To address the problem of frequent battery replacement for wearable sensors applied to fall detection among the elderly, a portable and low-cost triboelectric nanogenerator (TENG)-based self-powered sensor for human gait monitoring is proposed. The main fabrication materials of the TENG are polytetrafluoroethylene (PTFE) film, aluminum (Al) foil, and polyimide (PI) film, where PTFE and Al are the friction layer materials and the PI film is used to improve the output performance. Exploiting the ability of TENGs to monitor changes in environmental conditions, a self-powered sensor based on the TENG is placed in an insole to collect gait information. Since a TENG does not require a power source to convert physical and mechanical signals into electrical signals, the electrical signals can be used as sensing signals to be analyzed by a computer to recognize daily human activities and fall status. Experimental results show that the accuracy of the TENG-based sensor for recognizing human gait is 97.2%, demonstrating superior sensing performance and providing valuable insights for future monitoring of fall events in the elderly population.

HIGHLIGHTS

  • A triboelectric nanogenerator (TENG) does not require a power source to convert physical signals into electrical signals.

  • TENG-based sensors can monitor human gait to identify daily activities and falls.

  • This work represents an important development of nanotechnology for wearable flexible sensors.

The National Bureau of Statistics reports that China’s society began to age in 2013. Against this backdrop of an aging population, incidents of falls among the elderly occur frequently. As physical functions deteriorate, the likelihood of falling increases,1 and in severe cases, falls may pose a risk to the safety and even the life of an elderly person. Therefore, studying the gait of the elderly is of significant practical significance.

There are two data-gathering strategies for human gait identification: video-based human gait perception and sensor-based human gait recognition. Video-based human gait perception involves capturing a collection of images or video clips of human motion using cameras and analyzing them through image recognition. Human gait is identified by utilizing image recognition to assess the profile or joint body model as a gait characteristic. Sensor-based gait recognition, on the other hand, involves collecting sequential data on human motion, particularly walking, by placing sensors at key locations such as the waist and knees. The collected data are then processed for feature extraction, and ultimately human gait recognition is achieved.2 Wearable sensors offer convenient data collection, at a lower cost compared with video sequences, they have no spatial limitations, and they respect the privacy of the elderly.3 Commonly used sensors include pressure sensors, accelerometers, gyroscopes, and magnetometers.4–6 

With the emergence of a wide variety of wearable devices and implantable sensors for health monitoring, miniaturization and wearability of electronic devices have imposed more stringent requirements on traditional batteries, and although they have provided people with a richer experience, this has been at the cost of a shortened battery life. Frequent replacement of batteries can cause environmental pollution. Nanoscale energy technologies and self-powered sensors provide a solution to this problem. Triboelectric nanogenerators (TENGs) can convert mechanical energy into electrical energy and serve as self-powered sensors for monitoring the local environment and obtaining quantitative information.7 It is also possible to collect output signal datasets for different actions by wearing a TENG and a response mechanism for mechanical contact and voltage signals in the areas that need to be monitored. There are a wide variety of friction materials available for triboelectric nanogenerators. Any two materials with different friction electrical sequences,8 and with one of them at least being an insulator, can be used as friction electrical materials for triboelectric nanogenerators, such as nylon, metal, rubber, and cotton cloth.

