Search

In the medical field, there are several very different movement disorders, such as tremors, Parkinson's disease, or Huntington's disease. A wide range of motor and non-motor symptoms characterizes them. It is evident that in the modern era, the use of smart wrist devices, such as smartwatches, wristbands, and smart bracelets is spreading among all categories of people. This diffusion is justified by the limited costs, ease of use, and less invasiveness (and consequently greater acceptability) than other types of sensors used for health status monitoring. This systematic review aims to synthesize research studies using smart wrist devices for a specific class of movement disorders. Following PRISMA-S guidelines, 130 studies were selected and analyzed. For each selected study, information is provided relating to the smartwatch/wristband/bracelet model used (whether it is commercial or not), the number of end-users involved in the experimentation stage, and finally the characteristics of the benchmark dataset possibly used for testing. Moreover, some articles also reported the type of raw data extracted from the smart wrist device, the implemented designed algorithmic pipeline, and the data classification methodology. It turned out that most of the studies have been published in the last ten years, showing a growing interest in the scientific community. The selected articles mainly investigate the relationship between smart wrist devices and Parkinson's disease. Epilepsy and seizure detection are also research topics of interest, while there are few papers analyzing gait disorders, Huntington's Disease, ataxia, or Tourette Syndrome. However, the results of this review highlight the difficulties still present in the use of the smartwatch/wristband/bracelet for the identified categories of movement disorders, despite the advantages these technologies could bring in the dissemination of low-cost solutions usable directly within living environments and without the need for caregivers or medical personnel.

This paper proposes a machine learning approach to detect threats using short-term PPG (photoplethysmogram) signals from a commercial smartwatch. In supervised learning, having accurately annotated training data is essential. However, a key challenge in the threat detection problem is the uncertainty regarding how accurately data labeled as 'threat' reflect actual threat responses since participants may react differently to the same experiments. In this paper, Gaussian Mixture Models are learned to remove ambiguously labeled training, and those models are also used to remove ambiguous test data. For the realistic test scenario, PPG measurements are collected from participants playing a horror VR (Virtual Reality) game, and the proposed method validates the superiority of our proposed approach in comparison with other methods. Also, the proposed filtering with GMM improves prediction accuracy by 23% compared to the method that does not incorporate the filtering.

Monitoring activities of daily living (ADLs) plays an important role in measuring and responding to a person's ability to manage their basic physical needs. Effective recognition systems for monitoring ADLs must successfully recognize naturalistic activities that also realistically occur at infrequent intervals. However, existing systems primarily focus on either recognizing more separable, controlled activity types or are trained on balanced datasets where activities occur more frequently. In our work, we investigate the challenges associated with applying machine learning to an imbalanced dataset collected from a fully in-the-wild environment. This analysis shows that the combination of preprocessing techniques to increase recall and postprocessing techniques to increase precision can result in more desirable models for tasks such as ADL monitoring. In a user-independent evaluation using in-the-wild data, these techniques resulted in a model that achieved an event-based F1-score of over 0.9 for brushing teeth, combing hair, walking, and washing hands. This work tackles fundamental challenges in machine learning that will need to be addressed in order for these systems to be deployed and reliably work in the real world.

In recent years, there has been a growing interest in developing portable and personal devices for measuring air quality and surrounding pollutants, partly due to the need for ventilation in the aftermath of COVID-19 situation. Moreover, the monitoring of hazardous chemical agents is a focus for ensuring compliance with safety standards and is an indispensable component in safeguarding human welfare. Air quality measurement is conducted by public institutions with high precision but costly equipment, which requires constant calibration and maintenance by highly qualified personnel for its proper operation. Such devices, used as reference stations, have a low spatial resolution since, due to their high cost, they are usually located in a few fixed places in the city or region to be studied. However, they also have a low temporal resolution, providing few samples per hour. To overcome these drawbacks and to provide people with personalized and up-to-date air quality information, a personal device (smartwatch) based on MEMS gas sensors has been developed. The methodology followed to validate the performance of the prototype was as follows: firstly, the detection capability was tested by measuring carbon dioxide and methane at different concentrations, resulting in low detection limits; secondly, several experiments were performed to test the discrimination capability against gases such as toluene, xylene, and ethylbenzene. principal component analysis of the data showed good separation and discrimination between the gases measured.

Assessing mobility in daily life can provide significant insights into several clinical conditions, such as Chronic Obstructive Pulmonary Disease (COPD). In this paper, we present a comprehensive analysis of wearable devices' performance in gait speed estimation and explore optimal device combinations for everyday use. Using data collected from smartphones, smartwatches, and smart shoes, we evaluated the individual capabilities of each device and explored their synergistic effects when combined, thereby accommodating the preferences and possibilities of individuals for wearing different types of devices. Our study involved 20 healthy subjects performing a modified Six-Minute Walking Test (6MWT) under various conditions. The results revealed only little performance differences among devices, with the combination of smartwatches and smart shoes exhibiting superior estimation accuracy. Particularly, smartwatches captured additional health-related information and demonstrated enhanced accuracy when paired with other devices. Surprisingly, wearing all devices concurrently did not yield optimal results, suggesting a potential redundancy in feature extraction. Feature importance analysis highlighted key variables contributing to gait speed estimation, providing valuable insights for model refinement.

Smartwatch health sensor data are increasingly utilized in smart health applications and patient monitoring, including stress detection. However, such medical data often comprise sensitive personal information and are resource-intensive to acquire for research purposes. In response to this challenge, we introduce the privacy-aware synthetization of multi-sensor smartwatch health readings related to moments of stress, employing Generative Adversarial Networks (GANs) and Differential Privacy (DP) safeguards. Our method not only protects patient information but also enhances data availability for research. To ensure its usefulness, we test synthetic data from multiple GANs and employ different data enhancement strategies on an actual stress detection task. Our GAN-based augmentation methods demonstrate significant improvements in model performance, with private DP training scenarios observing an 11.90-15.48% increase in F1-score, while non-private training scenarios still see a 0.45% boost. These results underline the potential of differentially private synthetic data in optimizing utility-privacy trade-offs, especially with the limited availability of real training samples. Through rigorous quality assessments, we confirm the integrity and plausibility of our synthetic data, which, however, are significantly impacted when increasing privacy requirements.

Variations in Global Positioning Systems (GPSs) have been used for tracking users' locations. However, when location tracking is needed for an indoor space, such as a house or building, then an alternative means of precise position tracking may be required because GPS signals can be severely attenuated or completely blocked. In our approach to indoor positioning, we developed an indoor localization system that minimizes the amount of effort and cost needed by the end user to put the system to use. This indoor localization system detects the user's room-level location within a house or indoor space in which the system has been installed. We combine the use of Bluetooth Low Energy beacons and a smartwatch Bluetooth scanner to determine which room the user is located in. Our system has been developed specifically to create a low-complexity localization system using the Nearest Neighbor algorithm and a moving average filter to improve results. We evaluated our system across a household under two different operating conditions: first, using three rooms in the house, and then using five rooms. The system was able to achieve an overall accuracy of 85.9% when testing in three rooms and 92.106% across five rooms. Accuracy also varied by region, with most of the regions performing above 96% accuracy, and most false-positive incidents occurring within transitory areas between regions. By reducing the amount of processing used by our approach, the end-user is able to use other applications and services on the smartwatch concurrently.

