Athletic sports posture measurement algorithm based on multi-sensor combination

Abstract

With the arrival of sports wave, it is increasingly important to measure the posture of human body in movement. Aiming at the problems of low measurement accuracy and long measurement time in the traditional sports posture measurement algorithm, a sports posture measurement algorithm based on multi-sensor combination is proposed. The acceleration sensor, gyroscope and magnetoresistive sensor are combined to collect the sports posture data. The sliding mean filtering method and Z-score standardization method are used to reduce the noise and standardize the collected data. Based on the data processing results, the sports state is divided according to the dispersion characteristics, and the posture of any joint node is judged. According to the judgment results, the sports posture is measured. The simulation results show that the proposed method has less error, higher measurement accuracy and shorter measurement time. It can provide theoretical support for the practical application of motion attitude measurement.

Keywords

Multisensor joint athletic sports attitude measurement acceleration sensor gyroscope magnetoresistive sensor moving mean filtering

1. Introduction

With the rapid development of modern social economy and the continuous improvement of material living standards, people begin to pursue a healthy lifestyle, and sports gradually become a fashion [1]. At the same time, we have developed a lot of instruments to study human movement, to correct the defects of athletes’ movement details by obtaining the relevant data of life movement posture, so as to improve athletes’ competitive level. Motion capture system is the main tool for digitization of human motion information. It is widely used in virtual reality, human-computer interactive games, animation, film special effects, medical rehabilitation, sports and other fields [2]. Attitude measurement is an important basic link in the process of motion capture. Attitude measurement methods can be divided into mechanical tracking method, electromagnetic tracking method, acoustic tracking method, optical tracking method and acceleration tracking method, all of which have advantages and disadvantages [3]. As the current motion capture system is rich and diverse, it is very meaningful and practical to study the motion attitude measurement based on the new platform [4].

Sun et al. [5] use dual IMU to measure the attitude of an object relative to the coordinate system of the moving platform. Combining the characteristics of fast IMU measurement speed and high short-term accuracy with the characteristics that the visual measurement error does not diverge with time, the dual IMU and visual combined attitude measurement algorithm on the moving platform is studied. A multi rate adaptive extended Kalman filter (MAEKF) algorithm is proposed to fuse the measurement results of IMU and vision. The coordinate system calibration in inertial measurement is realized by orthogonal double vector calibration method and Q-method. Experimental results show that the combined attitude measurement algorithm can measure the attitude of the object relative to the moving platform, but the accuracy of the measurement results still needs to be improved. Tang and Chen [6] proposed a pure attitude measurement algorithm based on monocular vision odometer. Based on the classical monocular vision odometer calculation method, the calculation and optimization algorithm of attitude is directly initialized and modified by giving a depth, which solves the problem that it cannot work under the condition of pure rotation, and the actual problem that the camera cannot be accurately installed at the rotation center of the measured object. The installation bias model is established, the attitude measurement algorithm is improved, and the installation bias problem is effectively solved. At the same time, an off-line measurement scheme is proposed to identify the installation bias parameters. This algorithm solves the problem that can only occur under certain circumstances, but the overall effect of human body posture measurement is poor. Wang et al. [7] propose an attitude measurement algorithm based on acceleration separation algorithm. Firstly, the ellipsoid fitting method and modeling method are used to compensate the errors of accelerometer and gyroscope respectively to ensure the accuracy of the initial measurement data of MEMS sensor. Secondly, a method of separating motion acceleration is proposed to eliminate the influence of motion on the measurement data of accelerometer. Finally, combined with the acceleration separation algorithm, the high-precision attitude solution based on Kalman filter is realized. The attitude measurement result obtained by this algorithm is more accurate, but the measurement process consumes too long, which affects the overall work efficiency. Lao et al. [8] propose an automatic attitude measurement algorithm based on monocular vision. Firstly, the pose parameters of the stereo target are calculated by using the EPNP algorithm, and the pose parameters are brought in as the iterative initial value of the softpost algorithm. Secondly, the iterative process of calculating the pose of the softpost algorithm is combined with the stereo target to realize the automatic attitude measurement. Finally, in order to verify the accuracy of attitude measurement results, taking the high-precision two-dimensional turntable as the benchmark, the three-dimensional target is installed in the two-dimensional turntable, and the measurement data of target attitude are obtained by controlling the rotation angle of the turntable. Although the algorithm realizes the motion attitude measurement, but the measurement time is long, resulting in low measurement efficiency, still needs to be improved.

