A new approach for contact-free detection of human gait based on continuous-wave bio-radar

Abstract

The human gait is a complex mechanism by which a variety of muscles and bones must work together in precise coordination to create characteristic human locomotion, which can be used as an identification or clinical gait characterization for detecting pathological conditions. In this paper, a contact-free detection system for human gait based on continuous-wave bio-radar is introduced. Through the spectrum analysis of the radar echo signal generated during human walking, the main frequency of the human gait signal is extracted. Correlation analyses are made with the data obtained from the acceleration sensor. The experimental results show that the bio-radar can detect the gait signal of human walking, and it is consistent with the result of acceleration and video monitoring.

Keywords

Human gait contact-free continuous-wave bio-radar correlation analyses

1. Introduction

The detection and identification of human body, which has wide applications in disaster relief, battlefield ambulance, counter-terrorism and security, is mainly using various advanced techniques to obtain human physiological or behavioral characteristics [1]. Traditional identification technology generally relies on fingerprint, iris, face, etc. However, the biological characteristics that these techniques relied on are easy to forge and it requires the subjects to be cooperative, which cannot be used in special situations [2, 3]. The gait detection, which don’t need cooperation and contact, has advantages such as not violated, easy acquisition as well as anti-optoelectronic camouflage, etc. [4, 5], thus attracts the attention of the scholars all over the world.

Various techniques have been developed to analyze and extract gait parameters, including visible video and ultrasound [6, 7]. The detection technology based on optical image contains various technologies such as target tracking, video acquisition, image processing and recognition, etc. In order to classification and recognition, the system traces and checks the targets captured by the video system. Then image processing technologies are applied to extract the gait information of the target and then compare with the sample data stored in the database [10].The detection technology based on ultrasonic, which utilizes the Doppler effect of mechanical wave, illuminates walking human body with high frequency mechanical wave. The echo will contain various Doppler information modulated by different human body components motions, such as legs and arms, which could be exploited to generate the human gait parameters by conducting signal processing technology on the echo [11, 12, 13].

Both contact-free detection methods above are able to obtain human gait parameters, but affected obviously by external factors such as light, visibility and obstacles [8, 9, 10, 11, 12]. However, bio-radar prevents light influence and can be all-weather identification. It also prevents the effect of smoke, dust and fog due to its penetrability, which is able to penetrate clothes, disguise and walls [11]. Therefore, the human gait contact-free detection technology based on bio-radar research has important research significance.

A single-frequency continuous wave (CW) radar prevents range estimates, but can capture an extremely high-resolution Doppler data, which can be used to extract gait features. Geisheimer J. L, etc., used a fully coherent, CW radar, which operated at a frequency near 10.5 GHz, to extract the Doppler signature from human walking [12]. Short time Fourier transform (STFT) and Chirplet Transform (CT) were employed to processing signals and various gait characteristics were captured [13]. Michael Otero, etc., introduced a method to capture gait characteristics by means of a CW radar system and Fourier Transformation (FT). They hold the point that the gait characteristics are unique and can be used as a classification [14]. Zhang Jun applied different time-frequency analysis methods to multi-component non-stationary signal analysis of the human gait radar echo, which laid a foundation for future research [15].

This paper discusses using a continuous wave bio-radar combine with acceleration sensors and video monitoring system to extract human gait information.

2. Materials and method

2.1 Experimental subject

In the study, 15 volunteers took part in the experiment after being told the purpose and the damage that may produce. Ethical approval was obtained from the Ethics Committee Board of the university for this study’s methods and conduct.

2.2 Experimental system

Experiment system is mainly composed of bio-radar system, acceleration sensors and video monitoring system. When human walking, all parts of the body including torso and limbs will generate typical characteristics of Doppler signals. We used bio-radar to collect the signals so that we could extract the human gait features. At the same time, acceleration sensors were used to get the movement parameters. The compare between two sets of data could find the relation. Video monitoring system was used in this experiment to confirm the condition of the human target as well as the relative position of the human target and the biological radar.

