Gait cycle phase recognition method and system based on extreme gradient lifting algorithm

Abstract

The invention provides a gait cycle phase recognition method and system based on an extreme gradient lifting algorithm, and belongs to the technical field of health detection. Comprising the following steps: respectively preprocessing corresponding three-axis gravitational acceleration components, pressure values and step width data in a sole movement process; and carrying out gait recognition on the preprocessed data based on an extreme gradient lifting algorithm, visualizing a classification result through a normalized confusion matrix, dividing a flat ground walking period into a plurality of motion stages with obvious characteristics, determining starting and ending moments of each motion stage by identifying amplitude change inflection points of pressure signals, and carrying out gait recognition. And the recognition of the gait cycle phase is realized. According to the method, on the basis of ensuring the data acquisition quality, accurate recognition of the gait cycle phase can be realized, and a guidance basis is provided for health analysis of a tester.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of health detection, and particularly relates to a gait cycle phase recognition method and system based on an extreme gradient boosting algorithm. BACKGROUND

[0002] The statements in this section merely provide background information related to the present application and do not necessarily constitute prior art.

[0003] With the acceleration of the aging process and the enhancement of the national health concept, people's demand for improving the quality of life and enhancing the physical quality is increasing. With the rapid development of intelligent wearable devices in the field of health monitoring, gait analysis technology based on multi-sensor fusion has become an important means for evaluating motor function, preventing fall risk and guiding rehabilitation training.

[0004] However, the existing technology still has the following key technical problems in realizing high-precision gait cycle phase recognition: (1) The existing gait recognition mostly uses support vector machine, recurrent neural network and other algorithms. However, the support vector machine has poor generalization ability in high-dimensional gait data (such as multi-sensor fusion time sequence features) processing and is difficult to capture nonlinear relationships; the recurrent neural network is prone to gradient disappearance problem and is not sufficient for modeling long-time sequence gait cycle features. Moreover, such algorithms cannot optimize the periodicity and stage characteristics of gait data, resulting in large recognition error for the transition stage from the "initial landing period" to the "swing period", and cannot meet the demand for accurate division of gait stages in clinical rehabilitation.

[0005] (2) The existing gait data collection is mostly limited to plantar pressure data (such as heel and forefoot pressure) and foot Euler angle (inclination angle), and the data dimension is single, which leads to incomplete description of gait cycle. For example, relying only on pressure data cannot accurately distinguish the transition characteristics between "mid-stance" and "propulsion", which further affects the accuracy of gait stage recognition.

[0006] (3) The existing intelligent shoe products for gait recognition mostly use high-cost sensors (such as professional-grade inertial measurement units and high-density pressure sensor arrays) in hardware design; and complex signal processing modules are needed, which further increases the manufacturing cost. SUMMARY

[0007] In order to overcome the above-mentioned deficiencies of the prior art, the present application provides a gait cycle phase recognition method and system based on an extreme gradient boosting algorithm, which can realize accurate recognition of gait cycle phase on the basis of ensuring data collection quality, and further provide guidance for health analysis of testers.

[0008] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions: The first aspect of this invention provides a gait period phase recognition method based on an extreme gradient boosting algorithm.

[0009] A gait periodic phase recognition method based on the extreme gradient boosting algorithm includes: The tilt angle during foot movement is collected, and a unidirectional three-axis matrix rotation is performed to obtain the three-axis gravitational acceleration components; at the same time, the analog values ​​of pressure at the first metatarsal bone, the fourth metatarsal bone and the heel are collected, as well as the step width data during heel movement. The obtained triaxial gravitational acceleration components, pressure values, and step width data are preprocessed. The gait identification is performed on the preprocessed data based on the extreme gradient boosting algorithm, and the classification results are visualized by normalizing the confusion matrix. The flat walking cycle is divided into multiple movement stages with distinct characteristics. The start and end times of each movement stage are determined by identifying the inflection point of the amplitude change of the pressure signal, thus realizing the identification of the phase of the gait cycle.

