Detecting Walking Challenges in Gait Patterns Using a Capacitive Sensor Floor and Recurrent Neural Networks

A Systematic Evaluation of Feature Encoding Techniques for Gait Analysis Using Multimodal Sensory Data

This paper addresses the problem of feature encoding for gait analysis using multimodal time series sensory data. In recent years, the dramatic increase in the use of numerous sensors, e.g., inertial measurement unit (IMU), in our daily wearable devices has gained the interest of the research community to collect kinematic and kinetic data to analyze the gait. The most crucial step for gait analysis is to find the set of appropriate features from continuous time series data to accurately represent human locomotion. This paper presents a systematic assessment of numerous feature extraction techniques. In particular, three different feature encoding techniques are presented to encode multimodal time series sensory data. In the first technique, we utilized eighteen different handcrafted features which are extracted directly from the raw sensory data. The second technique follows the Bag-of-Visual-Words model; the raw sensory data are encoded using a pre-computed codebook and a locality-constrained linear encoding (LLC)-based feature encoding technique. We evaluated two different machine learning algorithms to assess the effectiveness of the proposed features in the encoding of raw sensory data. In the third feature encoding technique, we proposed two end-to-end deep learning models to automatically extract the features from raw sensory data. A thorough experimental evaluation is conducted on four large sensory datasets and their outcomes are compared. A comparison of the recognition results with current state-of-the-art methods demonstrates the computational efficiency and high efficacy of the proposed feature encoding method. The robustness of the proposed feature encoding technique is also evaluated to recognize human daily activities. Additionally, this paper also presents a new dataset consisting of the gait patterns of 42 individuals, gathered using IMU sensors.

Gait Variability at Different Walking Speeds

Gait variability (GV) is a crucial measure of inconsistency of muscular activities or body segmental movements during repeated tasks. Hence, GV might serve as a relevant and sensitive measure to quantify adjustments of walking control. However, it has not been clarified whether GV is associated with walking speed, a clarification needed to exploit effective better bilateral coordination level. For this aim, fourteen male students (age 22.4 ± 2.7 years, body mass 74.9 ± 6.8 kg, and body height 1.78 ± 0.05 m) took part in this study. After three days of walking 1 km each day at a self-selected speed (SS) on asphalt with an Apple Watch S. 7 (AppleTM, Cupertino, CA, USA), the participants were randomly evaluated on a treadmill at three different walking speed intensities for 10 min at each one, SS - 20%/SS + 20%/ SS, with 5 min of passive recovery in-between. Heart rate (HR) was monitored and normalized as %HRmax, while the rate of perceived exertion (RPE) (CR-10 scale) was asked after each trial. Kinematic analysis was performed, assessing the Contact Time (CT), Swing Time (ST), Stride Length (SL), Stride Cycle (SC), and Gait Variability as Phase Coordination Index (PCI). RPE and HR increased as the walking speed increased (p = 0.005 and p = 0.035, respectively). CT and SC decreased as the speed increased (p = 0.0001 and p = 0.013, respectively), while ST remained unchanged (p = 0.277). SL increased with higher walking speed (p = 0.0001). Conversely, PCI was 3.81 ± 0.88% (high variability) at 3.96 ± 0.47 km·h-1, 2.64 ± 0.75% (low variability) at SS (4.94 ± 0.58 km·h-1), and 3.36 ± 1.09% (high variability) at 5.94 ± 0.70 km·h-1 (p = 0.001). These results indicate that while the metabolic demand and kinematics variables change linearly with increasing speed, the most effective GV was observed at SS. Therefore, SS could be a new methodological approach to choose the individual walking speed, normalize the speed intensity, and avoid a gait pattern alteration.

Research on nonlinear calibration of mine catalytic-combustion-based combustible-gas sensor based on RBF neural network

After using a catalytic-combustion-based combustible-gas sensor (catalytic sensor) underground for a period of time, the sensitivity drifts due to environmental factors such as coal dust, temperature, and humidity. It is necessary to adjust the sensor regularly to ensure its accuracy. In this paper, RBF neural network technology is introduced to fit a nonlinear continuous function to solve the problem of the output error of the sensor being too large due to linear adjustment. Through experimental analysis, it is demonstrated that the RBF neural network model has a higher convergence speed and smaller error than other network models. By embedding the RBF network model into a sensor microcontroller, the error of traditional linear calibration can be reduced by two orders of magnitude and the measurement accuracy of the catalytic sensor can be greatly improved.