In view of the problems existing in the above algorithm, this paper proposes a sports posture measurement algorithm based on multi-sensor combination, and verifies the effectiveness and practicability of this algorithm through simulation experiments, improves the accuracy and efficiency of sports posture measurement, and provides effective data reference for the training of sports athletes and the later research of intelligent protective equipment.

2. Sports posture measurement algorithm based on multi-sensor combination

2.1 Sports posture data acquisition based on multi-sensor combination

As the first step in the process of sports posture measurement, data acquisition is not simple and has many limitations. For different sports scenes, we can’t choose sensors at will. First, we should evaluate the performance and universality of sensors. Because many actions in daily life are uncertain, the selected sensor combination should also adapt to various test schemes. In addition, the size, weight, signal, service life and other attributes of the sensor should also be considered. Wearable sensors commonly used in daily life include health monitoring sensors, inertial sensors and cameras. The sensors used for sports are three different types of sensors: acceleration sensors, gyroscopes and magnetoresistive sensors, which can qualitatively measure pressure, displacement and rotation [9].

Due to its own measurement principle and design defects, a single sensor cannot accurately measure the omni-directional sports posture information in the carrier coordinate system. The multi-sensor data combination technology can intelligently jointly process the original sports posture information measured by different types of posture measurement sensors. According to the different working environment of each sensor and the characteristics of data accuracy, the data features are extracted. After the joint algorithm is solved, the accurate carrier attitude information is finally obtained. Compared with a single attitude angle sensor, multi-sensor data joint technology can greatly improve the accuracy of attitude data, enhance the reliability and fault tolerance of measurement, and improve the real-time performance of output data and information utilization [10].

Therefore, this paper combines acceleration sensor, gyroscope and magnetoresistive sensor to collect sports posture data. After selecting the sensor, we need to arrange the sensor according to different motion scenes and recognition tasks. Because each part of the human body has different response ability to various movements, placing it in different parts will produce different performance. When deploying sensors, the same type of sensors can be placed in multiple parts of the body, or multiple types of sensors can be installed in the same part to establish a basic sports action recognition network. In short, the complexity of sensor layout is high, which is conducive to improving robustness [11].

Figure 1 shows several distributions of sensors on the human body. For example, the heart rate can be measured by pasting a heart rate sensor on the human chest, the sports information can be obtained by installing an accelerometer at the ankle, and the rotation angle can be obtained by installing a gyroscope on the wrist.

Figure 1.

Distribution of multi-sensor on human body.

In order to reduce the interference between sensors, it is necessary to design sensor platforms for different sports, such as smart bracelets and smart phones, to integrate the sensors required for specific measurement tasks, then select the appropriate tested population and easy to control test environment, formulate data acquisition standards and plan the scale of data, Finally, according to the signal transmission distance, the data collected by the sensor is sent to the processing end through wireless or wired communication [12]. Due to the mobility of sports, wireless communication technologies such as Wi Fi or Bluetooth are more suitable for data transmission in sports.

2.2 Data preprocessing

In the stage of sports posture data acquisition, the sports posture data signals collected by the sensor devices are usually affected by external or their own interference, including: the shaking generated by the body in the process of sports or the irregular signals generated by the peripheral environment; The multi-sensor itself has measurement error; Inaccurate signal caused by node position offset during movement. In practical application, the collected original data cannot be directly used for analysis and calculation. In order to obtain a more accurate signal, it is necessary to preprocess the signal after collecting it. Common preprocessing methods include denoising and normalization. The two preprocessing processes will be introduced below.