[0010] Furthermore, the unidirectional three-axis matrix rotation includes: defining the gravitational acceleration vector at the initial moment and multiplying it with the Z-axis rotation matrix to achieve the first rotation calculation; multiplying the matrix vector after the first rotation calculation with the Y-axis rotation matrix to achieve the second rotation calculation; multiplying the matrix vector after the second rotation calculation with the X-axis rotation matrix to achieve the third rotation calculation, and using the matrix vector after the third rotation calculation as the three-axis gravitational acceleration component.

[0011] Furthermore, after collecting analog values ​​of pressure at the first and fourth metatarsal bones and the heel, the obtained analog values ​​are converted into voltage values; an effective voltage range is defined, and the voltage values ​​are mapped to pressure values.

[0012] Furthermore, the stride width data during the heel movement is acquired based on an ultrasonic sensor. Specifically, the time-of-flight method is used for distance measurement, and the distance is measured by calculating the time difference between ultrasonic wave transmission and reception.

[0013] Furthermore, the preprocessing includes Gaussian filtering and linear normalization operations.

[0014] Furthermore, the extreme gradient boosting algorithm is implemented based on the XGBoost classifier, which contains multiple decision trees. The leaf nodes in each decision tree predict and determine the cycle phase of the current gait based on the additive training strategy under the set objective function.

[0015] Furthermore, the multiple motion phases include the initial landing phase, the mid-support phase, the propulsion phase, and the oscillation phase.

[0016] The second aspect of the present invention provides a gait period phase recognition system based on an extreme gradient boosting algorithm.

[0017] The gait cycle phase recognition system based on the extreme gradient boosting algorithm includes: an acrylic box, a PCB board, an IMU unit, multiple pressure sensors, an ultrasonic sensor, and an insole. The IMU unit is integrated on the PCB board; The PCB board is placed inside the acrylic box and connected to the pressure sensor and ultrasonic sensor outside the acrylic box via DuPont wires, respectively. Multiple pressure sensors are respectively arranged at the positions marked on the insole at the first metatarsal bone, the fourth metatarsal bone, and the heel.

[0018] Furthermore, the IMU unit is used to acquire the tilt angle during the foot movement process and perform unidirectional three-axis matrix rotation to obtain the three-axis gravitational acceleration components. The pressure sensor is used to collect analog values ​​of pressure at the first metatarsal bone, the fourth metatarsal bone, and the heel of the foot; The ultrasonic sensor is used to collect stride width data during heel movement. The PCB board integrates a data processing module and a gait recognition module.

[0019] Furthermore, the data processing module is used to preprocess the obtained triaxial gravitational acceleration components, pressure values, and step width data; The gait recognition module is used to perform gait recognition on preprocessed data based on the extreme gradient boosting algorithm, and visualize the classification results through a normalized confusion matrix. The flat walking cycle is divided into multiple movement stages with distinct characteristics. The starting and ending times of each movement stage are determined by identifying the inflection point of the amplitude change of the pressure signal, thereby realizing the recognition of the phase of the gait cycle. The above one or more technical solutions have the following beneficial effects: (1) This invention uses the XGBoost algorithm for gait recognition. This algorithm combines gradient boosting and regularization techniques, which can effectively handle high-dimensional gait data and large datasets. Compared with support vector machines and recurrent neural networks, it can significantly reduce overfitting and enhance the model's generalization ability. At the same time, by optimizing data quality through Gaussian filtering and normalization preprocessing, and combined with targeted modeling of the periodic and phased features of gait, it can accurately identify the transition phase from the "initial landing phase" to the "swing phase", greatly reducing recognition errors and meeting the need for accurate gait phase segmentation.

[0020] (2) The gait cycle phase recognition system designed in this invention integrates an inertial measurement unit, a pressure sensor, and an ultrasonic sensor, and can simultaneously collect multi-dimensional data such as foot Euler angles, plantar pressure distribution, and stride width signals. Among them, the introduction of stride width signals makes up for the shortcomings of existing technologies that rely solely on pressure and Euler angle data, and can more comprehensively reflect the transition characteristics between the "mid-support phase" and the "propulsion phase" in the gait cycle, thereby completely depicting the dynamic changes of the gait cycle and providing data support for improving the accuracy of gait phase recognition.