A variety of TENGs with different architectures based on the properties described above have been investigated. Zhou et al.9 designed and fabricated a high-output, stretchable TENG using elastic materials and a helical internal electrode. This architecture, enables wearable electronic devices to be powered continuously through human motion energy collection. However, it does not effectively utilize the self-powered sensing capability of the TENG. Liu et al.10 proposed a soft tubular TENG (ST-TENG) that can collect various biomechanical energies. By walking or jogging with the ST-TENG tubes, several portable electronics, such as a thermometer, digital watch, and calculator, can be continuously powered, but its application as a sensor has not been satisfactorily demonstrated. Li et al.11 fabricated a thin and flexible TENG that could be conformally worn on the wrist and palm for monitoring of body motion and for sensing of gripped objects (with the latter having potential applications to intelligent robotic sorting), but they did not apply this TENG to gait sensing and recognition. Li et al.12 proposed a bioinspired sweat-resistant wearable TENG (BSRW-TENG) for movement monitoring during exercise. However, it does not provide a sufficient number of types of monitoring. Zhang and Cai13 proposed a high-output TENG based on polytetrafluoroethylene and cotton that can function as a human motion sensor but has limited recognition capabilities. A TENG based on liquid metal droplets was proposed by Zhang et al.14 for self-powered sensing in human gait monitoring. However, it has low recognition accuracy, and its use of liquid mercury may endanger human health. Li et al.15 developed a stretchable and repairable water-based hydrogel TENG that can be used for mechanical energy harvesting and self-powered sensing. By arranging these TENGs on different parts of the body, it is possible to detect human gait, but this device is subject to significant detection errors. Gait recognition can be studied by processing the signals obtained from various TENG structures as self-powered sensors with algorithms such as machine learning and deep learning. Cai and Zhang16 utilized a spring-structured TENG for human gait monitoring, with the electrical output signals reflecting human walking and jumping states, but with low recognition accuracy. Ahmed et al.17 reported a flame-retardant TENG sensor that can be used to monitor human activity in the petroleum drilling industry and high-temperature applications. Human activity is judged through using different durations of average peak values to identify different motion states. The processing of these signals is based on classification of the original signal features and thus requires highly accurate transmission of the original signals. Collins et al.18 directly used outline information of the human body during walking for gait recognition, employing the k-nearest neighbor (KNN) method for classification. However, this approach has stringent requirements on viewing angles and low recognition efficiency. Currently, there are still problems regarding the state of development of recognition algorithms suitable for use with TENGs for human gait sensing recognition.

Building on the foundations established by the above studies, the present work fabricates a low-cost and high-output hemispherical TENG from PTFE and aluminum foil as the principal materials and uses its self-driving sensing characteristics for gait recognition. The most representative time series are extracted by piecewise aggregate approximation (PAA), and dynamic time warping (DTW) is then applied to measure the similarity between sequences of unequal length and thereby construct a motion template library that can effectively improve the accuracy of human gait recognition.

A TENG is a microgenerator that exploits the gain and loss of electrons that occur when two materials with different electron affinities come into contact to generate an electric current in an external circuit. Its operation involves two phenomena: triboelectrification and electrostatic induction. Additionally, it can serve as a self-powered sensor to detect and identify states of human motion. When such a device is to be used for gathering human data, it is necessary to consider its suitability, durability, waterproofness, and wearability during the structural design and material selection processes. According to triboelectric series theory, it is easier to generate triboelectric charge and obtain high output performance when two materials with a large difference in triboelectric series are selected. In this study, two commonly used and cost-effective materials, namely, polytetrafluoroethylene (PTFE) film and aluminum (Al) foil, are selected as the triboelectric layer materials. The roughness of the friction layer surface can be increased by placing it on an acrylic board and polishing it with 3000-mesh sandpaper. The electrode materials are aluminum foil and copper (Cu) foil. The encapsulation materials are acrylic board and semicircular silicone rubber. Additionally, a polyimide (PI) film is inserted between the triboelectric layer and the electrode layer. This allows the sensor to produce a corresponding output voltage even under small pressure, improving the output performance and enhancing the sensitivity of the frictional electric sensor.19 When PTFE comes into contact with the aluminum metal electrode, triboelectricity is generated between the materials. Owing to the stronger electron affinity of PTFE compared with aluminum, the aluminum electrode loses electrons, while the PTFE gains electrons, leading to charge transfer. As the two triboelectric materials make contact and separate during movement, an induced potential difference is generated between the two electrodes. We select a TENG with vertical contact separation distances of 0.5 cm and 1 cm, and we simulate the potential distribution during contact separation motion using COMSOL Multiphysics 5.5. A diagram of the working principle and the potential distribution are shown in Fig. 1.

FIG. 1.

(a) Diagram of TENG contact separation working principle. (b) Potential distribution maps for d = 0.5 cm and d = 1 cm.