(1) Denoising processing

Due to the influence of environmental factors and sensor’s own factors, especially the interference of mechanical vibration during human movement, each sensor will inevitably introduce noise in the measurement process, and the measured real-time signal will fluctuate randomly. Reducing the noise in the measured data is a crucial step in the subsequent attitude solution. Sensor noise is mainly divided into inherent noise and random noise. The inherent noise is mainly caused by the output offset of the sensor itself; Random noise is mainly disturbed by environmental factors, especially mechanical vibration during human movement [13]. Each sensor will inevitably introduce noise in the measurement process, and the measured real-time signal will fluctuate randomly. Therefore, it is necessary to denoise the sensor data before attitude measurement. Among them, the removal of fixed noise is realized through sensor calibration, and there is a linear relationship between the output of the sensor and the measurement parameters, which is expressed as:

$\displaystyle Y_{\textit{ADC}}=S\times T+B$ (1)

where $Y_{\textit{ADC}}$ represents the original output value of the sensor, $S$ is the scale factor, that is, the sensitivity of the sensor, which can be selected by setting the sensor, $T$ is the real value of the measured physical quantity, and $B$ is the zero bias. The purpose of calibration is to calculate the parameter $B$ so that the true value $T$ of the sensor can be obtained.

The calibration process is briefly described as follows:

Place the acceleration sensor horizontally upward so that the $z_{b}$ -axis of the sensor is parallel to the gravity direction and vertically upward. The real value of the acceleration corresponding to the $z_{b}$ -axis is 1 g and the real value of the acceleration corresponding to the $x_{b}$ , $y_{b}$ -axis is 0 g. Record the output of the three axes of the sensor respectively, and bring the relevant parameters into Eq. (1) to obtain the zero deviation $B$ of each axis of the sensor.

The removal of sensor random noise is realized by filtering. Common digital filters, such as wavelet transform, filtering and adaptive filtering, have excellent denoising effect, but the filtering algorithm is complex, which is mainly realized in host computer software and special digital hardware. Moving average filtering has the characteristics of simple algorithm, high smoothness and low sensitivity, and has a good inhibitory effect on periodic interference [14]. The process of sliding filtering is the process of solving the moving average. The moving average is to calculate the average of multiple continuous $m$ -term sequences from $n$ -term data sequence. The first item of the first continuous $m$ -item sequence starts from the starting point of the $n$ -item data sequence; The starting point of the second continuous $m$ -term sequence is the second term of the original $n$ -term data sequence; By analogy, the first term of the last $m$ -term sequence is the $n-m+1$ -term of the $n$ -term sequence [15]. Therefore, in this paper, the sliding mean filter is used to denoise the sensor signal to reduce the interference of random noise such as mechanical vibration on the sensor measurement. The expression is:

$\displaystyle y(i)=\sum\limits_{n=0}^{9}{x(i-n)}/10$ (2)

where $x(i)$ represents the data collected by the sensor in real time, and $y(i)$ represents the calculated moving average value, that is, the filtered data.

(2) Normalization processing

Because the range of motion data from different sensors is different, the scale of data will also affect their contribution to simultaneous interpreting. After standardized processing, the size and dimension of all sports posture data can be controlled, and can be compared in the same scale, which solves the problem of biased estimation. In a word, it is to treat all sports posture data equally, eliminate the differences between the data, and reduce the impact of the above reasons on the posture measurement effect [16].

Several common standardization methods include log function transformation, ATAN function transformation, min max standardization and Z-score standardization. The min max standardization is to make the original sensor data result fall into the [0, 1] interval through linear transformation:

$\displaystyle Y^{\prime}=\frac{Y-\min}{\max-\min}$ (3)

where $Y$ represents the original data and $\min$ , $\max$ represents the minimum and maximum values in $Y$ [17].

Z-score standardization is to process the standard deviation and average value of the original data, and its conversion function is:

$\displaystyle Y^{\prime}=\frac{Y-\bar{Y}}{T}$ (4)

where $\bar{Y}$ , $T$ represents the mean and standard deviation of data $Y$ respectively.

2.3 Division of sports status

Based on the preprocessed sports posture data, the sports status is divided. The degree of dispersion represents the difference between the values of the observed variables. The difference between the sample values of the sensor signal is defined as the degree of dispersion. The expression of dispersion $d_{n}^{x}$ is:

$\displaystyle d_{n}^{x}=\left|{\varpi_{n}^{x}-\varpi_{n-1}^{x}}\right|$ (5)

Taking the angular velocity as an example, $\varpi_{n}^{x}$ represents the angular velocity data of axis $x$ at time $n$ , $\varpi_{n-1}^{x}$ represents the angular velocity data of axis $x$ at time $n-1$ , and $d_{n}^{x}$ represents the angular velocity difference between the angular velocity of axis $x$ of the sensor at time $n$ and the previous time.