Unlimited access to ECGs using an over-the-counter smartwatch constitutes a real revolution for our discipline, and the application is rapidly expanding to include patients with cardiac implantable electronic devices (CIEDs) such as pacemakers (PMs) and implantable cardioverter defibrillators (ICDs). CIEDs require periodic evaluation and adjustment by healthcare professionals. In addition, implanted patients often present with symptoms that may be related to their PMs or ICDs. An ECG smartwatch could reveal information about device functioning, confirm normal device function, or aid in the case of device troubleshooting. In this review, we delve into the available evidence surrounding smartwatches with ECG registration and their integration into the care of patients with implanted pacemakers and ICDs. We explore safety considerations and the benefits and limitations associated with these wearables, drawing on relevant studies and case series from our own experience. By analyzing the current landscape of this emerging technology, we aim to provide a comprehensive overview that facilitates informed decision-making for both healthcare professionals and patients.

Advancement in smartwatch sensors and connectivity features demands low latency communication with a wide bandwidth. ISM bands below 6 GHz are reaching a threshold. The millimeter-wave (mmWave) spectrum is the solution for future smartwatch applications. Therefore, a compact dual-band antenna operating at 25.5 and 38 GHz is presented here. The characteristics mode theory (CMT) aids the antenna design process by exciting Mode 1 and 2 as well as Mode 1-3 at their respective bands. In addition, the antenna structure generates two traverse modes, TM 10 and TM 02 , at the lower and higher frequency bands. The antenna measured a bandwidth (BW) of 1.5 (25-26.5 GHz) and 2.5 GHz (37-39.5 GHz) with a maximum gain of 7.4 and 7.3 dBi, respectively. The antenna performance within the watch case (stainless steel) showed a stable |S 11 | with a gain improvement of 9.9 and 10.9 dBi and a specific absorption rate (SAR) of 0.063 and 0.0206 W/kg, respectively, at the lower and higher bands. The link budget analysis for various rotation angles of the watch indicated that, for a link margin of 20 dB, the antenna can transmit/receive 1 Gbps of data. However, significant fading was noticed at certain angles due to the shadowing effect caused by the watch case itself. Nonetheless, the antenna has a workable bandwidth, a high gain, and a low SAR, making it suitable for smartwatch and IoT applications.

Sensor data has been used in social security and welfare infrastructures such as insurance and medical care to provide personalized products and services; there is a risk that attackers can alter sensor data to obtain unfair benefits. We consider that one of the attack methods to modify sensor data is to attack the wearer's body to modify biometric information. In this study, we propose a noninvasive attack method to modify the sensor value of a photoplethysmogram. The proposed method can disappear pulse wave peaks by pressurizing the upper arm with air pressure to control blood volume. Seven subjects experiencing a rest environment and five subjects experiencing an after-exercise environment wore five different models of smartwatches, and three pressure patterns were performed. It was confirmed in both situations that the displayed heart rate decreased from the true heart rate.

Smartwatches equipped with automatic atrial fibrillation (AF) detection through electrocardiogram (ECG) recording are increasingly prevalent. We have recently reported the limitations of the Apple Watch (AW) in correctly diagnosing AF. In this study, we aim to apply a data science approach to a large dataset of smartwatch ECGs in order to deliver an improved algorithm. We included 723 patients (579 patients for algorithm development and 144 patients for validation) who underwent ECG recording with an AW and a 12-lead ECG (21% had AF and 24% had no ECG abnormalities). Similar to the existing algorithm, we first screened for AF by detecting irregularities in ventricular intervals. However, as opposed to the existing algorithm, we included all ECGs (not applying quality or heart rate exclusion criteria) but we excluded ECGs in which we identified regular patterns within the irregular rhythms by screening for interval clusters. This "irregularly irregular" approach resulted in a significant improvement in accuracy compared to the existing AW algorithm (sensitivity of 90% versus 83%, specificity of 92% versus 79%, p < 0.01). Identifying regularity within irregular rhythms is an accurate yet inclusive method to detect AF using a smartwatch ECG.

Although smartwatches are not considered medical devices, experimental validation of their accuracy in detecting hypoxemia is necessary due to their potential use in monitoring conditions manifested by a prolonged decrease in peripheral blood oxygen saturation (SpO 2 ), such as chronic obstructive pulmonary disease, sleep apnea syndrome, and COVID-19, or at high altitudes, e.g., during sport climbing, where the use of finger-sensor-based pulse oximeters may be limited. The aim of this study was to experimentally compare the accuracy of SpO 2 measurement of popular smartwatches with a clinically used pulse oximeter according to the requirements of ISO 80601-2-61. Each of the 18 young and healthy participants underwent the experimental assessment three times in randomized order-wearing Apple Watch 8, Samsung Galaxy Watch 5, or Withings ScanWatch-resulting in 54 individual experimental assessments and complete datasets. The accuracy of the SpO 2 measurements was compared to that of the Radical-7 (Masimo Corporation, Irvine, CA, USA) during short-term hypoxemia induced by consecutive inhalation of three prepared gas mixtures with reduced oxygen concentrations (14%, 12%, and 10%). All three smartwatch models met the maximum acceptable root-mean-square deviation (≤4%) from the reference measurement at both normal oxygen levels and induced desaturation with SpO 2 less than 90%. Apple Watch 8 reached the highest reliability due to its lowest mean bias and root-mean-square deviation, highest Pearson correlation coefficient, and accuracy in detecting hypoxemia. Our findings support the use of smartwatches to reliably detect hypoxemia in situations where the use of standard finger pulse oximeters may be limited.

Daily steps could be a valuable indicator of real-world ambulation in Parkinson's disease (PD). Nonetheless, no study to date has investigated the minimum number of days required to reliably estimate the average daily steps through commercial smartwatches in people with PD. Fifty-six patients were monitored through a commercial smartwatch for 5 consecutive days. The total daily steps for each day was recorded and the average daily steps was calculated as well as the working and weekend days average steps. The intraclass correlation coefficient (ICC) (3,k), standard error of measurement (SEM), Bland-Altman statistics, and minimum detectable change (MDC) were used to evaluate the reliability of the step count for every combination of 2-5 days. The threshold for acceptability was set at an ICC ≥ 0.8 with a lower bound of CI 95% ≥ 0.75 and a SAM < 10%. ANOVA and Mann-Whitney tests were used to compare steps across the days and between the working and weekend days, respectively. Four days were needed to achieve an acceptable reliability (ICC range: 0.84-0.90; SAM range: 7.8-9.4%). In addition, daily steps did not significantly differ across the days and between the working and weekend days. These findings could support the use of step count as a walking activity index and could be relevant to developing monitoring, preventive, and rehabilitation strategies for people with PD.