FIG. 1.

(a) Diagram of TENG contact separation working principle. (b) Potential distribution maps for d = 0.5 cm and d = 1 cm.

Close modal

From simulation experiments, it can be seen that the electric potential will increase as the distance between the two plates increases. This is due to the electrical principle on which a TENG is based. At the moment of contact between the two plates, there is no charge transfer activity between them. However, as the two plates are separated, owing to electrostatic induction and friction electrification effects, the charge between the plates begins to intensify. Therefore, the potential increases until both plates return to their original positions.

This study aims to collect human gait data while retaining wearability. A structure that combines a TENG with an insole is designed for gait sensing. Two 1.5 mm-thick ring-shaped acrylic plates with an inner diameter of 10 mm and outer diameters of 55 mm and 50 mm, respectively, were customized from the supplier as the bottom base and middle disk of the hemispherical TENG, respectively. First, a highly stretchable latex film was glued to the acrylic base. The second acrylic disk was fixed above the latex film so that the two layers of acrylic and latex film formed an air chamber. The aluminum foil, PTFE, and PI were formed into rings of the same dimensions as the middle acrylic disk, and were then glued in that sequence above it. Finally a hemispherical silicone rubber dome coated with copper foil was tightly fixed on top of the acrylic base structure. Figure 2 shows a schematic of the device structure, together with a photograph.

FIG. 2.

Schematic and photograph of hemispherical TENG.

FIG. 2.

Schematic and photograph of hemispherical TENG.

Close modal

The hemispherical TENG consists of two main components: the TENG structure and the air chamber. The TENG structure utilizes the contact-separation mode, which is one of the four working modes of a TENG. Contact separation process is achieved using the elastic air chamber. When an external force is applied, the two triboelectric layers come into contact with each other, and the latex film in the air chamber accommodates the compressed air from the TENG, causing the film to expand. The silicone rubber provides a good seal, isolating the overall structure from the external environment. When the external force is released, the air flows back into the raised TENG, resulting in separation of the two triboelectric layers. This design prevents water or sweat from escaping to the environment, making the sensor more durable in practical situations.

To verify the performance of the hemispherical TENG sensor, its pressure sensitivity and long-term stability were tested. At a constant frequency of 1 Hz, different pressures were applied through a DC motor. The output voltage curve is shown in Fig. 3(a). When the pressure is less than 30 N, the output voltage increases quite rapidly with increasing pressure, showing good pressure sensitivity (1.175 V/N), but when the pressure exceeds 30 N, there are only small further increases in output voltage with increasing pressure. The reason for this is that at lower pressures, a pressure change can produce a greater change in the contact area than at higher pressures, resulting in a more rapid increase in the output voltage. In addition, the stability of the TENG-based sensor was tested 1000 times at 2 Hz. It can be seen from Fig. 3(b) that the performance of the TENG remained stable after 1000 cycles, indicating that the device had excellent long-term reliability. Therefore, the proposed sensor based on a TENG showed both high sensitivity and excellent stability, indicating its practicality for gait monitoring.

FIG. 3.

Hemispherical TENG performance test: (a) pressure sensitivity; (b) long-term stability.

FIG. 3.

Hemispherical TENG performance test: (a) pressure sensitivity; (b) long-term stability.

Close modal

The working principle of a TENG-based sensor relies on contact and separation between two frictional layers. This design couples frictional electrification and electrostatic induction, generating an alternating flow of electrons. To verify the self-powered sensing performance of the proposed hemispherical TENG, it was combined with an insole. Data collection was performed using the experimental setup shown in Fig. 4. The copper electrode of the TENG was connected to the red probe of the oscilloscope, and the aluminum electrode was connected to the black probe. The oscilloscope was connected to a computer. When human motion behaviors simulating gait patterns were performed, the gait voltage signals were displayed as changing signals on the oscilloscope, and the computer exported and processed the collected data.

FIG. 4.

Experimental device and collection site.

FIG. 4.

Experimental device and collection site.