The action data includes angular velocity data and acceleration data. In order to realize the accurate division of action, it is necessary to comprehensively consider the data characteristics of each sensor. Use $D_{n}^{x}$ to represent the dispersion of acceleration sensor data at time $n$ , $d_{n}^{g}$ to represent the dispersion of angular velocity sensor data at time $n$ , and $d_{n}^{a_{x}},d_{n}^{a_{y}},d_{n}^{a_{z}},d_{n}^{g_{x}},d_{n}^{g_{y}},d_{n}^{g% _{z}}$ to represent the dispersion of acceleration and angular velocity of each axis respectively. Then the expression of $D_{n}^{a},D_{n}^{g}$ is:

$\displaystyle D_{n}^{a}=d_{n}^{a_{x}}+d_{n}^{a_{y}}+d_{n}^{a_{z}}$ (6) $\displaystyle D_{n}^{g}=d_{n}^{g_{x}}+d_{n}^{g_{y}}+d_{n}^{g_{z}}$ (7) $\displaystyle\gamma_{n}=\left\{{\begin{array}[]{l}0,D_{n}^{a}<\lambda_{a}\text% { and }D_{n}^{g}<\lambda_{g}\\ 1,D_{n}^{a}\geqslant\lambda_{a}\text{ or }D_{n}^{g}\geqslant\lambda_{g}\\ \end{array}}\right.$ (8)

In the static state, the dispersion of acceleration and angular velocity are kept below the threshold $\lambda_{a},\lambda_{g}$ respectively. In the sports state, the sensor data will change rapidly with the athletes’ actions, and the dispersion can reflect the difference degree of sensor data. Therefore, the division of athletes’ limb states can be realized according to the dispersion characteristics [18]. $\gamma_{n}$ represents the state of the athlete’s limbs at time $n$ , when $\gamma_{n}$ is 0, it represents the static state, and when $\gamma_{n}$ is 1, it represents the motion state.

2.4 Sports posture measurement

According to the above divided sports state, the sports posture is measured. Firstly, the attitude of any joint node is judged, the coordinate system is established on the joint node, and the information collected by the joint acceleration sensor, gyroscope and magnetoresistive sensor is transformed into the form of coordinates. The expression is:

$\displaystyle R_{a}=\left[{{\begin{array}[]{*{20}c}{\sigma_{a}}&0&0\\ 0&{\sigma_{a}}&0\\ 0&0&{\sigma_{a}}\\ \end{array}}}\right]$ (9)

where $R_{a}$ represents the coordinate representation of the information collected by the acceleration sensor. Similarly, the gyro collected data $R_{g}$ and the magnetoresistive sensor data $R_{m}$ are expressed in the form of the above formula.

Assuming that the sports posture of the research target is running, the joint speed and moving position of both legs are calculated [19]. The calculation formula of leg joint angular velocity is:

$\displaystyle\theta=\theta_{0}\bar{\omega}_{\theta}\cos(\bar{\omega}_{\theta}t)$ (10)

where $\theta$ represents the inclination angle of the global coordinate system, $\bar{\omega}_{\theta}$ represents the inclination angle of the leg joint relative to the ground, and $t$ represents the duration. Then the linear velocity of the inertial node is:

$\displaystyle v=L\theta=L\theta_{0}\bar{\omega}_{\theta}\cos(\bar{\omega}_{% \theta}t)$ (11)

where $L$ is the distance between the knee node of the leg and the ground. After superposition of gravity acceleration, the coordinates of leg joint movement can be expressed as:

$\displaystyle q=\left[{{\begin{array}[]{*{20}c}x\\ y\\ z\\ \end{array}}}\right]=\left[{{\begin{array}[]{*{20}c}{v_{x}-g\sin(\theta)}\\ {v_{y}-g\sin(\theta)}\\ 0\\ \end{array}}}\right]+\left[{{\begin{array}[]{*{20}c}{R_{a}}\\ {R_{g}}\\ {R_{m}}\\ \end{array}}}\right]$ (12)

