Search

In this study, an action recognition system was developed to identify fundamental basketball movements using a single Inertial Measurement Unit (IMU) sensor embedded in a wearable vest. This study aims to enhance basketball training by providing a high-performance, low-cost solution that minimizes discomfort for athletes. Data were collected from 21 collegiate basketball players, and movements such as dribbling, passing, shooting, layup, and standing still were recorded. The collected IMU data underwent preprocessing and feature extraction, followed by the application of machine learning algorithms including KNN, decision tree, Random Forest, AdaBoost, and XGBoost. Among these, the XGBoost algorithm with a window size of 250 and a 75% overlap yielded the highest accuracy of 96.6%. The system demonstrated superior performance compared to other single-sensor systems, achieving an overall classification accuracy of 96.9%. This research contributes to the field by presenting a new dataset of basketball movements, comparing the effectiveness of various feature extraction and machine learning methods, and offering a scalable, efficient, and accurate action recognition system for basketball. [ABSTRACT FROM AUTHOR]

Sensor-based Human Activity Recognition (HAR) is crucial in ubiquitous computing, analyzing behaviors through multi-dimensional observations. Despite research progress, HAR confronts challenges, particularly in data distribution assumptions. Most studies assume uniform data distributions across datasets, contrasting with the varied nature of practical sensor data in human activities. Addressing data heterogeneity issues can improve performance, reduce computational costs, and aid in developing personalized, adaptive models with fewer annotated data. This review investigates how machine learning addresses data heterogeneity in HAR by categorizing data heterogeneity types, applying corresponding suitable machine learning methods, summarizing available datasets, and discussing future challenges. [ABSTRACT FROM AUTHOR]

The work aims to leverage computer vision and artificial intelligence technologies to quantify key components in food distribution services. Specifically, it focuses on dish counting, content identification, and portion size estimation in a dining hall setting. An RGB camera is employed to capture the tray delivery process in a self-service restaurant, providing test images for plate counting and content identification algorithm comparison, using standard evaluation metrics. The approach utilized the YOLO architecture, a widely recognized deep learning model for object detection and computer vision. The model is trained on labeled image data, and its performance is assessed using a precision–recall curve at a confidence threshold of 0.5, achieving a mean average precision (mAP) of 0.873, indicating robust overall performance. The weight estimation procedure combines computer vision techniques to measure food volume using both RGB and depth cameras. Subsequently, density models specific to each food type are applied to estimate the detected food weight. The estimation model's parameters are calibrated through experiments that generate volume-to-weight conversion tables for different food items. Validation of the system was conducted using rice and chicken, yielding error margins of 5.07% and 3.75%, respectively, demonstrating the feasibility and accuracy of the proposed method. [ABSTRACT FROM AUTHOR]

System-to-system communication via Application Programming Interfaces (APIs) plays a pivotal role in the seamless interaction among software applications and systems for efficient and automated service delivery. APIs facilitate the exchange of data and functionalities across diverse platforms, enhancing operational efficiency and user experience. However, this also introduces potential vulnerabilities that attackers can exploit to compromise system security, highlighting the importance of identifying and mitigating associated security risks. By examining the weaknesses inherent in these APIs using security open-intelligence catalogues like CWE and CAPEC and implementing controls from NIST SP 800-53, organizations can significantly enhance their security posture, safeguarding their data and systems against potential threats. However, this task is challenging due to evolving threats and vulnerabilities. Additionally, it is challenging to analyse threats given the large volume of traffic generated from API calls. This work contributes to tackling this challenge and makes a novel contribution to managing threats within system-to-system communication through API calls. It introduces an integrated architecture that combines deep-learning models, i.e., ANN and MLP, for effective threat detection from large API call datasets. The identified threats are analysed to determine suitable mitigations for improving overall resilience. Furthermore, this work introduces transparency obligation practices for the entire AI life cycle, from dataset preprocessing to model performance evaluation, including data and methodological transparency and SHapley Additive exPlanations (SHAP) analysis, so that AI models are understandable by all user groups. The proposed methodology was validated through an experiment using the Windows PE Malware API dataset, achieving an average detection accuracy of 88%. The outcomes from the experiments are summarized to provide a list of key features, such as FindResourceExA and NtClose, which are linked with potential weaknesses and related threats, in order to identify accurate control actions to manage the threats. [ABSTRACT FROM AUTHOR]

Convolutional neural networks (CNNs) have made significant progress in the field of facial expression recognition (FER). However, due to challenges such as occlusion, lighting variations, and changes in head pose, facial expression recognition in real-world environments remains highly challenging. At the same time, methods solely based on CNN heavily rely on local spatial features, lack global information, and struggle to balance the relationship between computational complexity and recognition accuracy. Consequently, the CNN-based models still fall short in their ability to address FER adequately. To address these issues, we propose a lightweight facial expression recognition method based on a hybrid vision transformer. This method captures multi-scale facial features through an improved attention module, achieving richer feature integration, enhancing the network's perception of key facial expression regions, and improving feature extraction capabilities. Additionally, to further enhance the model's performance, we have designed the patch dropping (PD) module. This module aims to emulate the attention allocation mechanism of the human visual system for local features, guiding the network to focus on the most discriminative features, reducing the influence of irrelevant features, and intuitively lowering computational costs. Extensive experiments demonstrate that our approach significantly outperforms other methods, achieving an accuracy of 86.51% on RAF-DB and nearly 70% on FER2013, with a model size of only 3.64 MB. These results demonstrate that our method provides a new perspective for the field of facial expression recognition. [ABSTRACT FROM AUTHOR]

Human Activity Recognition (HAR) plays an important role in the automation of various tasks related to activity tracking in such areas as healthcare and eldercare (telerehabilitation, telemonitoring), security, ergonomics, entertainment (fitness, sports promotion, human–computer interaction, video games), and intelligent environments. This paper tackles the problem of real-time recognition and repetition counting of 12 types of exercises performed during athletic workouts. Our approach is based on the deep neural network model fed by the signal from a 9-axis motion sensor (IMU) placed on the chest. The model can be run on mobile platforms (iOS, Android). We discuss design requirements for the system and their impact on data collection protocols. We present architecture based on an encoder pretrained with contrastive learning. Compared to end-to-end training, the presented approach significantly improves the developed model's quality in terms of accuracy (F1 score, MAPE) and robustness (false-positive rate) during background activity. We make the AIDLAB-HAR dataset publicly available to encourage further research. [ABSTRACT FROM AUTHOR]