Background: Smart devices that are able to measure blood pressure (BP) are valuable for hypertension or heart failure management using digital technology. Data regarding their diagnostic accuracy in comparison to standard noninvasive measurement in accordance to Riva-Rocci are sparse. This study compared a wearable watch-type oscillometric BP monitor (Omron HeartGuide), a wearable watch-type infrared BP monitor (Smart Wear), a conventional ambulatory BP monitor, and auscultatory sphygmomanometry., Methods: Therefore, 159 consecutive patients (84 male, 75 female, mean age 64.33 ± 16.14 years) performed observed single measurements with the smart device compared to auscultatory sphygmomanometry (n = 109) or multiple measurements during 24 h compared to a conventional ambulatory BP monitor on the upper arm (n = 50). The two BP monitoring devices were simultaneously worn on the same arm throughout the monitoring period. In a subgroup of 50 patients, single measurements were also performed with an additional infrared smart device., Results: The intraclass correlation coefficient (ICC) between the difference and the mean of the oscillometric Omron HeartGuide and the conventional method for the single measurement was calculated for both systole (0.765) and diastole (0.732). This is exactly how the ICC was calculated for the individual mean values calculated over the 24 h long-term measurement of the individual patients for both systole (0.880) and diastole (0.829). The ICC between the infrared device and the conventional method was "bad" for SBP (0.329) and DBP (0.025). Therefore, no further long-term measurements were performed with the infrared device., Conclusion: The Omron HeartGuide device provided comparable BP values to the standard devices for single and long-term measurements. The infrared smart device failed to acquire valid measurement data.

The recognition of human activities (HAR) using wearable device data, such as smartwatches, has gained significant attention in the field of computer science due to its potential to provide insights into individuals' daily activities. This article aims to conduct a comparative study of deep learning techniques for recognizing activities of daily living (ADL). A mapping of HAR techniques was performed, and three techniques were selected for evaluation, along with a dataset. Experiments were conducted using the selected techniques to assess their performance in ADL recognition, employing standardized evaluation metrics, such as accuracy, precision, recall, and F1-score. Among the evaluated techniques, the DeepConvLSTM architecture, consisting of recurrent convolutional layers and a single LSTM layer, achieved the most promising results. These findings suggest that software applications utilizing this architecture can assist smartwatch users in understanding their movement routines more quickly and accurately.

Ubiquitous computing has been a green research area that has managed to attract and sustain the attention of researchers for some time now. As ubiquitous computing applications, human activity recognition and localization have also been popularly worked on. These applications are used in healthcare monitoring, behavior analysis, personal safety, and entertainment. A robust model has been proposed in this article that works over IoT data extracted from smartphone and smartwatch sensors to recognize the activities performed by the user and, in the meantime, classify the location at which the human performed that particular activity. The system starts by denoising the input signal using a second-order Butterworth filter and then uses a hamming window to divide the signal into small data chunks. Multiple stacked windows are generated using three windows per stack, which, in turn, prove helpful in producing more reliable features. The stacked data are then transferred to two parallel feature extraction blocks, i.e., human activity recognition and human localization. The respective features are extracted for both modules that reinforce the system's accuracy. A recursive feature elimination is applied to the features of both categories independently to select the most informative ones among them. After the feature selection, a genetic algorithm is used to generate ten different generations of each feature vector for data augmentation purposes, which directly impacts the system's performance. Finally, a deep neural decision forest is trained for classifying the activity and the subject's location while working on both of these attributes in parallel. For the evaluation and testing of the proposed system, two openly accessible benchmark datasets, the ExtraSensory dataset and the Sussex-Huawei Locomotion dataset, were used. The system outperformed the available state-of-the-art systems by recognizing human activities with an accuracy of 88.25% and classifying the location with an accuracy of 90.63% over the ExtraSensory dataset, while, for the Sussex-Huawei Locomotion dataset, the respective results were 96.00% and 90.50% accurate.

In the modern world, stress has become a pervasive concern that affects individuals' physical and mental well-being. To address this issue, many wearable devices have emerged as potential tools for stress detection and management by measuring heart rate, heart rate variability (HRV), and various metrics related to it. This literature review aims to provide a comprehensive analysis of existing research on HRV tracking and biofeedback using smartwatches pairing with reliable 3rd party mobile apps like Elite HRV, Welltory, and HRV4Training specifically designed for stress detection and management. We apply various algorithms and methodologies employed for HRV analysis and stress detection including time-domain, frequency-domain, and non-linear analysis techniques. Prominent smartwatches, such as Apple Watch, Garmin, Fitbit, Polar, and Samsung Galaxy Watch, are evaluated based on their HRV measurement accuracy, data quality, sensor technology, and integration with stress management features. We describe the efficacy of smartwatches in providing real-time stress feedback, personalized stress management interventions, and promoting overall well-being. To assist researchers, doctors, and developers with using smartwatch technology to address stress and promote holistic well-being, we discuss the data's advantages and limitations, future developments, and the significance of user-centered design and personalized interventions.

In order to ensure sufficient recovery of the human body and brain, healthy sleep is indispensable. For this purpose, appropriate therapy should be initiated at an early stage in the case of sleep disorders. For some sleep disorders (e.g., insomnia), a sleep diary is essential for diagnosis and therapy monitoring. However, subjective measurement with a sleep diary has several disadvantages, requiring regular action from the user and leading to decreased comfort and potential data loss. To automate sleep monitoring and increase user comfort, one could consider replacing a sleep diary with an automatic measurement, such as a smartwatch, which would not disturb sleep. To obtain accurate results on the evaluation of the possibility of such a replacement, a field study was conducted with a total of 166 overnight recordings, followed by an analysis of the results. In this evaluation, objective sleep measurement with a Samsung Galaxy Watch 4 was compared to a subjective approach with a sleep diary, which is a standard method in sleep medicine. The focus was on comparing four relevant sleep characteristics: falling asleep time, waking up time, total sleep time (TST), and sleep efficiency (SE). After evaluating the results, it was concluded that a smartwatch could replace subjective measurement to determine falling asleep and waking up time, considering some level of inaccuracy. In the case of SE, substitution was also proved to be possible. However, some individual recordings showed a higher discrepancy in results between the two approaches. For its part, the evaluation of the TST measurement currently does not allow us to recommend substituting the measurement method for this sleep parameter. The appropriateness of replacing sleep diary measurement with a smartwatch depends on the acceptable levels of discrepancy. We propose four levels of similarity of results, defining ranges of absolute differences between objective and subjective measurements. By considering the values in the provided table and knowing the required accuracy, it is possible to determine the suitability of substitution in each individual case. The introduction of a "similarity level" parameter increases the adaptability and reusability of study findings in individual practical cases.