Thus, the coordinates of all nodes are calculated by using the above formula to form the coordinates of the whole sports posture and form the coordinate matrix set $Q$ , which is expressed as:

$\displaystyle Q=[{q_{0},q_{1},\ldots,q_{15}}]^{T}$ (13)

And set $Q$ needs to meet the following conditions:

$\displaystyle|Q|^{2}=(q_{0}^{2}+q_{1}^{2}+,\ldots,+q_{17}^{2})+\mu^{2}=1$ (14)

where $\mu$ represents the value of the identity matrix. Assuming that the attitude angle of each joint is divided into pitch angle and roll angle, expressed by $\varphi,\phi$ respectively, the conversion relationship between joint position coordinates and attitude matrix is as follows:

$\displaystyle\left[{{\begin{array}[]{*{20}c}\mu\\ {q_{0}}\\ \vdots\\ {q_{15}}\\ \end{array}}}\right]=\left[{{\begin{array}[]{*{20}c}{\cos\frac{\phi}{2}\cos% \frac{\theta}{2}\cos\frac{\varphi}{2}+\sin\frac{\phi}{2}\sin\frac{\theta}{2}% \sin\frac{\varphi}{2}}\\ {\cos\frac{\phi}{2}\cos\frac{\theta}{2}\sin\frac{\varphi}{2}-\sin\frac{\phi}{2% }\sin\frac{\theta}{2}\cos\frac{\varphi}{2}}\\ \vdots\\ {\sin\frac{\phi}{2}\cos\frac{\theta}{2}\cos\frac{\varphi}{2}-\cos\frac{\phi}{2% }\sin\frac{\theta}{2}\sin\frac{\varphi}{2}}\\ \end{array}}}\right]$ (15)

The value of attitude angle can be obtained by converting the matrix. Then, the value of sports acceleration is calculated from the three directions $x, y, z$ , and the acceleration at time $i$ in the direction $x$ can be calculated by the following formula:

$\displaystyle\textit{SMA}=\sum\limits_{i=1}^{n}{(\textit{Accx}(i)-\textit{Accx% }(i-1))}$ (16)

According to the calculation method in the above formula, the acceleration values in the $y$ and $z$ directions are calculated. Taking the combined acceleration as the measurement index of sports intensity, the calculation formula is:

$\displaystyle\textit{ACC}=\sqrt{\textit{Accx}^{2}+\textit{Accy}^{2}+\textit{% Accz}^{2}}$ (17)

This is used as the standard to judge the posture, and the measurement results of sports posture are obtained.

3. Simulation experiment analysis

In order to verify the effectiveness of the sports posture measurement algorithm based on multi-sensor combination in practical application, a simulation experiment is carried out with basketball players as the experimental object.

The joint position of human body in sports is represented in the form of nodes. The node setting is shown in Fig. 2.

The mpu60506 axis sensor produced by nvensense company in the United States integrates acceleration sensor, gyroscope and magnetoresistive sensor, which can output the measured original inclination data through IC interface. Its main characteristics are shown in Table 1.

Table 1
Sensor parameter settings

	Parameter setting
Minimal volume	1524 mm $\times$ 1524 mm $\times$ 2 mm
Range	The range of acceleration sensor can reach $\pm$ 16 g, and the range of gyroscope about angular
	velocity can reach $\pm$ 2000 ${}^{\circ}$ /s
Resolving power stability	Acceleration sensor: 61 $\times$ 10 ${}^{\circ}$ /g
	Gyroscope accuracy: 76 $\times$ 10-3 ${}^{\circ}$ /s
Data output frequency	Acceleration: 0.001 g
Minimal volume	Gyroscope: 0.02 ${}^{\circ}$ /s
Range	100 Hz

Figure 2.

Distribution of human body nodes in sports.

There are many body parts involved in the process of sports, so the more sensor nodes are placed in the body, the richer the attitude information will be obtained. In this paper, the legs and arms are regarded as the key parts in the process of human movement, and the human attitude is inferred by observing the movement status of the limbs. Therefore, the placement position of the sensors is shown in Fig. 3. In this paper, four sensor nodes are attached to human arms and legs to detect the movements of upper and lower limbs in human movement. And send the collected data information to the base station through the wireless communication protocol. The base station completes the data acquisition and uploads it to the host computer through the serial port for further data analysis.