Close modal

During the data collection process, we observed that different human behaviors generated distinct and unique signals. Using the PTFE–Al triboelectric pair, we detected signals corresponding to human actions such as stepping, standing up and sitting down, jogging, fast running, walking, and falling, as shown in Figs. 5 and 6.

FIG. 5.

Output signal from falling during walking.

FIG. 5.

Output signal from falling during walking.

Close modal
FIG. 6.

Output signals generated by different gaits: (a) stepping; (b) walking; (c) standing up and sitting down; (d) jogging; (e) fast running; (f) falling.

FIG. 6.

Output signals generated by different gaits: (a) stepping; (b) walking; (c) standing up and sitting down; (d) jogging; (e) fast running; (f) falling.

Close modal

From the perspective of performance verification, whether the gait is normal or that associated with a fall, the TENG can serve as a self-driven sensor that can convert mechanical signals during human walking into corresponding voltage signals without the need for a power supply. In addition, the TENG can not only monitor changes in the generated signal but also track a fall signal during walking, as shown in Fig. 5, thus providing a unique way to monitor falling events among the elderly. At the same time, human gait signals can be characterized. As shown in Fig. 6, each gait frequency and output voltage are different. The slowest frequency of stepping (a) is about 1 Hz, and the output voltage is about 20–40 V, whereas the output voltage from walking (b) reaches 50 V, and the frequency is about 1.5 Hz. The output voltages for jogging (d) and fast running (e) range from 30 V to 60 V and 40 V to 80 V, respectively, with frequencies of 1.9 Hz and 2.4 Hz. Changes in output voltage and frequency reflect people’s daily activities. The strength of the output signal represents the strength of the motion. The output signal can also be expressed in terms of the physical parameters associated with human movement, such as the pressure response.

Time series signals are often lengthy, and employing them as inputs for PAA algorithms enables feature extraction to be performed. In combination with DTW, which measures the similarity between time series of unequal length, this enables the construction of a motion template library that can facilitate accurate recognition of both normal gait patterns and fall behaviors in humans.

The dimensionality reduction representation of time series involves extracting features from the original sequence through a certain method and representing the original time series in a lower number of dimensions, thereby simplifying the complexity of data models and algorithms and improving the efficiency of time classification. PAA achieves this by dividing the original time series into segments and representing each segment by its average value, thereby achieving the goal of feature extraction.20 

By transforming a sequence P=p1,p2,,pn of length n into another sequence Q=q1,q2,,qm of length m, where n > m, the feature representation and data dimensionality reduction of the time series are realized. Any element in the sequence Q satisfies
qi=1kj=ki1+1kipi,1im,
(1)
where k = n/m. By dividing the sequence P into segments and calculating the average value of each segment, a new sequence Q is formed, which represents the feature of the corresponding segment in sequence P. The new series Q obtained from PAA not only provides a better reflection of the overall trend of longer time series data but also enables dimensionality reduction of the time series data, reducing resource consumption and improving algorithm efficiency.
Dynamic time warping (DTW) was initially applied to speech recognition and is now commonly used for comparing the similarity between two time series.21 DTW is based on the concept of dynamic programming (DP). When two time series P=p1,p2,,pn and Q=q1,q2,,qm of lengths n and m are encountered, a matrix grid W of size n × m is constructed based on the lengths of the sequences. The matrix element w represents the distance dpi,qj between points pi and qj in the time series, where 0 < i < n and 0 < j < m. This can be denoted as follows:
dpi,qj=piqj.
(2)

The DP algorithm searches for an optimal path within the matrix grid. The points along the path correspond to the alignment of the two sequences. However, during the path-finding process, three constraints must be satisfied:

  1. Boundary constraint: the starting point must be w1,1, and the end point must be wn,m.

  2. Continuity: if wk1=a,b and wk=a,b, then aa′ ≤ 1 and bb′ ≤ 1.

  3. Monotonicity: if wk1=a,b and wk=a,b, then aa′ ≥ 0 and bb′ ≥ 0.

On the basis of these three conditions, a path w* is found satisfying
w*=mink=1KwkK,max(m,n)Km+n1
(3)
Based on the path w*, a new n × m matrix D is constructed to satisfy the dynamic programming equation, which can be denoted as follows:
d(i,j)=d(pi,qj)+minDi1,j1,Di1,j,Di,j1.
(4)

Its initial state is D0,0=0,Di,0=,D0,j=, where Dn,m indicates the degree of similarity of the two sequences P and Q and can also be represented as DTWP,Q, where a smaller value indicates a higher similarity between the two sequences.