[0021] (3) In the hardware design of the gait period phase recognition system, the present invention takes economy as an important consideration. It replaces professional-grade high-priced sensors by prioritizing the selection of cost-effective and stable components. At the same time, it simplifies the complexity of the signal processing module through hardware design of multi-sensor data fusion and efficient software algorithms such as Gaussian filtering. Under the premise of ensuring the functionality and reliability of the system, it significantly reduces the manufacturing cost.

[0022] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0023] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0024] Figure 1 This is a flowchart of the gait period phase recognition method based on the extreme gradient boosting algorithm in Embodiment 1 of the present invention.

[0025] Figure 2 This is a schematic diagram of unidirectional three-axis matrix rotation in Embodiment 1 of the present invention.

[0026] Figure 3 This is a schematic diagram of the results of the flat ground walking cycle division in Embodiment 1 of the present invention.

[0027] Figure 4 This is a modular structure diagram of the gait period phase recognition system based on the extreme gradient boosting algorithm in Embodiment 2 of the present invention. Detailed Implementation

[0028] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0029] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.

[0030] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0031] Example 1 This embodiment discloses a gait period phase recognition method based on the extreme gradient boosting algorithm.

[0032] like Figure 1 As shown, the gait periodic phase recognition method based on the extreme gradient boosting algorithm includes: Step S1: Collect the tilt angle during the foot movement process and perform a unidirectional three-axis matrix rotation to obtain the three-axis gravitational acceleration components; at the same time, collect the analog values ​​of the pressure values ​​at the first metatarsal bone, the fourth metatarsal bone and the heel, as well as the step width data during the heel movement process. Step S2: Preprocess the obtained triaxial gravitational acceleration components, pressure values, and step width data; Step S3: Based on the extreme gradient boosting algorithm, gait recognition is performed on the preprocessed data, and the classification results are visualized by normalizing the confusion matrix. The flat walking cycle is divided into multiple movement stages with distinct characteristics. The start and end times of each movement stage are determined by identifying the inflection point of the amplitude change of the pressure signal, thereby realizing the recognition of the phase of the gait cycle.

[0033] Based on the above process, this invention can achieve accurate identification of gait cycle phases while ensuring data acquisition quality, thereby providing guidance for the health analysis of test subjects. To facilitate understanding of the technical solution of this invention, the specific implementation methods of this invention will be further explained and described below.

[0034] In step S1, the tilt angle during the foot movement is collected, and a unidirectional three-axis matrix rotation is performed to obtain the three-axis gravitational acceleration components; at the same time, the analog values ​​of the pressure values ​​at the first metatarsal bone, the fourth metatarsal bone and the heel are collected, as well as the step width data during the heel movement.

[0035] This embodiment uses an inertial measurement unit (IMU), pressure sensors, and ultrasonic sensors as data acquisition devices. The insoles are marked with locations corresponding to the first metatarsal bone, fourth metatarsal bone, and heel of the human foot. In practice, the three pressure sensors are placed at the marked locations on the insoles corresponding to these locations to collect pressure data. The IMU is integrated with the lower-level computer on a PCB board. To protect the PCB and reduce external interference, it is placed in an acrylic box. The ultrasonic sensors are directly fixed to the side of the shoe to measure stride width data when walking on flat ground. This can be achieved through the following methods: Step S1-1: Based on the IMU unit, the tilt angle during the foot movement process is collected, and a unidirectional three-axis matrix rotation is performed to obtain the three-axis gravitational acceleration components.

[0036] In this embodiment, the IMU unit uses an MPU6050 inertial sensor designed via PCB and integrated with the lower-level computer on a single PCB board. The unidirectional three-axis matrix rotation includes: defining the initial gravitational acceleration vector and multiplying it by the Z-axis rotation matrix to achieve the first rotation calculation; multiplying the matrix vector obtained from the first rotation calculation by the Y-axis rotation matrix to achieve the second rotation calculation; multiplying the matrix vector obtained from the second rotation calculation by the X-axis rotation matrix to achieve the third rotation calculation, and using the matrix vector obtained from the third rotation calculation as the three-axis gravitational acceleration components.