Wearable devices are starting to gain popularity, which means that a large portion of the population is starting to acquire these products. This kind of technology comes with a lot of advantages, as it simplifies different tasks people do daily. However, as they recollect sensitive data, they are starting to be targets for cybercriminals. The number of attacks on wearable devices forces manufacturers to improve the security of these devices to protect them. Many vulnerabilities have appeared in communication protocols, specifically Bluetooth. We focus on understanding the Bluetooth protocol and what countermeasures have been applied during their updated versions to solve the most common security problems. We have performed a passive attack on six different smartwatches to discover their vulnerabilities during the pairing process. Furthermore, we have developed a proposal of requirements needed for maximum security of wearable devices, as well as the minimum requirements needed to have a secure pairing process between two devices via Bluetooth.

(1) Background: Consumer smartwatches may be a helpful tool to screen for atrial fibrillation (AF). However, validation studies on older stroke patients remain scarce. The aim of this pilot study from RCT NCT05565781 was to validate the resting heart rate (HR) measurement and the irregular rhythm notification (IRN) feature in stroke patients in sinus rhythm (SR) and AF. (2) Methods: Resting clinical HR measurements (every 5 min) were assessed using continuous bedside ECG monitoring (CEM) and the Fitbit Charge 5 (FC5). IRNs were gathered after at least 4 h of CEM. Lin's concordance correlation coefficient (CCC), Bland-Altman analysis, and mean absolute percentage error (MAPE) were used for agreement and accuracy assessment. (3) Results: In all, 526 individual pairs of measurements were obtained from 70 stroke patients-age 79.4 years (SD ± 10.2), 63% females, BMI 26.3 (IQ 22.2-30.5), and NIHSS score 8 (IQR 1.5-20). The agreement between the FC5 and CEM was good (CCC 0.791) when evaluating paired HR measurements in SR. Meanwhile, the FC5 provided weak agreement (CCC 0.211) and low accuracy (MAPE 16.48%) when compared to CEM recordings in AF. Regarding the accuracy of the IRN feature, analysis found a low sensitivity (34%) and high specificity (100%) for detecting AF. (4) Conclusion: The FC5 was accurate at assessing the HR during SR, but the accuracy during AF was poor. In contrast, the IRN feature was acceptable for guiding decisions regarding AF screening in stroke patients.

Balance assessment, or posturography, tracks and prevents health complications for a variety of groups with balance impairment, including the elderly population and patients with traumatic brain injury. Wearables can revolutionize state-of-the-art posturography methods, which have recently shifted focus to clinical validation of strictly positioned inertial measurement units (IMUs) as replacements for force-plate systems. Yet, modern anatomical calibration (i.e., sensor-to-segment alignment) methods have not been utilized in inertial-based posturography studies. Functional calibration methods can replace the need for strict placement of inertial measurement units, which may be tedious or confusing for certain users. In this study, balance-related metrics from a smartwatch IMU were tested against a strictly placed IMU after using a functional calibration method. The smartwatch and strictly placed IMUs were strongly correlated in clinically relevant posturography scores (r = 0.861-0.970, p < 0.001). Additionally, the smartwatch was able to detect significant variance ( p < 0.001) between pose-type scores from the mediolateral (ML) acceleration data and anterior-posterior (AP) rotation data. With this calibration method, a large problem with inertial-based posturography has been addressed, and wearable, "at-home" balance-assessment technology is within possibility.

In this work, a novel Window Score Fusion post-processing technique for biometric gait recognition is proposed and successfully tested. We show that the use of this technique allows recognition rates to be greatly improved, independently of the configuration for the previous stages of the system. For this, a strict biometric evaluation protocol has been followed, using a biometric database composed of data acquired from 38 subjects by means of a commercial smartwatch in two different sessions. A cross-session test (where training and testing data were acquired in different days) was performed. Following the state of the art, the proposal was tested with different configurations in the acquisition, pre-processing, feature extraction and classification stages, achieving improvements in all of the scenarios; improvements of 100% (0% error) were even reached in some cases. This shows the advantages of including the proposed technique, whatever the system.

Commercial smartwatches could be useful for step counting and monitoring ambulatory activity. However, in Parkinson's disease (PD) patients, an altered gait, pharmacological condition, and symptoms lateralization may affect their accuracy and potential usefulness in research and clinical routine. Steps were counted during a 6 min walk in 47 patients with PD and 47 healthy subjects (HS) wearing a Garmin Vivosmart 4 (GV4) on each wrist. Manual step counting was used as a reference. An inertial sensor (BTS G-Walk), placed on the lower back, was used to compute spatial-temporal gait parameters. Intraclass correlation coefficient (ICC) and mean absolute percentage error (MAPE) were used for accuracy evaluation and the Spearman test was used to assess the correlations between variables. The GV4 overestimated steps in PD patients with only a poor-to-moderate agreement. The OFF pharmacological state and wearing the device on the most-affected body side led to an unacceptable accuracy. The GV4 showed an excellent agreement and MAPE in HS at a self-selected speed, but an unacceptable performance at a slow speed. In PD patients, MAPE was not associated with gait parameters and clinical variables. The accuracy of commercial smartwatches for monitoring step counting might be reduced in PD patients and further influenced by the pharmacological condition and placement of the device.

Wrist-worn wearable sensors have attracted considerable research interest because of their potential in providing continuous, longitudinal, non-invasive measurements, leading to insights into Physical Activity (PA), sleep, and circadian variability. Three key practical considerations for research-grade wearables are as follows: (a) choosing an appropriate sample rate, (b) summarizing raw three-dimensional accelerometry data for further processing (accelerometry summary measures), and (c) accurately estimating PA levels and sleep towards understanding participants' 24-hour profiles. We used the CAPTURE-24 dataset, where 148 participants concurrently wore a wrist-worn three-dimensional accelerometer and a wearable camera over approximately 24 h to obtain minute-by-minute labels: sleep; and sedentary light, moderate, and vigorous PA. We propose a new acceleration summary measure, the Rate of Change Acceleration Movement (ROCAM), and compare its performance against three established approaches summarizing three-dimensional acceleration data towards replicating the minute-by-minute labels. Moreover, we compare findings where the acceleration data was sampled at 10, 25, 50, and 100 Hz. We demonstrate the competitive advantage of ROCAM towards estimating the five labels (80.2% accuracy) and building 24-hour profiles where the sample rate of 10 Hz is fully sufficient. Collectively, these findings provide insights facilitating the deployment of large-scale longitudinal actigraphy data processing towards 24-hour PA and sleep-profile assessment.