As shown in Fig. 7, the DTW method can effectively match sequence features with similar shapes at different time points, allowing for a better representation of the similarity between time series. The curve in Fig. 7(b) describes the process of selecting the optimal path according to the dynamic programming equation; that is, the path with the minimum distance from the starting point of the two sequences to the end point of the two sequences (from one end of the diagonal line to the other end of the matrix grid) is found. The path value represents the degree of similarity between the two sequences, and the smaller this value, the more similar are the two time series.

FIG. 7.

(a) Process of temporal alignment between two sequences. (b) Process of selecting the optimal path.

FIG. 7.

(a) Process of temporal alignment between two sequences. (b) Process of selecting the optimal path.

Close modal

Time series data often have large scales, and directly processing the raw data may reduce the reliability and accuracy of the algorithm. It may also lead to the problem of excessive bending of time. Therefore, before using the DTW algorithm, it is necessary to reduce the dimensionality of the time series. This not only represents the main information of the time series, but also approximates the overall information of the original time series. In addition, most of the current algorithms used for gait recognition require a large amount of training data to ensure convergence, which undoubtedly limits the universality of the algorithm.

In this study, a TENG was installed as a self-driven sensor between the insole and sole for collection of data on daily activity and fall behavior. The data were preprocessed and transmitted, and the extracted feature representation signal was measured for similarity with the sequence in the database. Based on the characteristic of DTW being smaller in distance, the minimum measurement distance was found to identify this type of activity. The algorithm has two functions: daily activity recognition and fall detection. We apply the DTW method to identify human activities. Two datasets were prepared for the classifier, namely, the database and test data. A block diagram of the algorithm is shown in Fig. 8.

FIG. 8.

Flow chart of human gait recognition.

FIG. 8.

Flow chart of human gait recognition.

Close modal

The PAA–DTW gait classification model consists of input, feature extraction, DTW classification, and output, as shown in Fig. 9. In the model, A, B, C, D, E, F, and G represent seven gait sequences: stepping, walking, jogging, fast running, standing up and sitting down, falling, and falling while walking. The value represents the degree of similarity of the test sequence and the corresponding sequence in the template library data after the PAA–DTW algorithm has been applied: the smaller the value, the higher the degree of similarity.

FIG. 9.

Gait classification model based on PAA–DTW.

FIG. 9.

Gait classification model based on PAA–DTW.

Close modal

The algorithm proposed here first uses TENG to collect human motion data, extracts features through PAA, and then uses DTW and a template library for classification and matching. The specific implementation is as follows.

Input. Using the self-powered TENG sensor, data on human motion are collected, extracted, and transmitted to the computer through the oscilloscope. By using low-pass filters for processing, useless noise can be filtered out and useful information can be extracted.

Feature extraction. PAA feature extraction preserves sufficient information from the data to reflect the fluctuating state of the original sequence, reduces data redundancy, and replaces the overall time series with the most discriminative subsequence. The main steps of feature extraction are as follows:
  • Step 1: Divide the time series into equal segments, with lengths kp = n/wp and kq = m/wq, respectively.

  • Step 2: According to the principle of PAA, each subsequence is averaged to obtain the PAA sequences P̄=p̄1,p̄2,,p̄wp and Q̄=q̄1,q̄2,,q̄wq. The elements of the two sets of feature sequences are denoted by

p̄i=1kpj=kpi1+1ikppj
and
q̄i=1kqj=kqi1+1ikqqj,
respectively.