In a dynamic production processes, mechanical degradation poses a significant challenge, impacting product quality and process efficiency. This paper explores a novel approach for monitoring degradation in the context of viscose fiber production, a highly dynamic manufacturing process. Using causal discovery techniques, our method allows domain experts to incorporate background knowledge into the creation of causal graphs. Further, it enhances the interpretability and increases the ability to identify potential problems via changes in causal relations over time. The case study employs a comprehensive analysis of the viscose fiber production process within a prominent textile industry, emphasizing the advantages of causal discovery for monitoring degradation. The results are compared with state-of-the-art methods, which are not considered to be interpretable, specifically LSTM-based autoencoder, UnSupervised Anomaly Detection on Multivariate Time Series (USAD), and Deep Transformer Networks for Anomaly Detection in Multivariate Time Series Data (TranAD), showcasing the alignment and validation of our approach. This paper provides valuable information on degradation monitoring strategies, demonstrating the efficacy of causal discovery in dynamic manufacturing environments. The findings contribute to the evolving landscape of process optimization and quality control. [ABSTRACT FROM AUTHOR]

This paper addresses issues concerning biometric authentication based on handwritten signatures. Our research aimed to check whether a handwritten signature acquired with a mobile device can effectively verify a user's identity. We present a novel online signature verification method using coordinates of points and pressure values at each point collected with a mobile device. Convolutional neural networks are used for signature verification. In this paper, three neural network models are investigated, i.e., two self-made light SigNet and SigNetExt models and the VGG-16 model commonly used in image processing. The convolutional neural networks aim to determine whether the acquired signature sample matches the class declared by the signer. Thus, the scenario of closed set verification is performed. The effectiveness of our method was tested on signatures acquired with mobile phones. We used the subset of the multimodal database, MobiBits, that was captured using a custom-made application and consists of samples acquired from 53 people of diverse ages. The experimental results on accurate data demonstrate that developed architectures of deep neural networks can be successfully used for online handwritten signature verification. We achieved an equal error rate (EER) of 0.63% for random forgeries and 6.66% for skilled forgeries. [ABSTRACT FROM AUTHOR]

Direct observation is a ground-truth measure for physical behavior, but the high cost limits widespread use. The purpose of this study was to develop and test machine learning methods to recognize aspects of physical behavior and location from videos of human movement: Adults (N = 26, aged 18–59 y) were recorded in their natural environment for two, 2- to 3-h sessions. Trained research assistants annotated videos using commercially available software including the following taxonomies: (1) sedentary versus non-sedentary (two classes); (2) activity type (four classes: sedentary, walking, running, and mixed movement); and (3) activity intensity (four classes: sedentary, light, moderate, and vigorous). Four machine learning approaches were trained and evaluated for each taxonomy. Models were trained on 80% of the videos, validated on 10%, and final accuracy is reported on the remaining 10% of the videos not used in training. Overall accuracy was as follows: 87.4% for Taxonomy 1, 63.1% for Taxonomy 2, and 68.6% for Taxonomy 3. This study shows it is possible to use computer vision to annotate aspects of physical behavior, speeding up the time and reducing labor required for direct observation. Future research should test these machine learning models on larger, independent datasets and take advantage of analysis of video fragments, rather than individual still images. [ABSTRACT FROM AUTHOR]

Deep networks can facilitate the monitoring of a balanced diet to help prevent various health problems related to eating disorders. Large, diverse, and clean data are essential for learning these types of algorithms. Although data can be collected automatically, the data cleaning process is time-consuming. This study aims to provide the model with the ability to learn even when the data are not completely clean. For this purpose, we extend the Attentive Feature MixUp method to enable its learning on noisy multi-label food data. The extension was based on the hypothesis that during the MixUp phase, when a pair of images are mixed, the resulting soft labels should be different for each ingredient, being larger for ingredients that are mixed with the background because they are better distinguished than when they are mixed with other ingredients. Furthermore, to address data perturbation, the incorporation of the Laplace approximation as a post-hoc method was analyzed. The evaluation of the proposed method was performed on two food datasets, where a notable performance improvement was obtained in terms of Jaccard index and F 1 score, which validated the hypothesis raised. With the proposed MixUp, our method reduces the memorization of noisy multi-labels, thereby improving its performance. [ABSTRACT FROM AUTHOR]

Recent advances in wearable systems have made inertial sensors, such as accelerometers and gyroscopes, compact, lightweight, multimodal, low-cost, and highly accurate. Wearable inertial sensor-based multimodal human activity recognition (HAR) methods utilize the rich sensing data from embedded multimodal sensors to infer human activities. However, existing HAR approaches either rely on domain knowledge or fail to address the time-frequency dependencies of multimodal sensor signals. In this paper, we propose a novel method called deep wavelet convolutional neural networks (DWCNN) designed to learn features from the time-frequency domain and improve accuracy for multimodal HAR. DWCNN introduces a framework that combines continuous wavelet transforms (CWT) with enhanced deep convolutional neural networks (DCNN) to capture the dependencies of sensing signals in the time-frequency domain, thereby enhancing the feature representation ability for multiple wearable inertial sensor-based HAR tasks. Within the CWT, we further propose an algorithm to estimate the wavelet scale parameter. This helps enhance the performance of CWT when computing the time-frequency representation of the input signals. The output of the CWT then serves as input for the proposed DCNN, which consists of residual blocks for extracting features from different modalities and attention blocks for fusing these features of multimodal signals. We conducted extensive experiments on five benchmark HAR datasets: WISDM, UCI-HAR, Heterogeneous, PAMAP2, and UniMiB SHAR. The experimental results demonstrate the superior performance of the proposed model over existing competitors. [ABSTRACT FROM AUTHOR]

