Human motion capture system in sports performance based on Internet of Things technology and wireless inertial sensor

,


R E T R A C T E D A R T I C L E
improvements and can accurately record multi-point 3D projection in space under highspeed sampling conditions.
At present, the Internet of Things has become another change after the new technological form of the Internet of Things.This will break the traditional fitness method and become a new fitness standard.Zhu W studied mass sports and fitness based on the Internet of Things and intelligent systems.The digital terminal can monitor the health status at any time through the intelligent information collection device.At the same time, on the basis of collecting and analyzing health data, he provides residents with a personalized consultation plan, allowing residents to participate in healthy diet and exercise, so as to achieve the goal of health.However, the data analysis of this system is not accurate enough, leading to errors in the results [1].Motion capture systems are more and more widely used.Qiu Y discussed a motion capture system that uses camera images to obtain 3D human motion data.In this system, in order to detect the feature points of human body parts, he used some color markers and human contour images to be able to move the human body without restriction.And, he uses two cameras to estimate the 3D position of the feature points through a stereo method.However, the design of the system's image capture function is not very complete and needs to be further strengthened [2].In order to solve the problem of errors in monitoring human movement, Zhu G Z proposed a method of human movement pattern recognition based on multiple sensors, such as the data fusion of acceleration sensors and gyroscopes.This method receives the human body motion information output by the gyroscope and extracts the acceleration sensor.Kalman filtering algorithm is used to synthesize multisensor information, which improves the accuracy of posture angle measurement.He also established a hidden Markov model based on human postures, to recognize various human motion bodies according to the daily activity state of the human body.However, the scope of application of this method is relatively small and not practical enough [3].
The novelty of this article is (1) error analysis of the motion capture system, error modeling and error analysis of experimental data by both hardware algorithms and software.
(2) On the established motion capture system, the original data characteristics of the three sensors were analyzed, which confirmed the necessity of fusing the three sensors.

Internet of Things
The Internet of Things is an important part of modern intelligent systems and plays an important role in the development of information flow.Integrated detection, storage, computer technology and its applications and design concepts of the Internet of Things play an important role in industry and commerce, processing and everyday life [4].The object-based Internet platform, with the support of information technology, establishes a wide range of connections between objects, forming an interactive three-dimensional network, connecting the virtual world and the physical world.The basic characteristics of the Internet of Things can be summarized in the following three points: (1) full perception: use radio frequency identification devices, sensors and QR codes to receive information on objects anytime and anywhere, (2) reliable transmission: real-time and accurate transmission of information on objects through the network and (3) intelligent processing: use intelligent computer technology to analyze and process a large amount of data and information and then check intelligent objects [5,6].

Motion capture technology
Motion capture technology uses video equipment, motion sensors and other equipment to monitor the movement of some or all of the joints of the human or animal body, measure joint motion information and provide reference data for gait recognition and film production, and television [7].This technology currently has a wide range of applications in film and television production, interactive games, virtual reality and personnel training.
Kalman filter is an unbiased, linear and minimum variance optimal estimation theory [8].Understanding the mathematical model of the state vector and the observation vector, the characteristics of the statistical noise of the state and the means of observation and the initial value of the state of the system, the measured data and the sensor state equation can be used to derive the relationship between the system state medium and the observation data.Kalman filtering is divided into two stages: prediction and information [9,10].In the prediction phase, the state estimation at each moment is estimated based on the previous state value.So get the prediction equation: That is, the state at each moment is transformed from the state at the previous moment through the state transformation matrix, plus the control amount at the current moment [11].Since this system does not contain the control quantity at every moment, it can be simplified as: In addition to the estimation of the system state at each moment, it is also necessary to estimate the accuracy of its estimation, so it is expressed in the form of covariance in formula (3): In this way, the next state can be predicted based on the previous state.Since Kalman filtering is an iterative process, the state needs to be updated after each prediction [12].
After obtaining the current state estimation value, the current measurement value can be combined to obtain an optimal estimation value of the current state.Among them, C x is the observed value, and K x is the Kalman gain, which can be obtained by formula (5): Finally, find a new covariance matrix to complete this iteration. (1)

Capture of wireless inertial sensor
The angle between the device and the gyro can be obtained by reading the direction pointed by the axis by the method; that is, the angular velocity can be obtained.The magnetometer is used to test the strength and direction of the magnetic field [13].The nine-axis inertial measurement sensors currently used for motion signal collection including three-axis gyroscope sensors, three-axis acceleration sensors and three-axis magnetic induction sensors are relatively mature, which can better realize the position and positioning of the sensor nodes of human body motion capture, thereby realizing actions capture accurately.
The technology is divided into two parts: front-end hardware and back-end software.The main functions and contents of raw materials include the use of motion capture sensors for the collection of human motion data and the transmission of this data to computer motion data [14,15].The main features and content of the back-end software use a computer to efficiently process the collected traffic data, so that the computer can automatically recognize the activity category of the captured object and use the computer to reproduce the action and human interaction.

Action data preprocessing
In order to obtain a better experimental model, a Butterworth low-pass filter and lowpass filter will be processed before the data processing.The following is the preprocessing of the data: first, use a low-pass filter to process the collected data, so as to remove the high-frequency interference caused by the body from the inertial sensor [16].Then, since the initial state and the end state of the data we collect are using the initial state of standing, these are the action signals that we do not need, and the signals in this respect need to be filtered out.