[0037] Figure 2 This diagram illustrates the unidirectional three-axis matrix rotation of an IMU element, rotating sequentially around the Z, Y, and X axes. Initially, the IMU element's three-axis orientation is aligned with the geodetic coordinate system. After three rotations, the following values ​​are generated: Figure 2 The three coordinate system attitudes shown have different components of triaxial gravitational acceleration in each attitude.

[0038] Specifically, the triaxial gravitational acceleration vector of the IMU unit at the initial moment is shown in the following equation: (1) in, , and These represent the gravitational acceleration components on the X, Y, and Z axes before rotation of the unidirectional three-axis matrix, respectively. The scalar (modulus) representing gravitational acceleration, unless otherwise specified, is taken as standard gravity. .

[0039] After the IMU module rotates around the Z-axis by a certain angle, its three-axis gravitational acceleration vector can be represented by the product of the rotation matrix and the gravitational acceleration vector at the initial moment, that is: (2) in, , and These represent the gravitational acceleration components on the X, Y, and Z axes after rotating a certain angle around the Z-axis. This represents the rotation matrix along the Z-axis.

[0040] Similarly, the IMU element is then rotated by a certain angle around the Y-axis and X-axis in sequence, and the representations of the three-axis gravitational acceleration vectors are shown in the following equations: (3) (4) in, , and These represent the gravitational acceleration components on the X, Y, and Z axes after rotating a certain angle along the Y-axis. Represents the Y-axis rotation matrix; , and These represent the gravitational acceleration components along the X, Y, and Z axes after rotation of a unidirectional three-axis matrix. This represents the rotation matrix along the X-axis.

[0041] Furthermore, the matrix forms of the X-axis rotation matrix, Y-axis rotation matrix, and Z-axis rotation matrix are respectively expressed as: (5) (6) (7) in, This represents the cosine value of the roll angle. Represents the sine value of the roll angle; This represents the cosine value of the pitch angle. Represents the sine value of the pitch angle; The cosine value of the yaw angle. This represents the sine value of the yaw angle. Based on this, substituting equations (5)-(7) into equation (4), we get: (8) (9) in, This represents the roll angle around the X-axis of the aircraft's coordinate system. This represents the pitch angle around the Y-axis of the body coordinate system. The right-hand rule and the Z–Y–X Euler angles order are used (…). The angle is measured in radians. The gyroscope is used to determine the angle. and , can be represented as: (10) (11) in, and These represent the angular velocities of the IMU element rotating around the X and Y axes, respectively. This indicates the point at which integration ends, i.e., the point up to the current calculation. Indicates the starting time of integration.

[0042] Step S1-2: Collect simulated pressure values ​​at the first metatarsal bone, the fourth metatarsal bone, and the heel of the foot.

[0043] After acquiring analog pressure values ​​at the first and fourth metatarsal bones and the heel using pressure sensors, the analog values ​​are converted into voltage values. An effective voltage range is defined, and the voltage values ​​are mapped to pressure values. Specifically: First, the analogRead() function is used to read the analog values ​​(0 ~ 1023) and convert them into voltage values ​​(mV): map(value, 0, 1023, 0, 3300). Here, analogRead() is an analog-to-digital converter (ADC) read interface that returns an integer count from 0 to 1023, which linearly corresponds to the reference voltage. Based on this, the map(value,0,1023,0,3300) function performs a linear mapping, converting the ADC count value into a voltage value of 0~3300mV. Subsequently, based on the pressure sensor specifications, the minimum and maximum ranges are defined as (PRESS_MIN, PRESS_MAX), and the effective voltage range is defined as (VOLTAGE_MIN, VOLTAGE_MAX). The map() function is then used to map the voltage to the pressure value, ultimately converting it into a pressure value (N): Fdata_A0 / 12.6. Here, Fdata_A0 is the raw pressure data calculated from channel A0, and 12.6 is the sensor sensitivity or proportionality coefficient (unit: mV / N), which is obtained through calibration experiments and used to convert the voltage signal into a force (pressure) value N.