Machine learning with deep neural networks (DNNs) is widely used for human activity recognition (HAR) to automatically learn features, identify and analyze activities, and to produce a consequential outcome in numerous applications. However, learning robust features requires an enormous number of labeled data. Therefore, implementing a DNN either requires creating a large dataset or needs to use the pre-trained models on different datasets. Multitask learning (MTL) is a machine learning paradigm where a model is trained to perform multiple tasks simultaneously, with the idea that sharing information between tasks can lead to improved performance on each individual task. This paper presents a novel MTL approach that employs combined training for human activities with different temporal scales of atomic and composite activities. Atomic activities are basic, indivisible actions that are readily identifiable and classifiable. Composite activities are complex actions that comprise a sequence or combination of atomic activities. The proposed MTL approach can help in addressing challenges related to recognizing and predicting both atomic and composite activities. It can also help in providing a solution to the data scarcity problem by simultaneously learning multiple related tasks so that knowledge from each task can be reused by the others. The proposed approach offers advantages like improved data efficiency, reduced overfitting due to shared representations, and fast learning through the use of auxiliary information. The proposed approach exploits the similarities and differences between multiple tasks so that these tasks can share the parameter structure, which improves model performance. The paper also figures out which tasks should be learned together and which tasks should be learned separately. If the tasks are properly selected, the shared structure of each task can help it learn more from other tasks. [ABSTRACT FROM AUTHOR]

Food and fluid intake monitoring are essential for reducing the risk of dehydration, malnutrition, and obesity. The existing research has been preponderantly focused on dietary monitoring, while fluid intake monitoring, on the other hand, is often neglected. Food and fluid intake monitoring can be based on wearable sensors, environmental sensors, smart containers, and the collaborative use of multiple sensors. Vision-based intake monitoring methods have been widely exploited with the development of visual devices and computer vision algorithms. Vision-based methods provide non-intrusive solutions for monitoring. They have shown promising performance in food/beverage recognition and segmentation, human intake action detection and classification, and food volume/fluid amount estimation. However, occlusion, privacy, computational efficiency, and practicality pose significant challenges. This paper reviews the existing work (253 articles) on vision-based intake (food and fluid) monitoring methods to assess the size and scope of the available literature and identify the current challenges and research gaps. This paper uses tables and graphs to depict the patterns of device selection, viewing angle, tasks, algorithms, experimental settings, and performance of the existing monitoring systems. [ABSTRACT FROM AUTHOR]

Machine learning techniques have progressively emerged as important and reliable tools that, when combined with machine condition monitoring, can diagnose faults with even superior performance than other condition-based monitoring approaches. Furthermore, statistical or model-based approaches are often not applicable in industrial environments with a high degree of customization of equipment and machines. Structures such as bolted joints are a key part of the industry; therefore, monitoring their health is critical to maintaining structural integrity. Despite this, there has been little research on the detection of bolt loosening in rotating joints. In this study, vibration-based detection of bolt loosening in a rotating joint of a custom sewer cleaning vehicle transmission was performed using support vector machines (SVM). Different failures were analyzed for various vehicle operating conditions. Several classifiers were trained to evaluate the influence of the number and location of accelerometers used and to determine the best approach between specific models for each operating condition or a single model for all cases. The results showed that using a single SVM model with data from four accelerometers mounted both upstream and downstream of the bolted joint resulted in more reliable fault detection, with an overall accuracy of 92.4%. [ABSTRACT FROM AUTHOR]

Human activity recognition (HAR) algorithms today are designed and evaluated on data collected in controlled settings, providing limited insights into their performance in real-world situations with noisy and missing sensor data and natural human activities. We present a real-world HAR open dataset compiled from a wristband equipped with a triaxial accelerometer. During data collection, participants had autonomy in their daily life activities, and the process remained unobserved and uncontrolled. A general convolutional neural network model was trained on this dataset, achieving a mean balanced accuracy (MBA) of 80%. Personalizing the general model through transfer learning can yield comparable and even superior results using fewer data, with the MBA improving to 85%. To emphasize the issue of insufficient real-world training data, we conducted training of the model using the public MHEALTH dataset, resulting in 100% MBA. However, upon evaluating the MHEALTH-trained model on our real-world dataset, the MBA drops to 62%. After personalizing the model with real-world data, an improvement of 17% in the MBA is achieved. This paper showcases the potential of transfer learning to make HAR models trained in different contexts (lab vs. real-world) and on different participants perform well for new individuals with limited real-world labeled data available. [ABSTRACT FROM AUTHOR]

A novel visual 3D reconstruction system, composed of a two-axis galvanometer scanner, a camera with a lens, and a set of control units, is introduced in this paper. By changing the mirror angles of the galvanometer scanner fixed in front of the camera, the boresight of the camera can be quickly adjusted. With the variable boresight, the camera can serve as a virtual multi-ocular system (VMOS), which captures the object at different perspectives. The working mechanism with a definite physical meaning is presented. A simple and efficient method for calibrating the intrinsic and extrinsic parameters of the VMOS is presented. The applicability of the proposed system for 3D reconstruction is investigated. Owing to the multiple virtual poses of the camera, the VMOS can provide stronger constraints in the object pose estimation than an ordinary perspective camera does. The experimental results demonstrate that the proposed VMOS is able to achieve 3D reconstruction performance competitive with that of a conventional stereovision system with a much more concise hardware configuration. [ABSTRACT FROM AUTHOR]

Systems engineering plays a key role in the naval sector, focusing on how to design, integrate, and manage complex systems throughout their life cycle; it is therefore difficult to conceive functional warships without it. To this end, specialized information systems for logistical support and the sustainability of material solutions are essential to ensure proper provisioning and to know the operational status of the frigate. However, based on an architecture composed of a set of logistics applications, this information system may require highly qualified operators with a deep knowledge of the behavior of onboard systems to manage it properly. In this regard, failure detection systems have been postulated as one of the main cutting-edge methods to address the challenge, employing intelligent techniques for observing anomalies in the normal behavior of systems without the need for expert knowledge. In this paper, the study is concerned to the scope of the Spanish navy, where a complex information system structure is responsible for ensuring the correct maintenance and provisioning of the vessels. In such context, we hereby suggest a comparison between different one-class techniques, such as statistical models, geometric boundaries, or dimensional reduction to face anomaly detection in specific subsystems of a warship, with the prospect of applying it to the whole ship. [ABSTRACT FROM AUTHOR]

Applications such as medical diagnosis, navigation, robotics, etc., require 3D images. Recently, deep learning networks have been extensively applied to estimate depth. Depth prediction from 2D images poses a problem that is both ill–posed and non–linear. Such networks are computationally and time–wise expensive as they have dense configurations. Further, the network performance depends on the trained model configuration, the loss functions used, and the dataset applied for training. We propose a moderately dense encoder–decoder network based on discrete wavelet decomposition and trainable coefficients (LL, LH, HL, HH). Our Nested Wavelet–Net (NDWTN) preserves the high–frequency information that is otherwise lost during the downsampling process in the encoder. Furthermore, we study the effect of activation functions, batch normalization, convolution layers, skip, etc., in our models. The network is trained with NYU datasets. Our network trains faster with good results. [ABSTRACT FROM AUTHOR]