In situ self-reports are a useful tool in the social sciences to supplement laboratory experiments. Smartwatches are a promising form factor to realize these methods. However, to date, no user-friendly, general-purpose solution has been available. This article therefore presents a newly developed, free and open-source firmware that facilitates the Experience Sampling Method and other self-report methods on a commercially-available, programmable smartwatch based on the ESP32 microcontroller. In a small-scale pilot study comparing this smartwatch and firmware to an equivalent design on smartphones, participants using the smartwatch showed increased compliance. The presented project demonstrates a useful tool for complementary tools like smartphones for self-reports.

The Apple Watch is capable of recording single-lead electrocardiograms (ECGs). To incorporate such devices in routine medical care, the reliability of such devices to obtain precordial leads needs to be validated. The purpose of this study was to assess the feasibility and reliability of a smartwatch (SW) to obtain precordial leads compared to standard ECGs. We included 100 participants (62 male, aged 62.8 ± 13.1 years) with sinus rhythm and recorded a standard 12-lead ECG and the precordial leads with the Apple Watch. The ECGs were quantitively compared. A total of 98 patients were able to record precordial leads without assistance. A strong correlation was observed between the amplitude of the standard and SW-ECGs' waves, in terms of P waves, QRS-complexes, and T waves (all p -values < 0.01). A significant correlation was observed between the two methods regarding the duration of the ECG waves (all p -values < 0.01). Assessment of polarity showed a significant and a strong concordance between the ECGs' waves in all six leads (91-100%, all p -values < 0.001). In conclusion, 98% of patients were able to record precordial leads using a SW without assistance. The SW is feasible and reliable for obtaining valid precordial-lead ECG recordings as a validated alternative to a standard ECG.

In this study, we develop a method for detecting the motions performed on a trampoline using an accelerometer mounted on a smartwatch. This method will lead to a system that can be used to promote trampoline exercise using a home trampoline by detecting motions on the trampoline using a smartwatch. We proposed a method based on the convolutional neural network to detect the motions on a trampoline. As a result of the performance evaluation by leave-one-subject-out cross-validation on eight subjects, our method achieves 78.8% estimation accuracy, which is the best estimation accuracy compared to the baseline methods. We also evaluate the inference time and the battery consumption when the model is actually running on a smartwatch. Our method is effective for on-device prediction.

This study is motivated by the fact that there are currently no widely used applications available to quantitatively measure a power wheelchair user's mobility, which is an important indicator of quality of life. To address this issue, we propose an approach that allows power wheelchair users to use their own mobile devices, e.g., a smartphone or smartwatch, to non-intrusively collect mobility data in their daily life. However, the convenience of data collection brings substantial challenges in data analysis because the data patterns associated with wheelchair maneuvers are not as strong as other activities, e.g., walking, running, etc. In addition, the built-in sensors in different mobile devices create significant heterogeneity in terms of sensitivity, noise patterns, sampling settings, etc. To address the aforementioned challenges, we developed a novel approach composed of algorithms that work collaboratively to reduce noise, identify patterns intrinsic to wheelchair maneuvers, and finalize mobility analysis by removing spikes and dips caused by abrupt maneuver changes. We conducted a series of experiments to evaluate the proposed approach. Experimental results showed that our approach could accurately determine wheelchair maneuvers regardless of the models and placements of the mobile devices.

Frequent spontaneous facial self-touches, predominantly during outbreaks, have the theoretical potential to be a mechanism of contracting and transmitting diseases. Despite the recent advent of vaccines, behavioral approaches remain an integral part of reducing the spread of COVID-19 and other respiratory illnesses. The aim of this study was to utilize the functionality and the spread of smartwatches to develop a smartwatch application to identify motion signatures that are mapped accurately to face touching. Participants ( n = 10, five women, aged 20-83) performed 10 physical activities classified into face touching (FT) and non-face touching (NFT) categories in a standardized laboratory setting. We developed a smartwatch application on Samsung Galaxy Watch to collect raw accelerometer data from participants. Data features were extracted from consecutive non-overlapping windows varying from 2 to 16 s. We examined the performance of state-of-the-art machine learning methods on face-touching movement recognition (FT vs. NFT) and individual activity recognition (IAR): logistic regression, support vector machine, decision trees, and random forest. While all machine learning models were accurate in recognizing FT categories, logistic regression achieved the best performance across all metrics (accuracy: 0.93 ± 0.08, recall: 0.89 ± 0.16, precision: 0.93 ± 0.08, F1-score: 0.90 ± 0.11, AUC: 0.95 ± 0.07) at the window size of 5 s. IAR models resulted in lower performance, where the random forest classifier achieved the best performance across all metrics (accuracy: 0.70 ± 0.14, recall: 0.70 ± 0.14, precision: 0.70 ± 0.16, F1-score: 0.67 ± 0.15) at the window size of 9 s. In conclusion, wearable devices, powered by machine learning, are effective in detecting facial touches. This is highly significant during respiratory infection outbreaks as it has the potential to limit face touching as a transmission vector.

Many accidents are caused by sudden changes in the physical conditions of professional drivers. Therefore, it is quite important that the driver monitoring system must not restrict or interfere with the driver's action. Applications that can measure a driver's heartbeat without restricting the driver's action are currently under development. In this review, examples of heartbeat-monitoring systems are discussed. In particular, methods for measuring the heartbeat through sensing devices of a wearable-type, such as wristwatch-type, ring-type, and shirt-type devices, as well as through devices of a nonwearable type, such as steering-type, seat-type, and other types of devices, are discussed. The emergence of wearable devices such as the Apple Watch is considered a turning point in the application of driver-monitoring systems. The problems associated with current smartwatch- and smartphone-based systems are discussed, as are the barriers to their practical use in vehicles. We conclude that, for the time being, detection methods using in-vehicle devices and in-vehicle cameras are expected to remain dominant, while devices that can detect health conditions and abnormalities simply by driving as usual are expected to emerge as future applications.

Despite prolific demands and sales, commercial sleep assessment is primarily limited by the inability to "measure" sleep itself; rather, secondary physiological signals are captured, combined, and subsequently classified as sleep or a specific sleep state. Using markedly different approaches compared with gold-standard polysomnography, wearable companies purporting to measure sleep have rapidly developed during recent decades. These devices are advertised to monitor sleep via sensors such as accelerometers, electrocardiography, photoplethysmography, and temperature, alone or in combination, to estimate sleep stage based upon physiological patterns. However, without regulatory oversight, this market has historically manufactured products of poor accuracy, and rarely with third-party validation. Specifically, these devices vary in their capacities to capture a signal of interest, process the signal, perform physiological calculations, and ultimately classify a state (sleep vs. wake) or sleep stage during a given time domain. Device performance depends largely on success in all the aforementioned requirements. Thus, this review provides context surrounding the complex hardware and software developed by wearable device companies in their attempts to estimate sleep-related phenomena, and outlines considerations and contributing factors for overall device success.