[0044] Steps S1-3: Collect stride width data during heel movement.

[0045] Stride width data during heel movement is acquired using the KS103 ultrasonic sensor. Specifically, the time-of-flight method is employed for distance measurement, calculating the time difference between ultrasonic wave transmission and reception. Compared to existing infrared and millimeter-wave solutions, the KS103 offers superior cost-effectiveness and interference resistance within a 5-20cm range.

[0046] In step S2, the obtained triaxial gravitational acceleration components, pressure values, and step width data are preprocessed, including Gaussian filtering and linear normalization.

[0047] When performing Gaussian filtering, the data processing module on the PCB first loads a CSV file containing sensor data, selects the first 7 columns of data (three-axis Euler angles, three sets of plantar forces, and gait phase labels) to be smoothed, and then uses the `gaussian_filterld` function from the SciPy library to perform one-dimensional Gaussian filtering on each column of data. In this embodiment, the sigma parameter is set to 0.8 to balance noise suppression and preservation of signal details. After filtering, the processed data is saved as a new CSV file.

[0048] Load the CSV file that has been processed by Gaussian filtering, explicitly specify the first 6 columns of data that need to be normalized, then calculate the minimum and maximum values ​​for each column individually. By normalizing (current value - minimum value) / (maximum value - minimum value), all data are linearly mapped to the [0,1] interval, achieving proportional scaling. After processing, save the normalized data as a new CSV file. The normalization calculation can be expressed as: (12) in, This indicates the current value of the data in this column. This indicates the minimum value of the data in this column. This indicates the maximum value of the data in this column.

[0049] In step S3, gait recognition is performed on the preprocessed data based on the extreme gradient boosting algorithm, and the classification results are visualized by normalizing the confusion matrix. The flat walking cycle is divided into multiple movement stages with distinct characteristics. The start and end times of each movement stage are determined by identifying the inflection point of the amplitude change of the pressure signal, thereby realizing the recognition of the phase of the gait cycle.

[0050] In practical implementation, firstly, the preprocessed CSV files are merged as input. Features (sensor data) and labels (gait categories) are separated, and LabelEncoder is used to convert the category labels into numerical values. The training and test sets are then divided in a 7:3 ratio to maintain consistent category distribution. LabelEncoder establishes a one-to-one integer encoding for discrete category labels, typically starting from 0. This encoding mapping remains consistent between training and testing, without changing the category meaning. Next, XGBoost classifier parameters are configured, and the model is trained. Performance metrics (accuracy, precision, recall, F1 score) are evaluated on the test set. Finally, the classification results are visualized using a normalized confusion matrix. This normalized confusion matrix is ​​normalized according to the number of true categories and rounding errors are corrected to ensure that the sum of each row is 1. In this embodiment, the configured XGBoost classifier parameters include: learning rate 0.17, maximum tree depth 14, subtree weight 0.7, and 400 decision trees.

[0051] The Extreme Gradient Boosting algorithm is based on the XGBoost classifier, which consists of multiple decision trees connected in series via a gradient boosting framework. Each new tree learns the prediction residuals of its predecessor tree. Its core components include: decision trees (feature space partitioning), an objective function (containing a loss function and a regularization term to control overfitting), and an additive training strategy (gradually optimizing predictions). The final classification is determined by voting based on the weights of the leaf nodes. Specifically, `colsample_bytree` (a hyperparameter of the XGBoost classifier during training) controls the feature sampling rate, `subsample` (another hyperparameter of the XGBoost classifier during training) adjusts the data subsampling ratio, and `learning_rate` (also known as `eta`, a hyperparameter of the XGBoost classifier during training) adjusts the contribution weight of each tree to the overall model. Together, they achieve efficient and accurate gait pattern recognition. Specifically, the leaf nodes in each decision tree, under the set objective function, predict and determine the phase of the current gait cycle based on the additive training strategy.

[0052] Based on the phase division criteria shown in Table 1, gait data in continuous gait cycles were meticulously labeled, providing sufficient training samples for subsequent gait recognition algorithms.