The moisture of bulk material has a significant impact on the energetic efficiency of dry grinding, resultant particle size distribution and particle shape, and conditions of powder transport. This research aims to develop computer vision and thermovision techniques for the on-site estimation of moisture content in copper ore, for use, e.g., in dry grinding installations. The influence of particle size on the results of moisture estimation is also studied. The tested granular material was copper ore of particle size 0–2 mm and relative moisture content of 0.5–11%. Both vision and thermovision images were taken at standard and macro scales. The results suggest that median-intensity vision images monotonically reflect copper ore moisture in the range of about 0.5–5%. Suitable models were identified and cross-validated here. In contrary, thermograms should not be analyzed simply for their mean temperature but treated with computer vision processing algorithms. [ABSTRACT FROM AUTHOR]

The importance of development of new methods for reconstruction of an object image given its sinogram and some additional information about the object stems from the possibility of artifact presence in the reconstructed image, or its insufficient sharpness when the used additional information does not hold. The problem of recovering artifact-free images of the studied object from tomography data is considered in the framework of the theory of computer-aided measuring systems. Methods for solving it are developed. They are based on narrowing the class of possible images using less artifact-inducing information. An example of such information is the natural condition of non-negativeness of the estimated brightnesses. The main problem that arises is the large dimensionality of the images, which prevents the use of direct algorithms. One proposed method is based on local approach, namely correction of the result of unfiltered backprojection by applying a locally (in the space of the output image) optimal linear transformation. Another method processes a sinogram directly, without using backprojection, using iterative implementation of the measurement reduction technique. Examples of use of the proposed methods for processing teeth sinograms are given. [ABSTRACT FROM AUTHOR]

Sensor-based food intake monitoring has become one of the fastest-growing fields in dietary assessment. Researchers are exploring imaging-sensor-based food detection, food recognition, and food portion size estimation. A major problem that is still being tackled in this field is the segmentation of regions of food when multiple food items are present, mainly when similar-looking foods (similar in color and/or texture) are present. Food image segmentation is a relatively under-explored area compared with other fields. This paper proposes a novel approach to food imaging consisting of two imaging sensors: color (Red–Green–Blue) and thermal. Furthermore, we propose a multi-modal four-Dimensional (RGB-T) image segmentation using a k-means clustering algorithm to segment regions of similar-looking food items in multiple combinations of hot, cold, and warm (at room temperature) foods. Six food combinations of two food items each were used to capture RGB and thermal image data. RGB and thermal data were superimposed to form a combined RGB-T image and three sets of data (RGB, thermal, and RGB-T) were tested. A bootstrapped optimization of within-cluster sum of squares (WSS) was employed to determine the optimal number of clusters for each case. The combined RGB-T data achieved better results compared with RGB and thermal data, used individually. The mean ± standard deviation (std. dev.) of the F1 score for RGB-T data was 0.87 ± 0.1 compared with 0.66 ± 0.13 and 0.64 ± 0.39, for RGB and Thermal data, respectively. [ABSTRACT FROM AUTHOR]

Smart tourism is the latest achievement of tourism development at home and abroad. It is also an essential part of the smart city. Promoting the application of computer and sensor technology in smart tourism is conducive to improving the efficiency of public tourism services and guiding the innovation of the tourism public service mode. In this paper, we have proposed a new method of using data collected by sensor networks. We have developed and deployed sensors to collect data, which are transmitted to the modular cloud platform, and combined with cluster technology and an Uncertain Support Vector Classifier (A-USVC) location prediction method to assist in emergency events. Considering the attraction of tourists, the system also incorporated human trajectory analysis and intensity of interaction as consideration factors to validate the spatial dynamics of different interests and enhance the tourists' experience. The system explored the innovative road of computer technology to boost the development of smart tourism, which helps to promote the high-quality development of tourism. [ABSTRACT FROM AUTHOR]

Three-dimensional human pose estimation from depth maps is a fast-growing research area in computer vision. The distal joints of the human body are more flexible than the proximal joints, making it more difficult to estimate the distal joints. However, most existing methods ignore the difference between the distal joints and proximal joints. Moreover, the distal joint can be constrained by the proximal joint on the same kinematic chain. In our work, we model the human skeleton as the tree structure called the human-tree. Then, motivated by the WPL (weighted path length) in the data structure, we propose a WPL-based loss function to constrain the distal joints with the proximal joints in a global-to-local manner. Extensive experiments on benchmarks demonstrate that our method can effectively improve the performance of the distal joints. [ABSTRACT FROM AUTHOR]

Medical images play an important role in medical diagnosis and treatment. Oncologists analyze images to determine the different characteristics of deadly diseases, plan the therapy, and observe the evolution of the disease. The objective of this paper is to propose a method for the detection of brain tumors. Brain tumors are identified from Magnetic Resonance (MR) images by performing suitable segmentation procedures. The latest technical literature concerning radiographic images of the brain shows that deep learning methods can be implemented to extract specific features of brain tumors, aiding clinical diagnosis. For this reason, most data scientists and AI researchers work on Machine Learning methods for designing automatic screening procedures. Indeed, an automated method would result in quicker segmentation findings, providing a robust output with respect to possible differences in data sources, mostly due to different procedures in data recording and storing, resulting in a more consistent identification of brain tumors. To improve the performance of the segmentation procedure, new architectures are proposed and tested in this paper. We propose deep neural networks for the detection of brain tumors, trained on the MRI scans of patients' brains. The proposed architectures are based on convolutional neural networks and inception modules for brain tumor segmentation. A comparison of these proposed architectures with the baseline reference ones shows very interesting results. MI-Unet showed a performance increase in comparison to baseline Unet architecture by 7.5 % in dice score, 23.91 % insensitivity, and 7.09 % in specificity. Depth-wise separable MI-Unet showed a performance increase by 10.83 % in dice score, 2.97 % in sensitivity, and 12.72 % in specificity as compared to the baseline Unet architecture. Hybrid Unet architecture achieved performance improvement of 9.71 % in dice score, 3.56 % in sensitivity, and 12.6 % in specificity. Whereas the depth-wise separable hybrid Unet architecture outperformed the baseline architecture by 15.45 % in dice score, 20.56 % in sensitivity, and 12.22 % in specificity. [ABSTRACT FROM AUTHOR]