2.2.1 CW bio-radar system

When a CW bio-radar illuminates the human target, the echo will be modulated and contain the Doppler shift, which is caused by the micro-motion of the human target. Signal processing is used to extract the key features from the echo. The radar operates at a frequency near 10.4 GHz and is fully coherent. The transmit power of the bio-radar was 1 mW and a bilateral antenna array with the gain of 17 dB was used.The real situation of CW bio-radar system is shown in Fig. 1.

Figure 1.

The real situation of CW bio-radar system.

2.2.2 Acceleration sensors

Acceleration sensors have a variety of specifications and parameters. In consideration of gait recognition, a low frequency (0–20 Hz) acceleration sensor can meet the requirements. In addition, according to the different output, the acceleration sensors can also be divided into analog and digital. An analog acceleration sensor, which is small and light, has an output value of voltage. A digital acceleration sensor, which has a digital output, integrates ADC circuit, but it is larger than an analog sensor as a low frequency, analog acceleration sensor, ADXL335, which has a small size, lightweight, and low power consumption is employed in the experiment. It can measure the static acceleration of gravity in the skew detection applications, as well as the dynamic movement, impact or vibration caused by acceleration. In order to processing singles, a Biopac multi-channel physiological signal acquisition system is used to convert analog signals to digital signals. Acceleration sensors paste positions in the human body surface and the real experimental situations are shown in Fig. 2.

Figure 2.

Acceleration sensors pasting locations in the human body surface and the real experimental situations.

2.2.3 Video monitoring system

Video monitoring system is composed of camera, transmission, control, display and record. In order to collect video signals of the human gait and extract gait information through video analyzing, a solution based on virtual line detection to compute the corresponding parameters is adopted. A compare between the gait data from video analyze and bio-radar was made to find the connection. This experiment adopts two 1080 p high-definition cameras to record the human target. The video monitoring system is shown in Fig. 3.

Figure 3.

The video monitoring system.

2.3 Data processing and arithmetic

2.3.1 Spectrum analysis

As a non-stationary signal, human gait is obvious periodic, which can be conducted by Fourier Transform. The Fourier Transform (FT) of continuous time signals $x(t)$ can be represented as:

$\displaystyle f(t)=\frac{1}{2\pi}\mathop{\int}\limits_{-\infty}^{\infty}F({j% \omega})e^{j\omega t}d\omega$ (1)

Signal is calculated as a continuous spectrum in the formula above. However, we always get the discrete sampling values $x(nT)$ of continuous signal $x(t)$ in practice so that the frequency spectrum of the signal $x(t)$ should be calculated by the discrete signal $x(nT)$ . The discrete Fourier transform of a finite discrete signal $x(n)$ , $n=0,1,\ldots,N-1$ is defined as:

$\displaystyle X(k)=\mathop{\sum}\limits_{n=0}^{N-1}x(n)W_{N}^{kn}\quad k=0,1,% \ldots,N-1\quad W_{N}=e^{-j\frac{2\pi}{N}}$ (2)

In order to extract spectrum analysis, we use Fast Fourier Transformation (FFT), which is equal to DFT, as a convenience method to processing signals.

Figure 4.

a. Bio-radar signal and spectrum analysis of human torso; b. Bio-radar signal and spectrum analysis of human elbow; c. Bio-radar signal and spectrum analysis of human wrist; d. Bio-radar signal and spectrum analysis of human knee.

Figure 4.

Continued. e. Bio-radar signal and spectrum analysis of human ankle.

Figure 5.

a. Acceleration sensor signal and spectrum analysis of human torso; b. Acceleration sensor signal and spectrum analysis of human elbow; c. Acceleration sensor signal and spectrum analysis of human wrist; d. Acceleration sensor signal and spectrum analysis of human knee.

Figure 5.