DTW classification. The timeline may experience nonlinear “distortion” due to stretching or contraction. DTW can be used to measure the similarity between these “distorted” sequences. After feature extraction, each gait sequence of A, B, C, D, E, F, and G is measured for similarity with the data in the template library to obtain DaaDag, DgaDgg;

Output. The gait classification model is used to select the minimum value for classification and obtain the results:
yt=minDaaDag,DgaDgg.
(5)

The following points should be noted regarding the above algorithm:

  1. Time series are segmented using equal partitioning to obtain corresponding subsequences.

  2. Each subsequence is represented by its mean value according to the principle of PAA. Numerous studies have shown that for long time series, the PAA method can preserve sufficient information from the data to reflect the fluctuating patterns of the original sequence.

  3. The distance measure function is used to quantify the similarity between the time series.

By combining these steps, the algorithm effectively captures the similarity between time series by performing PAA and DTW.

In this study, a prepared hemispherical TENG was placed in insoles as a self-powered sensor for collecting human gait data. To verify the feasibility of the algorithm, four volunteers with weight differences of no more than 2 kg, aged 23–26, and with heights ranging from 155 cm to 175 cm were invited to participate in the data collection experiment, simulating elderly individuals indoors. Different gait types were collected, including standing up and sitting down, stepping, walking, jogging, fast running, and falling during normal walking. A total of 700 human gait activity data points were obtained, with 200 samples being used for constructing the template library and 500 for validation. During the data collection process, the volunteers tried to simulate unconscious behavior to minimize human errors.

The specific operational steps were as follows:

  1. Ensuring that the TENG, oscilloscope, and computer were functioning properly.

  2. Wearing shoes and performing normal human daily activity for gait data collection.

  3. Data processing.

The experiment was conducted using a computer with the following specifications: Windows 10 64-bit operating system, an Intel Core i5 processor, and adequate RAM. The data processing was performed using the Python 3.7 and MATLAB 2020b programming languages. A SIGLENT SDS2302X fluorescent digital oscilloscope was used. Its parameters are listed in Table I.

TABLE I.

Oscilloscope parameters.

ParameterValue
Impedance 1 MΩ 
Probe 10 K 
Roll 500 ms/div 
Sampling rate 400 kSa/s 
ParameterValue
Impedance 1 MΩ 
Probe 10 K 
Roll 500 ms/div 
Sampling rate 400 kSa/s 

To verify the effectiveness of the TENG as a sensor for gait recognition, data from the four volunteers were collected for the gait recognition experiment. The specific experimental results are shown in Fig. 10 and in Tables II and III. On the basis of these results, the observations of correctly classified and misclassified instances, recognition accuracy for each gait type, and evaluation metrics are provided. Taking stepping as an example, there were a total of 72 instances, with 69 correctly classified and 3 misclassified as walking. The recognition accuracy is 95.8%, the precision is 0.95, the recall is 0.96, and the F1 score is 0.96. A higher F1 score closer to 1 indicates better recognition performance. Overall, whether for normal gait classification or identifying falling, the TENG, as a self-powered sensor, not only plays to its strengths but also builds a bridge between nanotechnology and the Internet of Things.

FIG. 10.

Confusion matrix thermodynamic diagram.

FIG. 10.

Confusion matrix thermodynamic diagram.

Close modal
TABLE II.

Detailed identification of each gait.

Gait typeTesting frequencyCorrect recognition timesNumber of errors identifiedAccuracy
Stepping 72 69 95.8% 
Standing up and sitting down 68 66 97.0% 
Walking 71 69 97.1% 
Jogging 78 77 98.7% 
Fast running 68 66 97.0% 
Falling 70 69 98.5% 
Falling during walking 73 70 95.8% 
Total 500 486 14 97.2% 
Gait typeTesting frequencyCorrect recognition timesNumber of errors identifiedAccuracy
Stepping 72 69 95.8% 
Standing up and sitting down 68 66 97.0% 
Walking 71 69 97.1% 
Jogging 78 77 98.7% 
Fast running 68 66 97.0% 
Falling 70 69 98.5% 
Falling during walking 73 70 95.8% 
Total 500 486 14 97.2% 
TABLE III.