An unhealthy diet is strongly linked to obesity and numerous chronic diseases. Currently, over two-thirds of American adults are overweight or obese. Although dietary assessment helps people improve nutrition and lifestyle, traditional methods for dietary assessment depend on self-report, which is inaccurate and often biased. In recent years, as electronics, information, and artificial intelligence (AI) technologies advanced rapidly, image-based objective dietary assessment using wearable electronic devices has become a powerful approach. However, research in this field has been focused on the developments of advanced algorithms to process image data. Few reports exist on the study of device hardware for the particular purpose of dietary assessment. In this work, we demonstrate that, with the current hardware design, there is a considerable risk of missing important dietary data owing to the common use of rectangular image screen and fixed camera orientation. We then present two designs of a new camera system to reduce data loss by generating circular images using rectangular image sensor chips. We also present a mechanical design that allows the camera orientation to be adjusted, adapting to differences among device wearers, such as gender, body height, and so on. Finally, we discuss the pros and cons of rectangular versus circular images with respect to information preservation and data processing using AI algorithms. [ABSTRACT FROM AUTHOR]

Time-series representation is the most important task in time-series analysis. One of the most widely employed time-series representation method is symbolic aggregate approximation (SAX), which converts the results from piecewise aggregate approximation to a symbol sequence. SAX is a simple and effective method; however, it only focuses on the mean value of each segment in the time-series. Here, we propose a novel time-series representation method—distance- and momentum-based symbolic aggregate approximation (DM-SAX)—that can secure time-series distributions by calculating the perpendicular distance from the time-axis to each data point and consider the time-series trend by adding a momentum factor reflecting the direction of previous data points. Experimental results for 29 highly imbalanced classification problems on the UCR datasets revealed that DM-SAX affords the optimal area under the curve (AUC) among competing time-series representation methods (SAX, extreme-SAX, overlap-SAX, and distance-based SAX). We statistically verified that performance improvements resulted in significant differences in the rankings. In addition, DM-SAX yielded the optimal AUC for real-world wire cutting and crimping process dataset. Meaningful data points such as outliers could be identified in a time-series outlier detection framework via the proposed method. [ABSTRACT FROM AUTHOR]

Road segmentation has been one of the leading research areas in the realm of autonomous driving cars due to the possible benefits autonomous vehicles can offer. Significant reduction of crashes, greater independence for the people with disabilities, and reduced traffic congestion on the roads are some of the vivid examples of them. Considering the importance of self-driving cars, it is vital to develop models that can accurately segment drivable regions of roads. The recent advances in the area of deep learning have presented effective methods and techniques to tackle road segmentation tasks effectively. However, the results of most of them are not satisfactory for implementing them into practice. To tackle this issue, in this paper, we propose a novel model, dubbed as TA-Unet, that is able to produce quality drivable road region segmentation maps. The proposed model incorporates a triplet attention module into the encoding stage of the U-Net network to compute attention weights through the triplet branch structure. Additionally, to overcome the class-imbalance problem, we experiment on different loss functions, and confirm that using a mixed loss function leads to a boost in performance. To validate the performance and efficiency of the proposed method, we adopt the publicly available UAS dataset, and compare its results to the framework of the dataset and also to four state-of-the-art segmentation models. Extensive experiments demonstrate that the proposed TA-Unet outperforms baseline methods both in terms of pixel accuracy and mIoU, with 98.74% and 97.41%, respectively. Finally, the proposed method yields clearer segmentation maps on different sample sets compared to other baseline methods. [ABSTRACT FROM AUTHOR]

Recent industrial robotics covers a broad part of the manufacturing spectrum and other human everyday life applications; the performance of these devices has become increasingly important. Positioning accuracy and repeatability, as well as operating speed, are essential in any industrial robotics application. Robot positioning errors are complex due to the extensive combination of their sources and cannot be compensated for using conventional methods. Some robot positioning errors can be compensated for only using machine learning (ML) procedures. Reinforced machine learning increases the robot's positioning accuracy and expands its implementation capabilities. The provided methodology presents an easy and focused approach for industrial in situ robot position adjustment in real-time during production setup or readjustment cases. The scientific value of this approach is a methodology using an ML procedure without huge external datasets for the procedure and extensive computing facilities. This paper presents a deep q-learning algorithm applied to improve the positioning accuracy of an articulated KUKA youBot robot during operation. A significant improvement of the positioning accuracy was achieved approximately after 260 iterations in the online mode and initial simulation of the ML procedure. [ABSTRACT FROM AUTHOR]

Wearable technology has advanced significantly and is now used in various entertainment and business contexts. Authentication methods could be trustworthy, transparent, and non-intrusive to guarantee that users can engage in online communications without consequences. An authentication system on a security framework starts with a process for identifying the user to ensure that the user is permitted. Establishing and verifying an individual's appearance usually requires a lot of effort. Recent years have seen an increase in the usage of activity-based user identification systems to identify individuals. Despite this, there has not been much research into how complex hand movements can be used to determine the identity of an individual. This research used a one-dimensional residual network with squeeze-and-excitation (SE) configurations called the 1D-ResNet-SE model to investigate hand movements and user identification. According to the findings, the SE modules have enhanced the one-dimensional residual network's identification ability. As a deep learning model, the proposed methodology is capable of effectively identifying features from the input smartwatch sensor and could be utilized as an end-to-end model to clarify the modeling process. The 1D-ResNet-SE identification model is superior to the other models. Hand movement assessment based on deep learning is an effective technique to identify smartwatch users. [ABSTRACT FROM AUTHOR]