Continued. e. Acceleration sensor signal and spectrum analysis of human ankle.

Figure 6.

a. PWVD of the signal from bio-radar; b. PWVD of the signal from the channel $z$ of acceleration sensors.

Figure 7.

a. Coherence analyze between the spectrums of the data from the bio-radar and from the channel $z$ of acceleration sensors on elbow; b. Coherence analyze between the spectrums of the data from the bio-radar and from the channel $z$ of acceleration sensors on wrist; c. Coherence analyze between the spectrums of the data from the bio-radar and from the channel $z$ of acceleration sensors on knee; d. Coherence analyze between the spectrums of the data from the bio-radar and from the channel $z$ of acceleration sensors on ankle; e. Coherence analyze between the spectrums of the data from the bio-radar and from the channel $z$ of acceleration sensors on torso.

2.3.2 Pseudo Wigner-Ville division (PWVD)

If the FT of signal $x(t)$ , $y(t)$ are $X({j\Omega})$ and $Y({j\Omega})$ , then the WVD of can be descirbe as:

$\displaystyle W_{x}({t,\Omega})=\mathop{\int}\limits_{-\infty}^{\infty}x\left(% {t+\frac{\tau}{2}}\right)x^{\ast}\left({t-\frac{\tau}{2}}\right)e^{-j\Omega% \tau}d\tau$ (3)

then the joint WVD of $x(t)$ and $y(t)$ can be describe as:

$\displaystyle W_{x,y}({t,\Omega})=\mathop{\int}\limits_{-\infty}^{\infty}x% \left({t+\frac{\tau}{2}}\right)y^{\ast}\left({t-\frac{\tau}{2}}\right)e^{-j% \Omega\tau}d\tau$ (4)

As an important tool to analysis non-stationary time-varying signals, WVD has improved some problems of STFT. WVD, which could be seen as a distribution of signal energy in time and frequency domain, has a clearly physical definition. By the convolution theorem, the WVD of multi-component signal will meet oscillate cross terms, which affect the time-frequency characteristics of the signal seriously. Thus how to suppress cross terms effectively is the main problem in the application of WVD [16, 17]. In this experiment, we used the PWVD algorithm to suppress the cross-term interference of multi-component signals.

2.3.3 Coherence analyze

Square coherence estimation which is usually used to estimate the consistency of the two groups of signals in each frequency, is a frequency function between 0 and 1 [18]. We adopt Welch ’s method to calculate the power spectral density of the signal $x$ and $y$ as $P_{xx}(f)$ and $P_{yy}(f)$ . Then the cross power spectrum density of $x$ and $y$ can be obtained by the following formula [19, 21]:

$\displaystyle C_{xy}(f)=\frac{|{P_{xy}(f)}|^{2}}{P_{xx}(f)P_{yy}(f)}$ (5)

According to the coherence analysis of MATLAB, we get coherence coefficient distributions in Fig. 4 based on the two groups of coherent signals in various frequencies. The horizontal axis represents the frequency and the vertical axis represents the correlation coefficient. When two signals are high consistent in some frequency components, the curve will show a peak.

3. Results

3.1 Bio-radar data

The measured signal is the combined reflections from all of the body components in motion, including arms, legs, torso, etc., all of which may be moving at different speed [20]. Signal processing will extract the features of Doppler frequency as well as the human gait parameters. In actual experiment, aluminum foil is pasted on the surface of human limbs, to enhance their ability of radar signal reflection.

Bio-radar data was collected at 1 KHz. Since human gait signal is usually less than 2 Hz, a 0.3–3 Hz band pass filter is adopted to signal pre-processing [22]. After FFT spectrum analysis, signal spectrums can be obtained.