Figure 3.

System topology.

Using the sports posture measurement algorithm based on multi-sensor combination proposed in this paper, the pure posture measurement algorithm based on monocular vision odometer proposed in Literature [6] and the posture measurement algorithm based on acceleration separation algorithm proposed in Literature [7], measure the sports posture angle, and compare the error between the measurement results and the actual test results. The comparison results are shown in Fig. 4.

Figure 4.

Comparison results of measurement errors of sports posture angle.

It can be seen from Fig. 4 that the motion attitude Angle measurement results based on the multi-sensor joint motion attitude measurement algorithm proposed in this paper are consistent with the actual test results. The pure attitude measurement algorithm based on the monocular visual odometer proposed in Literature [6] and the attitude measurement algorithm based on the acceleration separation algorithm proposed in Literature [7] have great fluctuations compared with the actual test results, which are quite different from the actual test results. The results show that the motion attitude measurement algorithm based on multi-sensor combination can accurately measure the motion attitude Angle and has high accuracy.

Using the sports posture measurement algorithm based on multi-sensor combination proposed in this paper, the pure posture measurement algorithm based on monocular vision odometer proposed in Literature [6] and the posture measurement algorithm based on acceleration separation algorithm proposed in Literature [7], measure the sports acceleration, and compare the error between the measurement results and the actual test results. The comparison results are shown in Fig. 5.

Figure 5.

Comparison results of measurement errors of sports acceleration.

According to Fig. 5, there is an error between the results of sports acceleration measurement based on the multi-sensor joint sports posture measurement algorithm proposed in this paper and the actual test results, but the error is small, while the pure posture measurement algorithm based on monocular vision odometer proposed in Literature [6] and the posture measurement algorithm based on acceleration separation algorithm proposed in Literature [7]. The error between the measurement results of sports acceleration and the actual test results is large, which shows that the sports posture measurement algorithm based on multi-sensor combination proposed in this paper can accurately measure sports acceleration.

The sports posture measurement algorithm based on multi-sensor combination proposed in this paper, the pure posture measurement algorithm based on monocular vision odometer proposed in Literature [6] and the posture measurement algorithm based on acceleration separation algorithm proposed in Literature [7] are used to compare and analyze the sports posture measurement accuracy. The comparison results are shown in Fig. 6.

Figure 6.

Comparison results of sports posture measurement accuracy of three algorithms.

According to Fig. 6, the maximum accuracy of the motion attitude measurement algorithm based on multi-sensor joints proposed in this paper is 100%, while the maximum accuracy of the pure attitude measurement algorithm based on monocular vision odometer proposed in Literature [6] is only 80% when it is used to measure the motion attitude. Literature [7] the separation algorithm based on acceleration of attitude measurement algorithm for motion measurement of the highest accuracy is 90%, and the method of minimum measuring precision is still higher than the other two methods of comparison of the highest accuracy, this paper put forward based on multi-sensor combination of motion measurement algorithm has higher precision and better measuring result.

In order to further verify the effectiveness of this algorithm, the sports posture measurement algorithm based on multi-sensor joint proposed in this paper, the pure posture measurement algorithm based on monocular vision odometer proposed in Literature [6] and the posture measurement algorithm based on acceleration separation algorithm proposed in Literature [7] are used to compare and analyze the sports posture measurement time, The comparison results are shown in Fig. 7.

Figure 7.

Comparison results of sports posture measurement time of three algorithms.

As can be seen from Fig. 7, the motion posture measurement algorithm based on multi-sensor joints proposed in this paper is significantly superior to the other two comparison methods in terms of measurement time, and consumes less 10 seconds than the motion posture measurement algorithm in Literature [6] and Literature [7], indicating that the proposed algorithm is more efficient in motion posture measurement.