Human motion model
According to the introduction of Kalman filter for the optimal estimation of carrier attitude, it can be seen that motion acceleration is one of the main sources of dynamic error, especially when the limbs are in motion [17].Assume that the motion acceleration at a certain joint is a, and the output of the accelerometer is b, which is used as a criterion for measuring and judging.Set the threshold value to T , then the effective output range of the accelerometer is shown in Eq. (7).
When the carrier is in motion and satisfies formula (8), the error of angle estimation is the maximum, and the following formula is obtained: Then, the maximum angle error judged according to the threshold is: Substituting different motion thresholds into Eqs.( 8) and ( 9), the angle estimation error is shown in Fig. 1.According to the threshold calculation result, the angle estimation error still exists, which cannot meet the actual usage requirements.Moreover, when there are different degrees of motion acceleration in the three axes, the amplitude may be close to 1 g.If only the acceleration amplitude is used to directly adjust the reliability of the X-, Y-and Z-axes, it is easy to cause misjudgment [18].

Inertial motion capture system
The current typical inertial motion capture system is composed of MEMS gyroscope, MEMS accelerometer and magnetometer, and sensors with different characteristics cooperate with each other.
Figure 2 shows a block diagram of a general motion capture system.Inertial motion capture systems usually consist of multiple rigid models to remove the human body.Install sensor nodes on each rigidity [19].A general sensor node is composed of a microprocessor, sensor, data transmission unit and power management unit.The microprocessor is responsible for reading and sending data and is the control section of the entire node.The sensor is used for 3D acquisition of acceleration, angular velocity and geomagnetic voltage.The power management unit is used to provide stable power to the entire node.The data collection unit collects data from all sensors and sends it to the (9) θ max = tan −1 (x/y) computer.According to actual needs, data acquisition units are generally into two categories: wired and wireless [20].The wireless data acquisition unit has no cable limitation, which is more convenient for users.Computer processing software uses related algorithms to calculate the position of each node and visualize the calculation results.

Kalman filtering algorithm analysis
Use sensors to collect two sets of data, static data and linear motion data.
The first is to analyze static data.Static data are to fix the sensor module on a horizontal tabletop and collect enough data for analysis.A total of 200 sets of data were collected.Select five groups of data as representative samples, as shown in Table 1.
From the front to the back are the acceleration Kalman filtering comparisons of the X-axis, Y-axis and Z-axis, where sample is the number of data sets.From the data in the table, it appears that the data of the three axes before filtering are around the three values of 0, 0 and 1, which can be considered as fluctuations caused by errors.After Kalman filter processing, the waveform becomes smoother, a lot of jitter is reduced, and the value is closer to the three values of 0, 0, and 1, and the error of the static state is reduced by 23.1%.
Then, the analysis of the linear motion is similar on the three axes, and for the convenience of calculation, the linear motion data are the sensor.
The module only moves 0.5 m in the positive direction of the X-axis, while the Y-axis and Z-axis should not move as much as possible.The movement collected 20 sets of data.Select 5 groups of data as representative samples, as shown in Table 2.
From the data in the table, it can be seen that the image of the system becomes smoother after Kalman filtering, and it is obviously reduced.At the same time, the shape of the original motion pattern is not lost, and the state of the motion process is basically maintained (stationary to acceleration to deceleration to zero speed).At the same time, the final distance after direct integration and Kalman filtering can be compared, and the Kalman data are closer to the real data.From this perspective, the effect of the Kalman filter can also be seen.

Effect of human capture
In order to show the movement of the limbs, the system arranges sensor nodes at the joints of the limbs to collect movement data.Take the arm as an example.The system arranges the sensor nodes at the three joints of the shoulder, elbow and wrist to collect the motion data of the three joints.Another node acts as an embedded gateway to collect the motion data sent by other nodes and then sends it to the data processing module through the serial port.This system distinguishes the data of different nodes through the different node number of each sensor node during data processing.And simultaneously process different node data separately.According to the initial position data processing module of the three joint points, the node position is initialized.After analyzing and processing the data, a three-dimensional drawing is performed according to the position of the three points at each moment, so as to obtain the corresponding movements of the limbs.
Let the arm bend and swing to the left to get the drawn three-dimensional animation (for display effect, select a part of the frame to draw the pattern) as shown in Fig. 3.

Results section
This article analyzes the background, application fields and development prospects of the limb motion capture system.The development status of wireless sensor network and body motion behavior capture is introduced.According to the requirements of the implementation of this system, the application of inertial navigation technology and wireless network of sensors, network technology for the recording of the motion detection pattern of the joints is proposed.In addition, this article uses inertial navigation technology to realize data processing, uses angular velocity data to correct the acceleration direction, maps the acceleration data on the sensor coordinate system to the geographic coordinate system and calculates the speed of each node by integrating the corrected acceleration data and displacement, so as to obtain the movement trajectory of each joint of the limb, and finally draw the movement state of the limb.Abbreviation 3D: Three-dimensional.

Authors' contributions
WX contributed to editing and data analysis.The author read and approved the final manuscript

Fig. 3
Fig. 3 Three-dimensional animation of arm movement

Table 1
Static data sample

Table 2
Samples of linear motion data