3.2 Comparison

3.2.1 Acceleration sensor

The $x$ axis of acceleration sensors is perpendicular upward to the direction of human movement, $y$ axis is perpendicular to the direction of human movement to the left and $z$ axis keeps the same as the direction of human movement. When the human target moves towards to the bio-radar, the channel $z$ of acceleration sensors will get the maximum signals, which has a better coherence with the signals from the bio-radar than the other channels. In order to prevent the disturbances from other acceleration sensors, we only collect the data from one acceleration sensor as well as the bio-radar at a time. Considering the movements of wrists, elbows, knees and ankles are symmetric, only one set of acceleration sensor of each position is analyzed. The signal and spectrum analysis of acceleration sensors on various human body surfaces are as follows:

The horizontal axis represents the frequency and the vertical axis represents the amplitude of signals. We found that there was a consistency between the data obtained from bio-radar and the acceleration sensors. The frequency of the human limbs concentrated at about 0.7 Hz, namely that the frequency of human gait concentrated at about 0.7 Hz. The frequency of torso concentrated at about 1.4 Hz, namely that twice as the frequency of limbs, which was consistent with human movement characteristics in theory.

In order to obtain the time-frequency distribution of signals, PWVD was used to analyze the signal of radar and acceleration sensors. The results were shown as follows.

According to the PWVD of the signals, the signals from the bio-radar have a harmonious with the frequency of the signals from acceleration sensors, which equals the frequency of human gait.

According to the coherence analyze, the signals of bio-radar and acceleration sensors have a high coherence in 0.6 Hz to 0.8 Hz, namely that the main components of them are consistent. The coherence of torso finds a maximum in about 1.4 Hz, which is twice as the frequency of limbs and consistent with the result of signal frequency component analyze above.

3.2.2 Video monitoring system

Two cameras are used to record the details of the experiment from the front and back of the target. According to the video record, the frequency of the human target motion is about 0.7 Hz in the experiment, which is consistent with the result from acceleration sensors as well as the bio-radar.

4. Conclusion and discussion

This paper introduced a new approach for human gait contact-free detection system based on CW bio-radar and human gait parameters were extracted by the system. After filtering, we employed welch’s method to calculate the power spectrum of the signal, which contained the frequency of human gait. According to the PWVD of the signals, the signals from the bio-radar have a harmonious with the frequency of the signals from acceleration sensors, which equals the frequency of human gait. According to the comparison between the data from the acceleration sensors and bio-radar, we found that the frequency of limbs of a walking human is about 0.7 Hz as well as the torso is about 1.4 Hz. The experimental conclusion was verified by the video monitoring system.

In clinical practice, there are a lot of diseases which cause human lower limb lesions may affect the gait. Bio-radar can classify and quantify abnormal gait and provide reference for analyze and treat the illness. The gait analysis can also record the patients’ gait parameters and compare with the parameters of the healthy people for health assessment and rehabilitation. In addition, the human gait contact-free detection based on bio-radar can get gait parameters without effect human movement, which provides a convenient for the research of basic process and mechanism of abnormal gait and the activity of the joints, bones, muscles.

The next step of research will mainly focus on the human gait signal extraction and applications. In order to extract the signal of human gait with less interference, the adaptive line enhancement and blind separation signal processing technique can be used for human gait signal detection [23]. To intelligently recognize the detail of gait while human subject is walking with several Doppler radar sensors, the technique of deep learning can be used in realistic smart home scenarios [24, 25].

Footnotes

Acknowledgments

This research was funded by the Key Research and Development Program of Shaanxi Province, China (No. 2021ZDLGY09-07).

References

Chen

Ling

. Time-frequency transforms for radar imaging and signal analysis. Artech House, 2002; 66-68.

Gopikrishnan

Santhanam

. Neural network based accurate biometric recognition and identification of human iris patterns. Journal of Computer Science. 2010; 6(10): 475-481.

Goldstein

Harmon

Lesk

. Identification of human faces Proceedings of the IEEE. 1971; 59(5): 748-760.

Wang

, et al. Contact-free detection of obstructive sleep apnea based on wavelet information entropy spectrum using bio-radar. Entropy. 2016; 18(8): 306.