4. Conclusion

Due to the problems of low measurement accuracy and long measurement time of the currently used algorithm for motion attitude measurement, a motion attitude measurement algorithm based on multi-sensor joint was proposed. The simulation results show that compared with the comparison method, the measurement results of motion attitude Angle proposed in this paper are consistent with the actual results. The error between the measured result and the actual result is small. High measurement accuracy, the highest can be close to 100%; The motion attitude measurement process took at least 10 seconds less than the other two methods. Based on the above results, the application effect of the proposed method is obviously better than the other two methods, indicating that the proposed method can quickly and accurately measure the motion posture. The algorithm can not only be applied to sports competition, but also play a certain role in bionic robots and film and television fields. Therefore, the motion attitude measurement algorithm based on multi-sensor combination proposed in this paper has strong application value.

References

Sun

Zhong

Zhang

. Simulation of helicopter landing real-time target positioning attitude prediction. Comp Simul. 2019; 36(6): 114-119.

Hamrah

Warier

Sanyal

. Finite-time stable estimator for attitude motion in the presence of bias in angular velocity measurements. Automatica. 2021; 132(2): 109815.

Fan

Jin

Wang

. Wearable motion attitude detection and data analysis based on Internet of Things. IEEE Access. 2019; 81: 1327-1338.

Jin

Yuan

Chen

. Design and implementation of attitude and heading reference system with extended Kalman filter based on MEMS multi-sensor fusion. Int J Uncertainty, Fuzz Knowl-Based Syst. 2021; 29(1): 157-180.

Sun

Huang

Wang

Guo

. Dual IMU and vision combined attitude measurement algorithm for moving platform. J Sens Technol. 2018; 31(9): 1365-1372, 1376.

Tang

Chen

. Design of pure attitude measurement algorithm based on monocular vision odometer. J Instrum. 2019; 40(5): 249-256.

Wang

Zhang

Liu

. Research on attitude measurement method based on acceleration separation algorithm. J Sens Technol. 2018; 31(12): 1848-1852.

Lao

Zhang

Xiong

Zhou

. Automatic attitude measurement method based on monocular vision. J Photonics. 2019; 48(3): 159-165.

Zhang

Liu

Yang

Liu

Chen

Yuan

. Attitude measurement and simulation of two-dimensional position sensitive detector. Optik. 2019; 184: 399-411.

10.

Shan

Zhao

Pan

Wang

. Sea-sky line and its nearby ships detection based on the motion attitude of visible light sensors. Sens. 2019; 19(18): 4004.

11.

Sun

Wang

Cai

Yang

Wen

. In-motion attitude and position alignment for odometer-aided SINS based on backtracking scheme. IEEE Access. 2019; 7: 20211-20224.

12.

Hui

. Detection algorithm of regional peak motion based on acceleration sensor. Comput Commun. 2020; 154: 173-179.

13.

Eismont

Kovalenko

Nazarov

Nazirov

Korotkov

Pogodin

Mzhelskii

Mikhailov

Ditrikh

Tregubov

. Orbital motion and attitude control of the Spectrum-Rntgen-Gamma space observatory. Astron Lett. 2020; 46(4): 263-274.

14.

Zhou

Xiong

Wang

. Multi MEMS sensor attitude measurement system based on EKF. J Tsinghua U: Naivt Sci Ed. 2006; 46(1): 1857-1859, 1863.

15.

Wang

Cheng

You

Zhang

Tang

Yang

. Motion attitude measurement algorithm based on multi-sensor. Comr Syst Appl. 2015; 24(9): 134-139.

16.

Xia

Luo

. Indoor location algorithm based on multi-sensor data fusion. J Shanghai Norm Univ: Nat Sci Ed. 2015; 44(1): 65-72.

17.

Wan

Kang

Lin

. Attitude calculation of four rotor aircraft based on multi-sensor data fusion. Sci Technol Guide. 2014; 32(19): 31-35.

18.

Yang

Liu

. Robot attitude measurement system based on multi-sensor information fusion. Comput Eng Design. 2016; 37(6): 1681-1686.

19.

Sheng

Zhang

Xing

. Adaptive attitude measurement algorithm based on MEMS gyroscope and accelerometer. J Tes Technol. 2018; 32(4): 277-284.

Athletic sports posture measurement algorithm based on multi-sensor combination

Abstract

Keywords

1. Introduction

2. Sports posture measurement algorithm based on multi-sensor combination

2.1 Sports posture data acquisition based on multi-sensor combination

(1) Denoising processing

(2) Normalization processing

Table 1 Sensor parameter settings

References

Table 1
Sensor parameter settings