Identification result classification report form.

Gait typePrecisionRecallF1
Stepping 0.95 0.96 0.96 
Standing up and siting down 0.96 0.97 0.98 
Jogging 0.97 0.99 0.98 
Fast running 0.96 0.97 0.97 
Walking 0.92 0.97 0.95 
Falling 0.97 0.99 0.98 
Falling during walking 0.97 0.96 0.97 
Gait typePrecisionRecallF1
Stepping 0.95 0.96 0.96 
Standing up and siting down 0.96 0.97 0.98 
Jogging 0.97 0.99 0.98 
Fast running 0.96 0.97 0.97 
Walking 0.92 0.97 0.95 
Falling 0.97 0.99 0.98 
Falling during walking 0.97 0.96 0.97 

This study has presented a low-cost and efficient triboelectric nanogenerator (TENG) as a self-powered sensor to address the energy consumption issue of wearable sensors, and it can be used as a sensor to collect human motion data and to monitor gait. A method for gait recognition based on PAA together with DTW has also been proposed. By using PAA to perform data dimensionality reduction and feature representation on time series, the overall sequence is reflected in a smaller amount of data, reducing data redundancy. Extracted feature data of different lengths are then used for DTW processing to construct a template library, effectively exploiting the sensing capabilities of the TENG. Furthermore, the gait data collected in this study can be applied to analyze the gait of patients for rehabilitation assessment and to help provide medical care for elderly patients. The proposed device can also detect falling incidents and immediately send alerts, contributing to activity recognition, medical monitoring, and research on other health assessment devices.

This work was supported by the Xi’an Science and Technology Plan Project (No. 2020KJRC0108).

The authors have no conflicts to disclose.

The data that support the findings of this study are available on request from the corresponding author.