Recently, recognizing a user's daily activity using a smartphone and wearable sensors has become a popular issue. However, in contrast with the ideal definition of an experiment, there could be numerous complex activities in real life with respect to its various background and contexts: time, space, age, culture, and so on. Recognizing these complex activities with limited low-power sensors, considering the power and memory constraints of the wearable environment and the user's obtrusiveness at once is not an easy problem, although it is very crucial for the activity recognizer to be practically useful. In this paper, we recognize activity of eating, which is one of the most typical examples of a complex activity, using only daily low-power mobile and wearable sensors. To organize the related contexts systemically, we have constructed the context model based on activity theory and the "Five W's", and propose a Bayesian network with 88 nodes to predict uncertain contexts probabilistically. The structure of the proposed Bayesian network is designed by a modular and tree-structured approach to reduce the time complexity and increase the scalability. To evaluate the proposed method, we collected the data with 10 different activities from 25 volunteers of various ages, occupations, and jobs, and have obtained 79.71% accuracy, which outperforms other conventional classifiers by 7.54-14.4%. Analyses of the results showed that our probabilistic approach could also give approximate results even when one of contexts or sensor values has a very heterogeneous pattern or is missing. [ABSTRACT FROM AUTHOR]

This study introduces the One-Class K-means with Randomly-projected features Algorithm (OCKRA). OCKRA is an ensemble of one-class classifiers built over multiple projections of a dataset according to random feature subsets. Algorithms found in the literature spread over a wide range of applications where ensembles of one-class classifiers have been satisfactorily applied; however, none is oriented to the area under our study: personal risk detection. OCKRA has been designed with the aim of improving the detection performance in the problem posed by the Personal RIsk DEtection(PRIDE) dataset. PRIDE was built based on 23 test subjects, where the data for each user were captured using a set of sensors embedded in a wearable band. The performance of OCKRA was compared against support vector machine and three versions of the Parzen window classifier. On average, experimental results show that OCKRA outperformed the other classifiers for at least 0.53% of the area under the curve (AUC). In addition, OCKRA achieved an AUC above 90% for more than 57% of the users. [ABSTRACT FROM AUTHOR]

Human Activity Recognition (HAR) is an important part of ambient intelligence systems since it can provide user-context information, thus allowing a greater personalization of services. One of the problems with HAR systems is that the labeling process for the training data is costly, which has hindered its practical application. A common approach is to train a general model with the aggregated data from all users. The problem is that for a new target user, this model can perform poorly because it is biased towards the majority type of users and does not take into account the particular characteristics of the target user. To overcome this limitation, a user-dependent model can be trained with data only from the target user that will be optimal for this particular user; however, this requires a considerable amount of labeled data, which is cumbersome to obtain. In this work, we propose a method to build a personalized model for a given target user that does not require large amounts of labeled data. Our method uses data already labeled by a community of users to complement the scarce labeled data of the target user. Our results showed that the personalized model outperformed the general and the user-dependent models when labeled data is scarce. [ABSTRACT FROM AUTHOR]

Parkinson's Disease (PD) is characterized as the commonest neurodegenerative illness that gradually degenerates the central nervous system. The goal of this review is to come out with a summary of the recent progress of numerous forms of sensors and systems that are related to diagnosis of PD in the past decades. The paper reviews the substantial researches on the application of technological tools (objective techniques) in the PD field applying different types of sensors proposed by previous researchers. In addition, this also includes the use of clinical tools (subjective techniques) for PD assessments, for instance, patient self-reports, patient diaries and the international gold standard reference scale, Unified Parkinson Disease Rating Scale (UPDRS). Comparative studies and critical descriptions of these approaches have been highlighted in this paper, giving an insight on the current state of the art. It is followed by explaining the merits of the multiple sensor fusion platform compared to single sensor platform for better monitoring progression of PD, and ends with thoughts about the future direction towards the need of multimodal sensor integration platform for the assessment of PD. [ABSTRACT FROM AUTHOR]

In this paper, we present human pose estimation and gesture recognition algorithms that use only depth information. The proposed methods are designed to be operated with only a CPU (central processing unit), so that the algorithm can be operated on a low-cost platform, such as an embedded board. The human pose estimation method is based on an SVM (support vector machine) and superpixels without prior knowledge of a human body model. In the gesture recognition method, gestures are recognized from the pose information of a human body. To recognize gestures regardless of motion speed, the proposed method utilizes the keyframe extraction method. Gesture recognition is performed by comparing input keyframes with keyframes in registered gestures. The gesture yielding the smallest comparison error is chosen as a recognized gesture. To prevent recognition of gestures when a person performs a gesture that is not registered, we derive the maximum allowable comparison errors by comparing each registered gesture with the other gestures. We evaluated our method using a dataset that we generated. The experiment results show that our method performs fairly well and is applicable in real environments. [ABSTRACT FROM AUTHOR]

This paper describes the framework and components of an experimental platform for an advanced driver assistance system (ADAS) aimed at providing drivers with a feedback about traffic violations they have committed during their driving. The system is able to detect some specific traffic violations, record data associated to these faults in a local data-base, and also allow visualization of the spatial and temporal information of these traffic violations in a geographical map using the standard Google Earth tool. The test-bed is mainly composed of two parts: a computer vision subsystem for traffic sign detection and recognition which operates during both day and nighttime, and an event data recorder (EDR) for recording data related to some specific traffic violations. The paper covers firstly the description of the hardware architecture and then presents the policies used for handling traffic violations. [ABSTRACT FROM AUTHOR]

It is known that parameter selection for data sampling frequency and segmentation techniques (including different methods and window sizes) has an impact on the classification accuracy. For Ambient Assisted Living (AAL), no clear information to select these parameters exists, hence a wide variety and inconsistency across today's literature is observed. This paper presents the empirical investigation of different data sampling rates, segmentation techniques and segmentation window sizes and their effect on the accuracy of Activity of Daily Living (ADL) event classification and computational load for two different accelerometer sensor datasets. The study is conducted using an ANalysis Of VAriance (ANOVA) based on 32 different window sizes, three different segmentation algorithm (with and without overlap, totaling in six different parameters) and six sampling frequencies for nine common classification algorithms. The classification accuracy is based on a feature vector consisting of Root Mean Square (RMS), Mean, Signal Magnitude Area (SMA), Signal Vector Magnitude (here SMV), Energy, Entropy, FFTPeak, Standard Deviation (STD). The results are presented alongside recommendations for the parameter selection on the basis of the best performing parameter combinations that are identified by means of the corresponding Pareto curve. [ABSTRACT FROM AUTHOR]