In this work we performed a comparison between two different approaches to track a person in indoor environments using a locating system based on BLE technology with a smartphone and a smartwatch as monitoring devices. To do so, we provide the system architecture we designed and describe how the different elements of the proposed system interact with each other. Moreover, we have evaluated the system's performance by computing the mean percentage error in the detection of the indoor position. Finally, we present a novel location prediction system based on neural embeddings, and a soft-attention mechanism, which is able to predict user's next location with 67% accuracy.

As many as 40% to 50% of patients do not adhere to long-term medications for managing chronic conditions, such as diabetes or hypertension. Limited opportunity for medication monitoring is a major problem from the perspective of health professionals. The availability of prompt medication error reports can enable health professionals to provide immediate interventions for patients. Furthermore, it can enable clinical researchers to modify experiments easily and predict health levels based on medication compliance. This study proposes a method in which videos of patients taking medications are recorded using a camera image sensor integrated into a wearable device. The collected data are used as a training dataset based on applying the latest convolutional neural network (CNN) technique. As for an artificial intelligence (AI) algorithm to analyze the medication behavior, we constructed an object detection model (Model 1) using the faster region-based CNN technique and a second model that uses the combined feature values to perform action recognition (Model 2). Moreover, 50,000 image data were collected from 89 participants, and labeling was performed on different data categories to train the algorithm. The experimental combination of the object detection model (Model 1) and action recognition model (Model 2) was newly developed, and the accuracy was 92.7%, which is significantly high for medication behavior recognition. This study is expected to enable rapid intervention for providers seeking to treat patients through rapid reporting of drug errors.

As important sensors in smart sensing systems, smartwatches are becoming more and more popular. Authentication can help protect the security and privacy of users. In addition to the classic authentication methods, behavioral factors can be used as robust measures for this purpose. This study proposes a lightweight authentication method for smartwatches based on edge computing, which identifies users by their tapping rhythms. Based on the DBSCAN clustering algorithm, a new classification method called One-Class DBSCAN is presented. It first seeks core objects and then leverages them to perform user authentication. We conducted extensive experiments on 6110 real data samples collected from more than 600 users. The results show that our method achieved the lowest Equal Error Rate (EER) of only 0.92%, which was lower than those of other state-of-the-art methods. In addition, a statistical method for detecting the security level of a tapping rhythm is proposed. It can prevent users from setting a simple tapping rhythm password, and thus improve the security of smartwatches.

Understanding people's eating habits plays a crucial role in interventions promoting a healthy lifestyle. This requires objective measurement of the time at which a meal takes place, the duration of the meal, and what the individual eats. Smartwatches and similar wrist-worn devices are an emerging technology that offers the possibility of practical and real-time eating monitoring in an unobtrusive, accessible, and affordable way. To this end, we present a novel approach for the detection of eating segments with a wrist-worn device and fusion of deep and classical machine learning. It integrates a novel data selection method to create the training dataset, and a method that incorporates knowledge from raw and virtual sensor modalities for training with highly imbalanced datasets. The proposed method was evaluated using data from 12 subjects recorded in the wild, without any restriction about the type of meals that could be consumed, the cutlery used for the meal, or the location where the meal took place. The recordings consist of data from accelerometer and gyroscope sensors. The experiments show that our method for detection of eating segments achieves precision of 0.85, recall of 0.81, and F1-score of 0.82 in a person-independent manner. The results obtained in this study indicate that reliable eating detection using in the wild recorded data is possible with the use of wearable sensors on the wrist.

Internet of things (IoT) cloud-based applications deliver advanced solutions for smart cities to decrease traffic accidents caused by driver fatigue while driving on the road. Environmental conditions or driver behavior can ultimately lead to serious roadside accidents. In recent years, the authors have developed many low-cost, computerized, driver fatigue detection systems (DFDs) to help drivers, by using multi-sensors, and mobile and cloud-based computing architecture. To promote safe driving, these are the most current emerging platforms that were introduced in the past. In this paper, we reviewed state-of-the-art approaches for predicting unsafe driving styles using three common IoT-based architectures. The novelty of this article is to show major differences among multi-sensors, smartphone-based, and cloud-based architectures in multimodal feature processing. We discussed all of the problems that machine learning techniques faced in recent years, particularly the deep learning (DL) model, to predict driver hypovigilance, especially in terms of these three IoT-based architectures. Moreover, we performed state-of-the-art comparisons by using driving simulators to incorporate multimodal features of the driver. We also mention online data sources in this article to test and train network architecture in the field of DFDs on public available multimodal datasets. These comparisons assist other authors to continue future research in this domain. To evaluate the performance, we mention the major problems in these three architectures to help researchers use the best IoT-based architecture for detecting DFDs in a real-time environment. Moreover, the important factors of Multi-Access Edge Computing (MEC) and 5th generation (5G) networks are analyzed in the context of deep learning architecture to improve the response time of DFD systems. Lastly, it is concluded that there is a research gap when it comes to implementing the DFD systems on MEC and 5G technologies by using multimodal features and DL architecture.

Heart rate variability (HRV) analysis can be a useful tool to detect underlying heart or even general health problems. Currently, such analysis is usually performed in controlled or semi-controlled conditions. Since many of the typical HRV measures are sensitive to data quality, manual artifact correction is common in literature, both as an exclusive method or in addition to various filters. With proliferation of Personal Monitoring Devices with continuous HRV analysis an opportunity opens for HRV analysis in a new setting. However, current artifact correction approaches have several limitations that hamper the analysis of real-life HRV data. To address this issue we propose an algorithm for automated artifact correction that has a minimal impact on HRV measures, but can handle more artifacts than existing solutions. We verify this algorithm based on two datasets. One collected during a recreational bicycle race and another one in a laboratory, both using a PMD in form of a GPS watch. Data include direct measurement of electrical myocardial signals using chest straps and direct measurements of power using a crank sensor (in case of race dataset), both paired with the watch. Early results suggest that the algorithm can correct more artifacts than existing solutions without a need for manual support or parameter tuning. At the same time, the error introduced to HRV measures for peak correction and shorter gaps is similar to the best existing solution (Kubios-inspired threshold-based cubic interpolation) and better than commonly used median filter. For longer gaps, cubic interpolation can in some cases result in lower error in HRV measures, but the shape of the curve it generates matches ground truth worse than our algorithm. It might suggest that further development of the proposed algorithm may also improve these results.

The World Health Organization (WHO) in 2016 considered m-health as: "the use of mobile wireless technologies including smart devices such as smartphones and smartwatches for public health". WHO emphasizes the potential of this technology to increase its use in accessing health information and services as well as promoting positive changes in health behaviours and overall management of diseases. In this regard, the capability of smartphones and smartwatches for m-health monitoring through the collection of patient data remotely, has become an important component in m-health system. It is important that the integrity of the data collected is verified continuously through data authentication before storage. In this research work, we extracted heart rate variability (HRV) and decomposed the signals into sub-bands of detail and approximation coefficients. A comparison analysis is done after the classification of the extracted features to select the best sub-bands. An architectural framework and a used case for m-health data authentication is carried out using two sub-bands with the best performance from the HRV decomposition using 30 subjects' data. The best sub-band achieved an equal error rate (EER) of 12.42%.