[0053] Table 1 Phase Division Criteria

[0054] In the process of analyzing gait characteristics, this invention employs time-domain signal analysis to focus on examining all the periodic fluctuation patterns of the plantar pressure curve. For example... Figure 3 As shown, the flat walking cycle is divided into four distinct movement stages. Each stage exhibits different pressure changes and gait characteristics. By identifying the inflection point of the amplitude change of the pressure signal, the start and end times of each stage can be accurately determined, thus achieving precise gait identification.

[0055] Specifically, such as Figure 3 As shown, the characteristics of Phase 1 are quite obvious: the pressure sensor data in the heel area shows a steep upward trend, indicating that the body's center of gravity begins to shift towards the supporting lower limb; subsequently, the sensor values ​​in the fourth metatarsal area also increase. In Phase 2, it can be seen that the pressure values ​​detected by the various pressure sensors on the sole of the foot continue to rise, and the center of gravity gradually shifts from the heel to the toes. The characteristics of Phase 3 are also quite prominent: the pressure data in the first and fourth metatarsal areas show a significant decrease and eventually tend to stabilize, indicating that the body's center of gravity has completed the forward shift process and the foot begins to generate propulsion. In Phase 4, it is clear that the readings of the various pressure sensors on the sole of the foot return to zero, indicating that no significant force has been applied, meaning that the foot has left the ground and is preparing for the next gait cycle.

[0056] Example 2 This embodiment discloses a gait cycle phase recognition system based on the extreme gradient boosting algorithm.

[0057] like Figure 4 As shown, the gait cycle phase recognition system based on the extreme gradient boosting algorithm includes an acrylic box, a PCB board, an IMU unit, multiple pressure sensors, an ultrasonic sensor, and an insole. The IMU unit is integrated on the PCB board; The PCB board is placed inside an acrylic box and connected to the pressure sensor and ultrasonic sensor outside the box via DuPont wires. The acrylic box is 3mm thick. Openings are made on the side of the box to allow the DuPont wires of the pressure and ultrasonic sensors to be routed out. During data collection, it was found that the wiring of the pressure and ultrasonic sensors was prone to loosening when shoes were worn. Therefore, a soldering process was used to fix the wiring, reducing external interference and ensuring the quality of the collected data. A 4000mAh rechargeable lithium battery from Suming Power, with an output voltage of 5V, was used, meeting the power supply requirements and allowing it to fit smoothly into the acrylic box.

[0058] Multiple pressure sensors are respectively positioned on the insole at the first metatarsal, fourth metatarsal, and heel locations marked thereon; the ultrasonic sensor is directly fixed to the side of the shoe worn by the tester to measure stride width data when walking on flat ground.

[0059] Furthermore, the IMU unit is used to acquire the tilt angle during the foot movement process and perform unidirectional three-axis matrix rotation to obtain the three-axis gravitational acceleration components. The pressure sensor is used to collect analog values ​​of pressure at the first metatarsal bone, the fourth metatarsal bone, and the heel of the foot; The ultrasonic sensor is used to collect stride width data during heel movement. The PCB board integrates a data processing module and a gait recognition module.