Human Pose Recovery has been studied in the field of Computer Vision for the last 40 years. Several approaches have been reported, and significant improvements have been obtained in both data representation and model design. However, the problem of Human Pose Recovery in uncontrolled environments is far from being solved. In this paper, we define a general taxonomy to group model based approaches for Human Pose Recovery, which is composed of five main modules: appearance, viewpoint, spatial relations, temporal consistence, and behavior. Subsequently, a methodological comparison is performed following the proposed taxonomy, evaluating current SoA approaches in the aforementioned five group categories. As a result of this comparison, we discuss the main advantages and drawbacks of the reviewed literature. [ABSTRACT FROM AUTHOR]

Smartphone-based activity recognition (SP-AR) recognizes users' activities using the embedded accelerometer sensor. Only a small number of previous works can be classified as online systems, i.e., the whole process (pre-processing, feature extraction, and classification) is performed on the device. Most of these online systems use either a high sampling rate (SR) or long data-window (DW) to achieve high accuracy, resulting in short battery life or delayed system response, respectively. This paper introduces a real-time/online SP-AR system that solves this problem. Exploratory data analysis was performed on acceleration signals of 6 activities, collected from 30 subjects, to show that these signals are generated by an autoregressive (AR) process, and an accurate AR-model in this case can be built using a low SR (20 Hz) and a small DW (3 s). The high within class variance resulting from placing the phone at different positions was reduced using kernel discriminant analysis to achieve position-independent recognition. Neural networks were used as classifiers. Unlike previous works, true subject-independent evaluation was performed, where 10 new subjects evaluated the system at their homes for 1 week. The results show that our features outperformed three commonly used features by 40% in terms of accuracy for the given SR and DW. [ABSTRACT FROM AUTHOR]

Automated methods of real-time, unobtrusive, human ambulation, activity, and wellness monitoring and data analysis using various algorithmic techniques have been subjects of intense research. The general aim is to devise effective means of addressing the demands of assisted living, rehabilitation, and clinical observation and assessment through sensor-based monitoring. The research studies have resulted in a large amount of literature. This paper presents a holistic articulation of the research studies and offers comprehensive insights along four main axes: distribution of existing studies; monitoring device framework and sensor types; data collection, processing and analysis; and applications, limitations and challenges. The aim is to present a systematic and most complete study of literature in the area in order to identify research gaps and prioritize future research directions. [ABSTRACT FROM AUTHOR]

Traffic accidents are still one of the main health problems in the World. A number of measures have been applied in order to reduce the number of injuries and fatalities in roads, i.e., implementation of Advanced Driver Assistance Systems (ADAS) based on image processing. In this paper, a real time speed supervisor based on road sign recognition that can work both in urban and non-urban environments is presented. The system is able to recognize 135 road signs, belonging to the danger, yield, prohibition obligation and indication types, and sends warning messages to the driver upon the combination of two pieces of information: the current speed of the car and the road sign symbol. The core of this paper is the comparison between the two main methods which have been traditionally used for detection and recognition of road signs: template matching (TM) and neural networks (NN). The advantages and disadvantages of the two approaches will be shown and commented. Additionally we will show how the use of well-known algorithms to avoid illumination issues reduces the amount of images needed to train a neural network. [ABSTRACT FROM AUTHOR]

A road sign recognition system based on adaptive image pre-processing models using two fuzzy inference schemes has been proposed. The first fuzzy inference scheme is to check the changes of the light illumination and rich red color of a frame image by the checking areas. The other is to check the variance of vehicle's speed and angle of steering wheel to select an adaptive size and position of the detection area. The Adaboost classifier was employed to detect the road sign candidates from an image and the support vector machine technique was employed to recognize the content of the road sign candidates. The prohibitory and warning road traffic signs are the processing targets in this research. The detection rate in the detection phase is 97.42%. In the recognition phase, the recognition rate is 93.04%. The total accuracy rate of the system is 92.47%. For video sequences, the best accuracy rate is 90.54%, and the average accuracy rate is 80.17%. The average computing time is 51.86 milliseconds per frame. The proposed system can not only overcome low illumination and rich red color around the road sign problems but also offer high detection rates and high computing performance. [ABSTRACT FROM AUTHOR]

This paper presents a complete traffic sign recognition system based on vision sensor onboard a moving vehicle which detects and recognizes up to one hundred of the most important road signs, including circular and triangular signs. A restricted Hough transform is used as detection method from the information extracted in contour images, while the proposed recognition system is based on Support Vector Machines (SVM). A novel solution to the problem of discarding detected signs that do not pertain to the host road is proposed. For that purpose infrastructure-to-vehicle (I2V) communication and a stereo vision sensor are used. Furthermore, the outputs provided by the vision sensor and the data supplied by the CAN Bus and a GPS sensor are combined to obtain the global position of the detected traffic signs, which is used to identify a traffic sign in the I2V communication. This paper presents plenty of tests in real driving conditions, both day and night, in which an average detection rate over 95% and an average recognition rate around 93% were obtained with an average runtime of 35 ms that allows real-time performance. [ABSTRACT FROM AUTHOR]

Parkinson's disease (PD) and progressive supranuclear palsy (PSP) are neurodegenerative movement disorders associated with cognitive dysfunction. The Luria's Alternating Series Test (LAST) is a clinical tool sensitive to both graphomotor problems and perseverative tendencies that may suggest the dysfunction of prefrontal and/or frontostriatal areas and may be used in PD and PSP assessment. It requires the participant to draw a series of alternating triangles and rectangles. In the study, two clinical groups—51 patients with PD and 22 patients with PSP—were compared to 32 neurologically intact seniors. Participants underwent neuropsychological assessment. The LAST was administered in a paper and pencil version, then scanned and preprocessed. The series was automatically divided into characters, and the shapes were recognized as rectangles or triangles. In the feature extraction step, each rectangle and triangle was regarded both as an image and a two-dimensional signal, separately and as a part of the series. Standard and novel features were extracted and normalized using characters written by the examiner. Out of 71 proposed features, 51 differentiated the groups (p < 0.05). A classifier showed an accuracy of 70.5% for distinguishing three groups. [ABSTRACT FROM AUTHOR]

