used to evaluate two pulse rate estimators; one for frequency domain and the second one for time domain using the autocorrelation function. The results show that the mean difference between the reference measurements and estimated pulse rate values are about 2bpm. In the analysis of short signals the estimator working in time domain has better accuracy since it is not limited by spectral leakage observed in the analysis of spectrum for short length signals.

In this study we evaluated system architecture for the use of smart glasses as a viewer of information, as a source of medical data (vital sign measurements: temperature, pulse rate, and respiration rate), and as a filter of healthcare information. All activities were based on patient/device identification procedures using graphical markers or features based on visual appearance. The architecture and particular use cases were implemented and verified using smart glasses prototypes developed under the eGlasses project and using a reference Health Level 7 Fast Healthcare Interoperability Resources (HL7 FHIR) server. The results show that information about the identified patient can be quickly retrieved from FHIR servers and annotated using voice recognition services. Smart glasses can be used in the measurement of vital signs of the observed patient, providing values of body temperature, pulse rate, and respiration rate by means of non-contact measurements. Such measurements are sufficiently reliable for medical screening and for fast data exchange using HL7 FHIR actions.

Two methods of object position and movement estimation in relation to the user of smart glasses were investigated. An active infrared and ultrasonic methods of the obstacle detection were presented and compared. Application of these methods depend on active transducers type (physical medium used), geometry and surface properties of detected objects and their movement direction and speed. In the article properties of both detectors were compared and applicability of both methods in mobile, battery operated environment such as eGlasses platform were compared.

In this paper we evaluated the possible application of a passive, optical sensor as an interface for human-smart glasses interactions. The designed proximity sensor is composed of set of photodiodes and the appropriate hardware and software components. First, experiments were performed for the estimations of such parameters as distance to an object, its width and velocity. Achieved results were satisfactory. Therefore, next, a set of static and dynamic hand gestures was proposed together with the related recognition method. The method was verified during experiments with volunteers. Experiments have shown that in appropriate lighting conditions it is possible to detect static and dynamic simple hand gestures using inexpensive and computationally efficient sensor. These features enable the use of such a sensor as one of the smart glasses interfaces.

In this paper we investigate the effectiveness of the gaze interaction within near-to-eye display. The practical aspect of the paper is about combining an eye tracker with smart glasses. Presented research is related with the eGlasses project, which is focused on the development of an open platform in the form of multisensory electronic glasses and related interaction methods. One of the implemented interaction methods is the one based on eye tracking module. Both, the advantages and limitation of the method are discussed. We also considering the calibration free method of fixation points estimation within the near-to-eye display.

In this paper we propose to cast the problem of identification of people, objects or places into an application for smart glasses that decodes information from graphical markers. We focus on analyzing different factors that can have influence on the processes of the automatic recognition of information from a code. The research we present aims at reviewing recognition performances in function of: size of a marker, distance from/to a marker, position of a marker and hardware used to decode information. Additionally, we describe a colorful graphical marker which was created for comparative analysis with existing monochrome codes. We also analyze accuracy of the detection of markers colors under different illumination conditions and different scanning distances. Moreover, considering possible usages of graphical markers, we present a prototype application, which can potentially provide faster access to patients' information. We believe that it may be very useful, because health care have to deal with a lot of various assets that could be identify or classify by using eyewear devices. Our first conclusions are the facts that following factors may have influence on monochrome code recognition (i) distance from the camera, (ii) rotation and slope of code, (iii) quality and size of the code. Distance from the camera should be adapted to the device that is used to scan the code. Under preliminary studies with colorful codes, tested light conditions and distances did not affect the classification of colors.

Recently, different smart glasses solutions have been proposed on the market. The rapid development of this wearable technology has led to several research projects related to applications of smart glasses in healthcare. In this paper we propose a general architecture of the system enabling data integration for the recognized person. In the proposed system smart glasses integrates data obtained for the recognized patient from health care information systems, from devices connected to the patient and from the patient himself. The data integration is possible, if proper patient recognition procedure is used. Therefore, we evaluated three identification methods based on face recognition and using the recognition of graphical markers (i.e. QR-codes and proposed color-based codes). The results show that it is possible to obtain reliable and fast recognition results during the video acquisition by the smart glasses camera.

Bar-codes and QR-codes (Quick Response) are often used in healthcare. In this paper an application of QR-codes to exchange of laboratory results is presented. The secure data exchange is proposed between a laboratory and a patient and between a patient and Electronic Health Records. Advanced Encryption Standard was used to provide security of data encapsulated within a QR-code. The experimental setup, named labSeq is described. Additionally, a mobile application which enables decoding and storage of the enciphered information is presented. This work is also a preliminary research in supplying electronic glasses device, eGlasses, a functionality to process data from graphical patterns. In this paper a mobile phone is used as a simulator of electronic glasses. The exchange of data between the mobile application and Electronic Health Records is based on Health Level 7 standards or using JavaScript Object Notation. This enables fast integration of data between a patient's device and the information system used by a healthcare professional.

An influence of head movements on the gaze estimation accuracy when using a head mounted eye tracking system is discussed in the paper. This issue has been examined for a multi-display environment. It was found that head movement (rotation) to some extent does not influence on the gaze estimation accuracy seriously. Acceptable results were obtained when using eye-tracker to communicate with a computer via in two displays simultaneously.

In this paper we present a new human-computer interface for the quantitative measurement of blowing activities. The interface can measure the air flow and air pressure during the blowing activity. The measured values are stored and used to control the state of the graphical objects in the graphical user interface. In speech therapy children will find easier to play attractive therapeutic games than to perform repetitive and tedious, traditional blowing actions. Therefore, we also propose a set of games that can be used together with the blowing device. The paper focuses on the design of the interface, preliminary experiments and results. Additionally, fours blowing games are described.

Implicit interaction combined with object recognition techniques opens a new possibility for gathering data and analyzing user behavior for activity and context recognition. The electronic eyewear platform, eGlasses, is being developed, as an integrated and autonomous system to provide interactions with smart environment. In this paper we present a method for the interactions with the recognized objects that can be used for electronic eyewear. The design of the control node is presented. The control node is the extension of the power socket, equipped with a set of sensors and ZigBee communication interface. The node is used to control electronic objects through user interactions with a mobile device. The preliminary results are presented based on the Android mobile phone used to recognize controllable objects based on the captured image sequences.

Diagnosis of obstructive sleep apnea usually involves polysomnographic analysis, which unfortunately requires overnight stay in a specialized clinic and is very uncomfortable for a patient. This paper describes the method and apparatus for recording a set of signals to detect sleep apnea. The device records the following signals simultaneously: three-channel ECG, respiratory functions, signals from the accelerometer, and snoring sounds. Measurements are carried out during the night, in the patient's home. Algorithm for automatic analysis of the data can be used to detect sleep apnea and hypopnea.

An example of an infra-red remote control device is presented in this paper. The device is equipped with additional medical diagnostic features for physiological and mental training. It is primarily a diagnostic health-care unit to support the elderly. To improve the acceptance of the device, the additional features are hidden in the shape of an ordinary, well-known device.