Falls are a significant threat to the health and independence of elderly people and represent an enormous burden on the healthcare system. Successfully predicting falls could be of great help, yet this requires a timely and accurate fall risk assessment. Gait abnormalities are one of the best predictive signs of underlying locomotion conditions and precursors of falls. The advent of wearable sensors and wrist-worn devices provides new opportunities for continuous and unobtrusive monitoring of gait during daily activities, including the identification of unexpected changes in gait. To this end, we present in this paper a novel method for determining gait abnormalities based on a wrist-worn device and a deep neural network. It integrates convolutional and bidirectional long short-term memory layers for successful learning of spatiotemporal features from multiple sensor signals. The proposed method was evaluated using data from 18 subjects, who recorded their normal gait and simulated abnormal gait while wearing impairment glasses. The data consist of inertial measurement unit (IMU) sensor signals obtained from smartwatches that the subjects wore on both wrists. Numerous experiments showed that the proposed method provides better results than the compared methods, achieving 88.9% accuracy, 90.6% sensitivity, and 86.2% specificity in the detection of abnormal walking patterns using data from an accelerometer, gyroscope, and rotation vector sensor. These results indicate that reliable fall risk assessment is possible based on the detection of walking abnormalities with the use of wearable sensors on a wrist.

Some of the recently released smartwatch products feature a single-lead electrocardiogram (ECG) recording capability. The reliability of obtaining 3-lead ECG with smartwatches is yet to be confirmed in a large study. This study aimed to assess the feasibility and reliability of smartwatch to obtain 3-lead ECG recordings, the classical Einthoven ECG leads I-III compared to standard ECG. To record lead I, the watch was worn on the left wrist and the right index finger was placed on the digital crown for 30 s. For lead II, the watch was placed on the lower abdomen and the right index finger was placed on the digital crown for 30 s. For lead III, the same process was repeated with the left index finger. Spearman correlation and Bland-Altman tests were used for data analysis. A total of 300 smartwatch ECG tracings were successfully obtained. ECG waves' characteristics of all three leads obtained from the smartwatch had a similar duration, amplitude, and polarity compared to standard ECG. The results of this study suggested that the examined smartwatch (Apple Watch Series 4) could obtain 3-lead ECG tracings, including Einthoven leads I, II, and III by placing the smartwatch on the described positions.

Smartwatch battery limitations are one of the biggest hurdles to their acceptability in the consumer market. To our knowledge, despite promising studies analyzing smartwatch battery data, there has been little research that has analyzed the battery usage of a diverse set of smartwatches in a real-world setting. To address this challenge, this paper utilizes a smartwatch dataset collected from 832 real-world users, including different smartwatch brands and geographic locations. First, we employ clustering to identify common patterns of smartwatch battery utilization; second, we introduce a transparent low-parameter convolutional neural network model, which allows us to identify the latent patterns of smartwatch battery utilization. Our model converts the battery consumption rate into a binary classification problem; i.e., low and high consumption. Our model has 85.3% accuracy in predicting high battery discharge events, outperforming other machine learning algorithms that have been used in state-of-the-art research. Besides this, it can be used to extract information from filters of our deep learning model, based on learned filters of the feature extractor, which is impossible for other models. Third, we introduce an indexing method that includes a longitudinal study to quantify smartwatch battery quality changes over time. Our novel findings can assist device manufacturers, vendors and application developers, as well as end-users, to improve smartwatch battery utilization.

Background: Smartwatches that are able to record a bipolar ECG and Einthoven leads were recently described. Nevertheless, for detection of ischemia or other cardiac diseases more leads are required, especially Wilson's chest leads., Objectives: Feasibility study of six single-lead smartwatch (Apple Watch Series 4) ECG recordings including Einthoven (I, II, III) and Wilson-like pseudo-unipolar chest leads (Wr, Wm, Wl)., Methods: In 50 healthy subjects (16 males; age: 36 ± 11 years, mean ± SD) without known cardiac disorders, a standard 12-lead ECG and a six single-lead ECG using an Apple Watch Series 4 were performed under resting conditions. Recording of Einthoven I was performed with the watch on the left wrist and the right index finger on the crown, Einthoven II was recorded with the watch on the left lower abdomen and the right index finger on the crown, Einthoven III was recorded with the watch on the left lower abdomen and the left index finger on the crown. Wilson-like chest leads were recorded corresponding to the locations of V1 (Wr), V4 (Wm) and V6 (Wl) in the standard 12-lead ECG. Wr was recorded in the fourth intercostal space right parasternal, Wm was recorded in the fifth intercostal space on the midclavicular line, and Wl was recorded in the fifth intercostal space in left midaxillary line. For all Wilson-like chest lead recordings, the smartwatch was placed on the described three locations on the chest, the right index finger was placed on the crown and the left hand encompassed the right wrist. Both hands and forearms also had contact to the chest. Three experienced cardiologists were independently asked to allocate three bipolar limb smartwatch ECGs to Einthoven I-III leads, and three smartwatch Wilson-like chest ECGs (Wr, Wm, Wl) to V1, V4 and V6 in the standard 12-lead ECG for each subject., Results: All 300 smartwatch ECGs showed a signal quality useable for diagnostics with 281 ECGs of good signal quality (143 limb lead ECGs (95%), 138 chest lead ECGs (92%). Nineteen ECGs had a moderate signal quality (7 limb lead ECGs (5%), 12 chest lead ECGs (8%)). One-hundred percent of all Einthoven and 92% of all Wilson-like smartwatch ECGs were allocated correctly to corresponding leads from 12-lead ECG. Forty-six subjects (92%) were assigned correctly by all cardiologists. Allocation errors were due to similar morphologies and amplitudes in at least two of the three recorded Wilson-like leads. Despite recording with a bipolar smartwatch device, morphology of all six leads was identical to standard 12-lead ECG. In two patients with acute anterior myocardial infarction, all three cardiologists recognized the ST-elevations in Wilson-like leads and assumed an occluded left anterior descending coronary artery correctly., Conclusion: Consecutive recording of six single-lead ECGs including Einthoven and Wilson-like leads by a smartwatch is feasible with good ECG signal quality. Thus, this simulated six-lead smartwatch ECG may be useable for the detection of cardiac diseases necessitating more than one ECG lead like myocardial ischemia or more complex cardia arrhythmias.