Human activity recognition (HAR) using wearable sensors is an increasingly active research topic in machine learning, aided in part by the ready availability of detailed motion capture data from smartphones, fitness trackers, and smartwatches. The goal of HAR is to use such devices to assist users in their daily lives in application areas such as healthcare, physical therapy, and fitness. One of the main challenges for HAR, particularly when using supervised learning methods, is obtaining balanced data for algorithm optimisation and testing. As people perform some activities more than others (e.g., walk more than run), HAR datasets are typically imbalanced. The lack of dataset representation from minority classes hinders the ability of HAR classifiers to sufficiently capture new instances of those activities. We introduce three novel hybrid sampling strategies to generate more diverse synthetic samples to overcome the class imbalance problem. The first strategy, which we call the distance-based method (DBM), combines Synthetic Minority Oversampling Techniques (SMOTE) with Random_SMOTE, both of which are built around the k-nearest neighbors (KNN). The second technique, referred to as the noise detection-based method (NDBM), combines SMOTE Tomek links (SMOTE_Tomeklinks) and the modified synthetic minority oversampling technique (MSMOTE). The third approach, which we call the cluster-based method (CBM), combines Cluster-Based Synthetic Oversampling (CBSO) and Proximity Weighted Synthetic Oversampling Technique (ProWSyn). We compare the performance of the proposed hybrid methods to the individual constituent methods and baseline using accelerometer data from three commonly used benchmark datasets. We show that DBM, NDBM, and CBM reduce the impact of class imbalance and enhance F1 scores by a range of 9–20 percentage point compared to their constituent sampling methods. CBM performs significantly better than the others under a Friedman test, however, DBM has lower computational requirements. [ABSTRACT FROM AUTHOR]

Due to the wide application of human activity recognition (HAR) in sports and health, a large number of HAR models based on deep learning have been proposed. However, many existing models ignore the effective extraction of spatial and temporal features of human activity data. This paper proposes a deep learning model based on residual block and bi-directional LSTM (BiLSTM). The model first extracts spatial features of multidimensional signals of MEMS inertial sensors automatically using the residual block, and then obtains the forward and backward dependencies of feature sequence using BiLSTM. Finally, the obtained features are fed into the Softmax layer to complete the human activity recognition. The optimal parameters of the model are obtained by experiments. A homemade dataset containing six common human activities of sitting, standing, walking, running, going upstairs and going downstairs is developed. The proposed model is evaluated on our dataset and two public datasets, WISDM and PAMAP2. The experimental results show that the proposed model achieves the accuracy of 96.95%, 97.32% and 97.15% on our dataset, WISDM and PAMAP2, respectively. Compared with some existing models, the proposed model has better performance and fewer parameters. [ABSTRACT FROM AUTHOR]

This article reviews the use of wearable sensors for the monitoring of physical activity (PA) for different purposes, including assessment of gait and balance, prevention and/or detection of falls, recognition of various PAs, conduction and assessment of rehabilitation exercises and monitoring of neurological disease progression. The article provides in-depth information on the retrieved articles and discusses study shortcomings related to demographic factors, i.e., age, gender, healthy participants vs patients, and study conditions. It is well known that motion patterns change with age and the onset of illnesses, and that the risk of falling increases with age. Yet, studies including older persons are rare. Gender distribution was not even provided in several studies, and others included only, or a majority of, men. Another shortcoming is that none of the studies were conducted in real-life conditions. Hence, there is still important work to be done in order to increase the usefulness of wearable sensors in these areas. The article highlights flaws in how studies based on previously collected datasets report on study samples and the data collected, which makes the validity and generalizability of those studies low. Exceptions exist, such as the promising recently reported open dataset FallAllD, wherein a longitudinal study with older adults is ongoing. [ABSTRACT FROM AUTHOR]

Cardiovascular diseases (CVDs) remain the leading cause of death worldwide. An effective management and treatment of CVDs highly relies on accurate diagnosis of the disease. As the most common imaging technique for clinical diagnosis of the CVDs, US imaging has been intensively explored. Especially with the introduction of deep learning (DL) techniques, US imaging has advanced tremendously in recent years. Photoacoustic imaging (PAI) is one of the most promising new imaging methods in addition to the existing clinical imaging methods. It can characterize different tissue compositions based on optical absorption contrast and thus can assess the functionality of the tissue. This paper reviews some major technological developments in both US (combined with deep learning techniques) and PA imaging in the application of diagnosis of CVDs. [ABSTRACT FROM AUTHOR]

Inertial sensors are widely used in the field of human activity recognition (HAR), since this source of information is the most informative time series among non-visual datasets. HAR researchers are actively exploring other approaches and different sources of signals to improve the performance of HAR systems. In this study, we investigate the impact of combining bio-signals with a dataset acquired from inertial sensors on recognizing human daily activities. To achieve this aim, we used the PPG-DaLiA dataset consisting of 3D-accelerometer (3D-ACC), electrocardiogram (ECG), photoplethysmogram (PPG) signals acquired from 15 individuals while performing daily activities. We extracted hand-crafted time and frequency domain features, then, we applied a correlation-based feature selection approach to reduce the feature-set dimensionality. After introducing early fusion scenarios, we trained and tested random forest models with subject-dependent and subject-independent setups. Our results indicate that combining features extracted from the 3D-ACC signal with the ECG signal improves the classifier's performance F1-scores by 2.72 % and 3.00 % (from 94.07 % to 96.80 % , and 83.16 % to 86.17 % ) for subject-dependent and subject-independent approaches, respectively. [ABSTRACT FROM AUTHOR]

Wearable sensors are widely used in activity recognition (AR) tasks with broad applicability in health and well-being, sports, geriatric care, etc. Deep learning (DL) has been at the forefront of progress in activity classification with wearable sensors. However, most state-of-the-art DL models used for AR are trained to discriminate different activity classes at high accuracy, not considering the confidence calibration of predictive output of those models. This results in probabilistic estimates that might not capture the true likelihood and is thus unreliable. In practice, it tends to produce overconfident estimates. In this paper, the problem is addressed by proposing deep time ensembles, a novel ensembling method capable of producing calibrated confidence estimates from neural network architectures. In particular, the method trains an ensemble of network models with temporal sequences extracted by varying the window size over the input time series and averaging the predictive output. The method is evaluated on four different benchmark HAR datasets and three different neural network architectures. Across all the datasets and architectures, our method shows an improvement in calibration by reducing the expected calibration error (ECE)by at least 40%, thereby providing superior likelihood estimates. In addition to providing reliable predictions our method also outperforms the state-of-the-art classification results in the WISDM, UCI HAR, and PAMAP2 datasets and performs as good as the state-of-the-art in the Skoda dataset. [ABSTRACT FROM AUTHOR]