Elderly falls are a major concerning threat resulting in over 1.5-2 million elderly people experiencing severe injuries and 1 million deaths yearly. Falls experienced by Elderly people may lead to a long-term negative impact on their physical and psychological health conditions. Major healthcare research had focused on this lately to detect and prevent the fall. In this work, an Artificial Intelligence (AI) edge computing based wearable device is designed and developed for detection and prevention of fall of elderly people. Further, the various deep learning algorithms such as Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU) are utilized for activity recognition of elderly. Also, the CNN-LSTM, RNN-LSTM and GRU-LSTM with and without attention layer respectively are utilized and the performance metrics are analyzed to find the best deep learning model. Furthermore, the three different hardware boards such as Jetson Nano developer board, Raspberry PI 3 and 4 are utilized as an AI edge computing device and the best deep learning model is implemented and the computation time is evaluated. Results demonstrate that the CNN-LSTM with attention layer exhibits the accuracy, recall, precision and F1_Score of 97%, 98%, 98% and 0.98 respectively which is better when compared to other deep learning models. Also, the computation time of NVIDIA Jetson Nano is less when compared to other edge computing devices. This work appears to be of high societal relevance since the proposed wearable device can be used to monitor the activity of elderly and prevents the elderly falls which improve the quality of life of elderly people.

This work aims to compare the performance of Machine Learning (ML) and Deep Learning (DL) algorithms in detecting users\' heartbeats on a smart bed. Targeting non-intrusive, continuous heart monitoring during sleep time, the smart bed is equipped with a 3D solid-state accelerometer. Acceleration signals are processed through an STM 32-bit microcontroller board and transmitted to a PC for recording. A photoplethysmographic sensor is simultaneously checked for ground truth reference. A dataset has been built, by acquiring measures in a real-world set-up: 10 participants were involved, resulting in 120 min of acceleration traces which were utilized to train and evaluate various Artificial Intelligence (AI) algorithms. The experimental analysis utilizes K-fold cross-validation to ensure robust model testing across different subsets of the dataset. Various ML and DL algorithms are compared, each being trained and tested using the collected data. The Random Forest algorithm exhibited the highest accuracy among all compared models. While it requires longer training time compared to some ML models such as Naïve Bayes, Linear Discrimination Analysis, and K-Nearest Neighbour Classification, it keeps substantially faster than Support Vector Machine and Deep Learning models. The Random Forest model demonstrated robust performance metrics, including recall, precision, F1-scores, macro average, weighted average, and overall accuracy well above 90%. The study highlights the better performance of the Random Forest algorithm for the specific use case, achieving superior accuracy and performance metrics in detecting user heartbeats in comparison to other ML and DL models tested. The drawback of longer training times is not too relevant in the long-term monitoring target scenario, so the Random Forest model stands out as a viable solution for real-time ballistocardiographic heartbeat detection, showcasing potential for healthcare and wellness monitoring applications.

BACKGROUND: Human activity Recognition (HAR) using smartphone sensors suffers from two major problems: sensor orientation and placement. Sensor orientation and sensor placement problems refer to the variation in sensor signal for a particular activity due to sensors\' altering orientation and placement. Extracting orientation and position invariant features from raw sensor signals is a simple solution for tackling these problems. Using few heuristic features rather than numerous time-domain and frequency-domain features offers more simplicity in this approach. The heuristic features are features which have very minimal effects of sensor orientation and placement. In this study, we evaluated the effectiveness of four simple heuristic features in solving the sensor orientation and placement problems using a 1D-CNN-LSTM model for a data set consisting of over 12 million samples.
METHODS: We accumulated data from 42 participants for six common daily activities: Lying, Sitting, Walking, and Running at 3-Metabolic Equivalent of Tasks (METs), 5-METs and 7-METs from a single accelerometer sensor of a smartphone. We conducted our study for three smartphone positions: Pocket, Backpack and Hand. We extracted simple heuristic features from the accelerometer data and used them to train and test a 1D-CNN-LSTM model to evaluate their effectiveness in solving sensor orientation and placement problems.
RESULTS: We performed intra-position and inter-position evaluations. In intra-position evaluation, we trained and tested the model using data from the same smartphone position, whereas, in inter-position evaluation, the training and test data was from different smartphone positions. For intra-position evaluation, we acquired 70-73% accuracy; for inter-position cases, the accuracies ranged between 59 and 69%. Moreover, we performed participant-specific and activity-specific analyses.
CONCLUSIONS: We found that the simple heuristic features are considerably effective in solving orientation problems. With further development, such as fusing the heuristic features with other methods that eliminate placement issues, we can also achieve a better result than the outcome we achieved using the heuristic features for the sensor placement problem. In addition, we found the heuristic features to be more effective in recognizing high-intensity activities.

A study on the gearbox (speed reducer) defect detection models built from the raw vibration signal measured by a triaxial accelerometer and based on convolutional neural networks (CNNs) is presented. Gear faults such as localized pitting, localized wear on helical pinion tooth flanks, and lubricant low level are under observation for three rotating velocities of the actuator and three load levels at the reducer output. A deep learning approach, based on 1D-CNN or 2D-CNN, is employed to extract from the vibration image significant signal features that are used further to identify one of the four states (one normal and three defects) of the system, regardless of the selected load level or the speed. The best-performing 1D-CNN-based detection model, with a testing accuracy of 98.91%, was trained on the signals measured on the Y axis along the reducer input shaft direction. The vibration data acquired from the X and Z axes of the accelerometer proved to be less relevant in discriminating the states of the gearbox, the corresponding 1D-CNN-based models achieving 97.15% and 97% testing accuracy. The 2D-CNN-based model, built using the data from all three accelerometer axes, detects the state of the gearbox with an accuracy of 99.63%.