Liang

, et al. Detection and classification of finer-grained human activities based on stepped-frequency continuous-wave through-wall radar. Sensors. 2016; 16(6): 885.

Abdelgawwad

Mallofre

Patzold

. A Trajectory-Driven 3D Channel Model for Human Activity Recognition. IEEE ACCESS. 2021; 9: 103393-103406.

Zulcaffle

TMA

Kurugollu

Crookes

Bouridane

Farid

. Frontal view gait recognition with fusion of depth features from a time of flight camera. IEEE Transactions on Information Forensics and Security. 2019; 14(4): 1067-82.

Gluteal Activity During Gait in Patients with Chronic Ankle Instability Following Rehabilitation: A Randomized Controlled Trial. Journal of sport rehabilitation. 2022; 31(2): 158-164.

Zhu

Chen

Zhang

Zhao

Wei

. Rotating micro-doppler parameter estimation of ground wheeled vehicles based on SPWD and image enhancement. Optik. 2020; 219.

10.

Zhang

Jing

Jiang

. Device-free wireless sensing for human detection: The deep learning perspective. IEEE Internet of Things Journal. 2021; 8(4): 2517-39.

11.

Little

Boyd

. Recognizing people by their gait: The shape of motion. Videre: Journal of Computer Vision Research. 1998; 1(2): 1-32.

12.

Kalgaonkar

Raj

. Acoustic doppler sonar for gait recogination: Advanced Video and Signal Based Surveillance. in: Proc. IEEE Conference on Advanced Video and Signal Based Surveillance, 2007; pp. 27-32.

13.

Youngwook

Brian

. Hand gesture recognition using micro-doppler signatures with convolutional neural network. IEEE Access. 4: 7125-7130.

14.

Zhang

Pouliquen

Waxman

, et al. Acoustic micro-Doppler radar for human gait imaging. The Journal of the Acoustical Society of America. 2007; 121(3): 110-113.

15.

Zhang

. Basic gait analysis based on continuous wave radar. Gait & Posture. 2012; 36(4): 667-671.

16.

Rios

Youngwook

. Application of linear predictive coding for human activity classification based on micro-doppler signatures. Geoscience and Remote Sensing Letters IEEE. 2014; 11(10): 1831-1834.

17.

Mecklenbräuker

Hlawatsch

. The Wigner Distribution: Theory and Applications in Signal Processing. Elsevier. 1997; 121-145.

18.

Kay

. Modern spectral estimation. Pearson Education India. 1988.

19.

Rabiner

Gold

. Theory and application of digital signal processing. Englewood Cliffs, NJ, Prentice-Hall, Inc. 1975; 777-790.

20.

Welch

. The use of fast Fourier transform for the estimation of power spectra: A method based on time averaging over short, modified periodograms. IEEE Transactions on Audio and Electroacoustics. 1967; 15(2): 70-73.

21.

Sun

Wang

Zhang

Sun

Yuan

. Multiple walking human recognition based on radar micro-Doppler signatures. Science China Information Sciences. 2015; 58: 1-13.

22.

Carmine

Alessio

Karl

John

. Developments in target micro-Doppler signatures analysis: Radar imaging, ultrasound and through-the-wall radar. EURASIP J. Adv. Signal Process. 2013; 47: 1-18.

23.

Wang

Shen

Huangfu

, et al. Blind separation of doppler human gesture signals based on continuous-wave radar sensors. IEEE Transactions on Instrumentation and Measurement. 2019; 68(7): 2659-61.

24.

Xia

Ding

Wang

. Person Identification with Millimeter-Wave Radar in Realistic Smart Home Scenarios. IEEE Geoscience and Remote Sensing Letters. 2022; 19.

25.

Seifert

A-K

Grimmer

Zoubir

. Doppler radar for the extraction biomechanical parameters in gait analysis. IEEE Journal of Biomedical and Health Informatics. 2021; 25(2): 547-58.