[0060] Furthermore, the data processing module is used to preprocess the obtained triaxial gravitational acceleration components, pressure values, and step width data; The gait recognition module is used to perform gait recognition on preprocessed data based on the extreme gradient boosting algorithm, and visualize the classification results through a normalized confusion matrix. The flat walking cycle is divided into multiple movement stages with distinct characteristics. The starting and ending times of each movement stage are determined by identifying the inflection point of the amplitude change of the pressure signal, thereby realizing the recognition of the phase of the gait cycle. While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A gait cycle phase recognition method based on an extreme gradient boosting algorithm, characterized by, The method comprises the following steps: Collecting the inclination angle during the movement of the foot bottom and performing one-way three-axis matrix rotation to obtain three-axis gravity acceleration components; At the same time, collecting the analog quantity of the pressure value at the first metatarsal bone, the fourth metatarsal bone and the heel of the foot bottom and the step width data during the movement of the heel; Preprocessing the obtained three-axis gravity acceleration components, pressure values and step width data; Based on the extreme gradient boosting algorithm, the gait recognition is performed on the preprocessed data, and the classification results are visualized through the normalized confusion matrix. The flat walking period is divided into multiple movement stages with obvious characteristics. The starting and ending time of each movement stage is determined by identifying the amplitude change inflection point of the pressure signal, and the gait cycle phase is recognized. 2.The gait cycle phase recognition method based on an extreme gradient boosting algorithm according to claim 1, wherein, The one-way three-axis matrix rotation comprises the following steps: defining the gravity acceleration vector at the initial moment and multiplying it by the Z-axis rotation matrix to realize the first rotation calculation; multiplying the matrix vector after the first rotation calculation by the Y-axis rotation matrix to realize the second rotation calculation; multiplying the matrix vector after the second rotation calculation by the X-axis rotation matrix to realize the third rotation calculation, and taking the matrix vector after the third rotation calculation as the three-axis gravity acceleration component. 3.The gait cycle phase recognition method based on an extreme gradient boosting algorithm of claim 1, wherein, After collecting the analog quantity of the pressure value at the first metatarsal bone, the fourth metatarsal bone and the heel of the foot bottom, the obtained analog quantity is converted into a voltage value; an effective voltage range is defined, and the voltage value is mapped into a pressure value. 4.The gait cycle phase recognition method based on an extreme gradient boosting algorithm according to claim 1, wherein, The step width data during the movement of the heel is obtained based on an ultrasonic sensor. Specifically, the time-of-flight method is used for distance measurement, and the time difference between the emission and reception of ultrasonic waves is used to realize distance measurement. 5.The gait cycle phase recognition method based on an extreme gradient boosting algorithm according to claim 1, wherein, The preprocessing comprises Gaussian filtering and linear normalization operations. 6.The gait cycle phase recognition method based on an extreme gradient boosting algorithm according to claim 1, wherein, The extreme gradient boosting algorithm is realized based on an XGBoost classifier. The XGBoost classifier contains multiple decision trees. The leaf nodes in each decision tree predict and determine the current gait cycle phase based on the additive training strategy under the set target function.

7. The gait cycle phase recognition method based on an extreme gradient boosting algorithm according to claim 1, wherein, The multiple movement stages include the initial landing period, the support middle period, the propulsion period and the swing period.

8. A gait cycle phase recognition system based on an extreme gradient boosting algorithm, characterized in that, The device comprises an acrylic box body, a PCB board, an IMU unit, multiple pressure sensors, an ultrasonic sensor and an insole. The IMU unit is integrated on the PCB board. The PCB board is placed in the acrylic box body and connected with the pressure sensors and the ultrasonic sensor outside the acrylic box body through DuPont wires. The multiple pressure sensors are arranged at the positions of the first metatarsal bone, the fourth metatarsal bone and the heel marked on the insole. The IMU unit is used to collect the inclination angle during the movement of the foot bottom and perform one-way three-axis matrix rotation to obtain three-axis gravity acceleration components.

9. The gait cycle phase recognition system based on an extreme gradient boosting algorithm of claim 8, wherein, The pressure sensor is used to collect the analog quantity of the pressure value at the first metatarsal bone, the fourth metatarsal bone and the heel of the foot bottom. The ultrasonic sensor is used to collect the step width data during the movement of the heel. The PCB board is integrated with a data processing module and a gait recognition module. The data processing module is used to preprocess the obtained three-axis gravity acceleration components, pressure values and step width data. ​ 10. The gait cycle phase recognition system based on an extreme gradient boosting algorithm of claim 9, wherein, ​ ​ The gait recognition module is configured to recognize the gait of the preprocessed data based on an extreme gradient boosting algorithm, and to visualize the classification result through a normalized confusion matrix, divide a walking cycle on a flat ground into a plurality of motion stages with obvious features, determine the starting and ending time of each motion stage by means of a change inflection point of the amplitude of the identified pressure signal, and realize recognition of the phase of the gait cycle.

Images