The aim of this study was to investigate the feasibility of automatically assessing the 2-Minute Walk Distance (2MWD) for monitoring people with multiple sclerosis (pwMS). For 154 pwMS, MS-related clinical outcomes as well as the 2MWDs as evaluated by clinicians and derived from accelerometer data were collected from a total of 323 periodic clinical visits. Accelerometer data from a wearable device during 100 home-based 2MWD assessments were also acquired. The error in estimating the 2MWD was validated for walk tests performed at hospital, and then the correlation (r) between clinical outcomes and home-based 2MWD assessments was evaluated. Robust performance in estimating the 2MWD from the wearable device was obtained, yielding an error of less than 10% in about two-thirds of clinical visits. Correlation analysis showed that there is a strong association between the actual and the estimated 2MWD obtained either at hospital (r = 0.71) or at home (r = 0.58). Furthermore, the estimated 2MWD exhibits moderate-to-strong correlation with various MS-related clinical outcomes, including disability and fatigue severity scores. Automatic assessment of the 2MWD in pwMS is feasible with the usage of a consumer-friendly wearable device in clinical and non-clinical settings. Wearable devices can also enhance the assessment of MS-related clinical outcomes.

With the development of wearable devices such as smartwatches, several studies have been conducted on the recognition of various human activities. Various types of data are used, e.g., acceleration data collected using an inertial measurement unit sensor. Most scholars segmented the entire timeseries data with a fixed window size before performing recognition. However, this approach has limitations in performance because the execution time of the human activity is usually unknown. Therefore, there have been many attempts to solve this problem through the method of activity recognition by sliding the classification window along the time axis. In this study, we propose a method for classifying all frames rather than a window-based recognition method. For implementation, features extracted using multiple convolutional neural networks with different kernel sizes were fused and used. In addition, similar to the convolutional block attention module, an attention layer to each channel and spatial level is applied to improve the model recognition performance. To verify the performance of the proposed model and prove the effectiveness of the proposed method on human activity recognition, evaluation experiments were performed. For comparison, models using various basic deep learning modules and models, in which all frames were classified for recognizing a specific wave in electrocardiography data were applied. As a result, the proposed model reported the best F1-score (over 0.9) for all kinds of target activities compared to other deep learning-based recognition models. Further, for the improvement verification of the proposed CEF method, the proposed method was compared with three types of SW method. As a result, the proposed method reported the 0.154 higher F1-score than SW. In the case of the designed model, the F1-score was higher as much as 0.184.

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.

The last few decades have been characterised by a very active application of smart technologies in various fields of industry. This paper deals with industrial activities, such as injection molding, where it is required to monitor continuously the manufacturing process to identify both the effective running time and down-time periods. Supervised machine learning algorithms are developed to recognize automatically the periods of the injection molding machines. The former algorithm uses directly the features of the descriptive statistics, while the latter one utilizes a convolutional neural network. The automatic state recognition system is equipped with an 3D-accelerometer sensor whose datasets are used to train and verify the proposed algorithms. The novelty of our contribution is that accelerometer data-based machine learning models are used to distinguish producing and non-producing periods by means of recognition of key steps in an injection molding cycle. The first testing results show the approximate overall balanced accuracy of 72-92% that illustrates the large potential of the monitoring system with the accelerometer. According to the ANOVA test, there are no sufficient statistical differences between the comparative algorithms, but the results of the neural network exhibit higher variances of the defined accuracy metrics.

The purpose of this study is to develop an approach for automating quality assurance (QA) analysis for a six-degree-of-freedom (6DOF) couch using image displacement and an accelerometer sensor. A cubic phantom was fabricated using 3D printing and the accelerometer sensor was embedded in the phantom to measure the couch in the pitch and roll directions. The accuracy and reliability of image displacement and the accelerometer sensor were investigated prior to their practical use for 6DOF couch QA. Image displacement performance had an accuracy and reliability of 0.026 ± 0.026 mm for the translation direction and 0.021 ± 0.016° for the rotation direction. Accelerometer sensor performance had an accuracy and reliability of 0.023 ± 0.018° for pitch rotation and 0.051 ± 0.024° for roll rotation. Automating QA analysis was used to perform 6DOF couch QA, and the couch position errors measured using image displacement were less than 0.99 mm, 0.91 mm, 0.82 mm for the vertical, longitudinal, lateral translation in range between ±20 mm, and 0.07°, 0.23°, and 0.2° for pitch, roll, and yaw rotation in range between ±3° whereas the couch position errors measured using the accelerometer sensor were less than 0.1° and 0.19° for the pitch and roll rotation in range between ±3°.

Sign language recognition is challenging due to the lack of communication between normal and affected people. Many social and physiological impacts are created due to speaking or hearing disability. A lot of different dimensional techniques have been proposed previously to overcome this gap. A sensor-based smart glove for sign language recognition (SLR) proved helpful to generate data based on various hand movements related to specific signs. A detailed comparative review of all types of available techniques and sensors used for sign language recognition was presented in this article. The focus of this paper was to explore emerging trends and strategies for sign language recognition and to point out deficiencies in existing systems. This paper will act as a guide for other researchers to understand all materials and techniques like flex resistive sensor-based, vision sensor-based, or hybrid system-based technologies used for sign language until now.