Study of Pressure Distribution in Floor Tiles with Printed P(VDF:TrFE) Sensors for Smart Surface Applications

Pressure sensors integrated in surfaces, such as the floor, can enable movement, event, and object detection with relatively little effort and without raising privacy concerns, such as video surveillance. Usually, this requires a distributed array of sensor pixels, whose design must be optimized according to the expected use case to reduce implementation costs while providing sufficient sensitivity. In this work, we present an unobtrusive smart floor concept based on floor tiles equipped with a printed piezoelectric sensor matrix. The sensor element adds less than 130 µm in thickness to the floor tile and offers a pressure sensitivity of 36 pC/N for a 1 cm² pixel size. A floor model was established to simulate how the localized pressure excitation acting on the floor spreads into the sensor layer, where the error is only 1.5%. The model is valuable for optimizing the pixel density and arrangement for event and object detection while considering the smart floor implementation in buildings. Finally, a demonstration, including wireless connection to the computer, is presented, showing the viability of the tile to detect finger touch or movement of a metallic rod.

Identity and Gender Recognition Using a Capacitive Sensing Floor and Neural Networks

In recent publications, capacitive sensing floors have been shown to be able to localize individuals in an unobtrusive manner. This paper demonstrates that it might be possible to utilize the walking characteristics extracted from a capacitive floor to recognize subject and gender. Several neural network-based machine learning techniques are developed for recognizing the gender and identity of a target. These algorithms were trained and validated using a dataset constructed from the information captured from 23 subjects while walking, alone, on the sensing floor. A deep neural network comprising a Bi-directional Long Short-Term Memory (BLSTM) provided the most accurate identity performance, classifying individuals with an accuracy of 98.12% on the test data. On the other hand, a Convolutional Neural Network (CNN) was the most accurate for gender recognition, attaining an accuracy of 93.3%. The neural network-based algorithms are benchmarked against Support Vector Machine (SVM), which is a classifier used in many reported works for floor-based recognition tasks. The majority of the neural networks outperform SVM across all accuracy metrics.

Recent Advances in Flexible Sensors and Their Applications

Flexible sensors are low cost, wearable, and lightweight, as well as having a simple structure as per the requirements of engineering applications. Furthermore, for many potential applications, such as human health monitoring, robotics, wearable electronics, and artificial intelligence, flexible sensors require high sensitivity and stretchability. Herein, this paper systematically summarizes the latest progress in the development of flexible sensors. The review briefly presents the state of the art in flexible sensors, including the materials involved, sensing mechanisms, manufacturing methods, and the latest development of flexible sensors in health monitoring and soft robotic applications. Moreover, this paper provides perspectives on the challenges in this field and the prospect of flexible sensors.

Deep Neural Networks for Human's Fall-risk Prediction using Force-Plate Time Series Signal

Early and accurate identification of the balance deficits could reduce falls, in particular for older adults, a prone population. Our work investigates deep neural networks' capacity to identify human balance patterns towards predicting fall-risk. Human balance ability can be characterized based on commonly-used balance metrics, such as those derived from the force-plate time series. We hypothesized that low, moderate, and high risk of falling can be characterized based on balance metrics, derived from the force-plate time series, in conjunction with deep learning algorithms. Further, we predicted that our proposed One-One-One Deep Neural Networks algorithm provides a considerable increase in performance compared to other algorithms. Here, an open source force-plate dataset, which quantified human balance from a wide demographic of human participants (163 females and males aged 18-86) for varied standing conditions (eyes-open firm surface, eyes-closed firm surface, eyes-open foam surface, eyes-closed foam surface) was used. Classification was based on one of the several indicators of fall-risk tied to the fear of falling: the clinically-used Falls Efficacy Scale (FES) assessment. For human fall-risk prediction, the deep learning architecture implemented comprised of: Recurrent Neural Network (RNN), Long-Short Time Memory (LSTM), One Dimensional Convolutional Neural Network (1D-CNN), and a proposed One-One-One Deep Neural Network. Results showed that our One-One-One Deep Neural Networks algorithm outperformed the other aforementioned algorithms and state-of-the-art models on the same dataset. With an accuracy, precision, and sensitivity of 99.9%, 100%, 100%, respectively at the 12th epoch, we found that our proposed One-One-One Deep Neural Network model is the most efficient neural network in predicting human's fall-risk (based on the FES measure) using the force-plate time series signal. This is a novel methodology for an accurate prediction of human risk of fall.