1.
Xiang
X
,
Wang
Y
.
The current situation, characteristics, causes, and countermeasures of China’s aging population
.
Chin J Gerontol
2021
;
41
:
4149
4152
.
2.
Zhao
L
,
Chen
W
.
Detection and recognition of human body posture in motion based on sensor technology
.
IEEJ Trans Electr Electron Eng
2020
;
15
:
766
770
.
3.
Chen
D
,
Liu
X
.
Adaptive sample entropy for human activity recognition using wearable sensors
.
J Wuhan Univ Technol
2022
;
44
:
91
96
.
4.
Baek
WS
,
Kim
DM
,
Bashir
F
,
Pyun
JY
.
Real life applicable fall detection system based on wireless body area network
.
2013 IEEE 10th Consumer Communications and Networking Conference (CCNC)
.
Las Vegas, NV
:
IEEE
.
2013
. pp.
62
67
.
5.
Hwang
JY
,
Kang
JM
,
Jang
YW
,
Kim
H
.
Development of novel algorithm and real-time monitoring ambulatory system using Bluetooth module for fall detection in the elderly
.
The 26th Annual International Conference of the IEEE Engineering in Medicine and Biology Society
.
San Francisco, CA
:
IEEE
.
2004
. pp.
2204
2207
.
6.
Tolkiehn
M
,
Atallah
L
,
Lo
B
,
Yang
GZ
.
Direction sensitive fall detection using a triaxial accelerometer and a barometric pressure sensor
.
Annual International Conference of the IEEE Engineering in Medicine and Biology Society
:
IEEE
.
2011
. pp.
369
372
.
7.
Li
R
,
Wei
X
,
Xu
J
,
Chen
J
,
Li
B
,
Wu
Z
,
Wang
Z
.
Smart wearable sensors based on triboelectric nanogenerator for personal healthcare monitoring
.
Micromachines
2021
;
12
:
352
.
8.
Wang
ZL
.
Triboelectric nanogenerators as new energy technology for self-powered systems and as active mechanical and chemical sensors
.
ACS Nano
2013
;
7
:
9533
.
9.
Zhou
Z
,
Weng
L
, et al
Smart insole for robust wearable biomechanical energy harvesting in harsh environments
.
ACS Nano
2020
;
14
:
14126
14133
.
10.
Liu
G
,
Li
W
,
Liu
W
,
Bu
T
,
Guo
T
,
Jiang
D
,
Zhao
J
,
Xi
F
,
Hu
W
,
Zhang
C
.
Soft tubular triboelectric nanogenerator for biomechanical energy harvesting
.
Adv Sustainable Syst
2018
;
2
:
1800081
.
11.
Li
W
,
Lu
L
,
Yan
F
,
Palasantzas
G
,
Loos
K
,
Pei
Y
.
High-performance triboelectric nanogenerators based on TPU/mica nanofiber with enhanced tribo-positivity
.
Nano Energy
2023
;
114
:
108629
.
12.
Li
W
,
Lu
L
,
Kottapalli
AGP
,
Pei
Y
.
Bioinspired sweat-resistant wearable triboelectric nanogenerator for movement monitoring during exercise
.
Nano Energy
2022
;
95
:
107018
.
13.
Zhang
Z
,
Cai
J
.
High output triboelectric nanogenerator based on PTFE and cotton for energy harvester and human motion sensor
.
Curr Appl Phys
2021
;
22
:
1
5
.
14.
Zhang
B
,
Zhang
L
,
Deng
W
,
Jin
L
,
Chun
F
,
Pan
H
,
Gu
B
,
Zhang
H
,
Lv
Z
,
Yang
W
,
Wang
Z
.
Self-powered acceleration sensor based on liquid metal triboelectric nanogenerator for vibration monitoring
.
ACS Nano
2017
;
11
:
7440
7446
.
15.
Li
G
,
Li
L
,
Zhang
P
,
Chang
C
,
Xu
F
,
Pu
X
.
Ultra-stretchable and healable hydrogel-based triboelectric nanogenerators for energy harvesting and self-powered sensing
.
RSC Adv
2021
;
11
:
17437
17444
.
16.
Cai
J
,
Zhang
Z
.
A spring structure triboelectric nanogenerator for human gait monitoring system
.
Nano
2022
;
17
:
2250001
.
17.
Ahmed
A
,
El-Kady
MF
,
Hassan
I
,
Negm
A
,
Pourrahimi
AM
,
Muni
M
,
Selvaganapathy
PR
,
Kaner
RB
.
Fire-retardant, self-extinguishing triboelectric nanogenerators
,
Nano Energy
2019
;
59
:
336
345
.
18.
Collins
RT
,
Gross
R
,
Shi
J
.
Silhouette-based human identification from body shape and gait
.
Proceedings of the 5th IEEE International Conference on Automatic Face Gesture Recognition
.
Washington, DC: USA
:
IEEE
.
2002
. pp.
366
371
.
19.
Zhang
Y
,
Bai
Z
,
Li
J
,
Guo
J
.
Preparation and performance optimization of sealed wearable triboelectric sensor
.
J Donghua Univ (Netural Sci)
2022
;
48
:
40
47
.
20.
Sun
D
,
Qu
L
.
Survey on feature representation and similarity measurement of time series
.
J Front Comput Sci Technol
2021
;
15
:
195
205
.
21.
Okawa
M
.
Analysis of session variability for online signature verification using local stability-weighted DTW
.
2020 IEEE 9th Global Conference on Consumer Electronics (GCCE)
.
Kobe: Japan
:
IEEE
.
2020
. pp.
220
221
.

Gang Yang is an associate professor in the School of Communication and Information Engineering at Xi’an University of Posts and Telecommunications. His main research interests include ultra-wideband wireless positioning, location information services in the Internet of Things, and elderly fall detection and warning.

Lifang Wang is a graduate student in the School of Communication and Information Engineering at Xi’an University of Posts and Telecommunications. Her main research interests include sensing and energy collection of friction nanogenerators.

Jiayun Tian is a graduate student in the School of Communication and Information Engineering at Xi’an University of Posts and Telecommunications. Her main research interests are on wearable sensor-based fall warning technology for the elderly.