Aims: Feasibility study of accurate three lead ECG recording (Einthoven I, II and III) using an Apple Watch Series 4. Methods : In 50 healthy subjects (18 male; age: 40 ± 12 years) without known cardiac disorders, a 12-lead ECG and three bipolar ECGs, corresponding to Einthoven leads I, II and III were recorded using an Apple Watch Series 4. Einthoven I was recorded with the watch on the left wrist and the right index finger on the crown, Einthoven II with the watch on the left lower abdomen and the right index finger on the crown, Einthoven III with the watch on the left lower abdomen and the left index finger on the crown. Four experienced cardiologists were independently asked to assign the watch ECGs to Einthoven leads from 12-lead ECG for each subject. Results: All watch ECGs showed an adequate signal quality with 134 ECGs of good (89%) and 16 of moderate signal quality (11%). Ninety-one percent of all watch ECGs were assigned correctly to corresponding leads from 12-lead ECG. Thirty-nine subjects (78%) were assigned correctly by all cardiologists. All assignment errors occurred in patients with similar morphologies and amplitudes in at least two of the three recorded leads. Erroneous assignment of all watch ECGs to leads from standard ECG occurred in no patient. Conclusion: Recording of Einthoven leads I-III by a smartwatch is accurate and highly comparable to standard ECG. This might contribute to an earlier detection of cardiac disorders, which are associated with repolarization abnormalities or arrhythmias.

The negative effects of mental stress on human health has been known for decades. High-level stress must be detected at early stages to prevent these negative effects. After the emergence of wearable devices that could be part of our lives, researchers have started detecting extreme stress of individuals with them during daily routines. Initial experiments were performed in laboratory environments and recently a number of works took a step outside the laboratory environment to the real-life. We developed an automatic stress detection system using physiological signals obtained from unobtrusive smart wearable devices which can be carried during the daily life routines of individuals. This system has modality-specific artifact removal and feature extraction methods for real-life conditions. We further tested our system in a real-life setting with collected physiological data from 21 participants of an algorithmic programming contest for nine days. This event had lectures, contests as well as free time. By using heart activity, skin conductance and accelerometer signals, we successfully discriminated contest stress, relatively higher cognitive load (lecture) and relaxed time activities by using different machine learning methods.

Activity recognition using off-the-shelf smartwatches is an important problem in humanactivity recognition. In this paper, we present an end-to-end deep learning approach, able to provideprobability distributions over activities from raw sensor data. We apply our methods to 10 complexfull-body exercises typical in CrossFit, and achieve a classification accuracy of 99.96%. We additionallyshow that the same neural network used for exercise recognition can also be used in repetitioncounting. To the best of our knowledge, our approach to repetition counting is novel and performswell, counting correctly within an error of 1 repetitions in 91% of the performed sets.

The most significant barrier to success in human activity recognition is extracting and selecting the right features. In traditional methods, the features are chosen by humans, which requires the user to have expert knowledge or to do a large amount of empirical study. Newly developed deep learning technology can automatically extract and select features. Among the various deep learning methods, convolutional neural networks (CNNs) have the advantages of local dependency and scale invariance and are suitable for temporal data such as accelerometer (ACC) signals. In this paper, we propose an efficient human activity recognition method, namely Iss2Image (Inertial sensor signal to Image), a novel encoding technique for transforming an inertial sensor signal into an image with minimum distortion and a CNN model for image-based activity classification. Iss2Image converts real number values from the X , Y , and Z axes into three color channels to precisely infer correlations among successive sensor signal values in three different dimensions. We experimentally evaluated our method using several well-known datasets and our own dataset collected from a smartphone and smartwatch. The proposed method shows higher accuracy than other state-of-the-art approaches on the tested datasets.

This paper presents SmartFall, an Android app that uses accelerometer data collected from a commodity-based smartwatch Internet of Things (IoT) device to detect falls. The smartwatch is paired with a smartphone that runs the SmartFall application, which performs the computation necessary for the prediction of falls in real time without incurring latency in communicating with a cloud server, while also preserving data privacy. We experimented with both traditional (Support Vector Machine and Naive Bayes) and non-traditional (Deep Learning) machine learning algorithms for the creation of fall detection models using three different fall datasets (Smartwatch, Notch, Farseeing). Our results show that a Deep Learning model for fall detection generally outperforms more traditional models across the three datasets. This is attributed to the Deep Learning model's ability to automatically learn subtle features from the raw accelerometer data that are not available to Naive Bayes and Support Vector Machine, which are restricted to learning from a small set of extracted features manually specified. Furthermore, the Deep Learning model exhibits a better ability to generalize to new users when predicting falls, an important quality of any model that is to be successful in the real world. We also present a three-layer open IoT system architecture used in SmartFall, which can be easily adapted for the collection and analysis of other sensor data modalities (e.g., heart rate, skin temperature, walking patterns) that enables remote monitoring of a subject's wellbeing.

Accurate step counting is essential for indoor positioning, health monitoring systems, and other indoor positioning services. There are several publications and commercial applications in step counting. Nevertheless, over-counting, under-counting, and false walking problems are still encountered in these methods. In this paper, we propose to develop a highly accurate step counting method to solve these limitations by proposing four features: Minimal peak distance, minimal peak prominence, dynamic thresholding, and vibration elimination, and these features are adaptive with the user's states. Our proposed features are combined with periodicity and similarity features to solve false walking problem. The proposed method shows a significant improvement of 99.42% and 96.47% of the average of accuracy in free walking and false walking problems, respectively, on our datasets. Furthermore, our proposed method also achieves the average accuracy of 97.04% on public datasets and better accuracy in comparison with three commercial step counting applications: Pedometer and Weight Loss Coach installed on Lenovo P780, Health apps in iPhone 5s (iOS 10.3.3), and S-health in Samsung Galaxy S5 (Android 6.01).

In recent years, with an increase in the use of smartwatches among wearable devices, various applications for the device have been developed. However, the realization of a user interface is limited by the size and volume of the smartwatch. This study aims to propose a method to classify the user's gestures without the need of an additional input device to improve the user interface. The smartwatch is equipped with an accelerometer, which collects the data and learns and classifies the gesture pattern using a machine learning algorithm. By incorporating the convolution neural network (CNN) model, the proposed pattern recognition system has become more accurate than the existing model. The performance analysis results show that the proposed pattern recognition system can classify 10 gesture patterns at an accuracy rate of 97.3%.

All drivers have their own distinct driving habits, and usually hold and operate the steering wheel differently in different driving scenarios. In this study, we proposed a novel Gaussian mixture model (GMM)-based method that can improve the traditional GMM in modeling driving behavior. This new method can be applied to build a better driver authentication system based on the accelerometer and orientation sensor of a smartwatch. To demonstrate the feasibility of the proposed method, we created an experimental system that analyzes driving behavior using the built-in sensors of a smartwatch. The experimental results for driver authentication-an equal error rate (EER) of 4.62% in the simulated environment and an EER of 7.86% in the real-traffic environment-confirm the feasibility of this approach., Competing Interests: The authors declare no conflict of interest.