Gait recognition model construction method and system based on plantar pressure characteristics

Abstract

The application discloses a gait recognition model construction method and system based on plantar pressure characteristics, and comprises the following steps: collecting VGRF signals of left and right feet of a subject; calculating COP trajectory information of the left and right feet independently by using the VGRF signals; performing significant difference analysis on the COP trajectory information between a normal gait posture group and an abnormal gait posture group, and obtaining plantar pressure characteristics; converting the plantar pressure characteristics into a Gram angle field diagram, inputting the Gram angle field diagram into an SGAN network model for network model training, obtaining a trained SGAN network model, and taking the trained SGAN network model as a gait recognition model. The application can obtain a precise and efficient gait recognition model, has strong adaptability, and provides technical support for extraction of gait characteristics.

Description

Technical Field

[0001] This invention relates to the field of gait recognition, and more specifically to a method and system for constructing a gait recognition model based on plantar pressure features. Background Technology

[0002] With the continuous development of intelligent technology, gait recognition technology has become a research hotspot in many fields, including medicine, security, and sports. Gait recognition is an important technology for identity verification, health monitoring, and motion analysis by analyzing an individual's walking patterns. The key to gait recognition lies in building an efficient and accurate model.

[0003] Currently, traditional gait recognition technologies are mainly based on rules or specific parameter settings. However, these technologies have poor adaptability among different populations and in different environments. Their accuracy and reliability in extracting or classifying gait features are low. Therefore, there is a need for a gait recognition model construction method and system based on plantar pressure features that can solve the above problems. Summary of the Invention

[0004] In view of this, the purpose of this invention is to overcome the defects in the prior art and provide a method and system for constructing a gait recognition model based on plantar pressure features, which can obtain an accurate and efficient gait recognition model with strong adaptability, and provides technical support for the extraction of gait features.

[0005] The gait recognition model construction method based on plantar pressure features of the present invention includes the following steps:

[0006] S1. Collect VGRF signals from the left and right feet of the subject; the subject includes those with normal gait and those with abnormal gait.

[0007] S2. Calculate the independent COP trajectory information for the left and right feet using the VGRF signal;

[0008] S3. A significant difference analysis of COP trajectory information between individuals with normal gait and a control group with abnormal gait was performed to obtain plantar pressure characteristics;

[0009] S4. Convert the plantar pressure features into a Gram angle field map, and input the Gram angle field map into the SGAN network model for network model training to obtain a trained SGAN network model. Use the trained SGAN network model as a gait recognition model.

[0010] Furthermore, the COP trajectory information of the left foot is determined according to the following formula:

[0011] ;

[0012] ;

[0013] The origin is the center of symmetry of the left and right feet, the Y-axis is the axis of symmetry of the left and right feet, and the X-axis is the straight line passing through the origin and perpendicular to the Y-axis. The X-axis coordinate of the left foot COP trajectory is... The Y-axis coordinate of the left foot COP trajectory. This is the number of the pressure sensor for the left foot. This represents the number of pressure sensors on the left foot. Indicates the first The x-coordinate value of each sensor location, For the first The ordinate values ​​of each sensor location; Indicates the left foot The VGRF signal value corresponding to each sensor.

[0014] Furthermore, the COP trajectory information of the right foot is determined according to the following formula:

[0015] ;

[0016] ;

[0017] The origin is the center of symmetry of the left and right feet, the Y-axis is the axis of symmetry of the left and right feet, and the X-axis is the straight line passing through the origin and perpendicular to the Y-axis. Let X be the X-axis coordinate of the right foot COP trajectory. The Y-axis coordinate of the right foot COP trajectory; This is the number of the right foot pressure sensor. This represents the number of pressure sensors on the right foot. Indicates the first The x-coordinate value of each sensor location, For the first The ordinate value of each sensor location, Indicates the right foot's first The VGRF signal value corresponding to each sensor.

[0018] Furthermore, step S3 specifically includes:

[0019] Based on the X and Y axis coordinates of the COP trajectory information of the left and right feet, the control groups were divided into 4 categories, with each control group including those with normal gait and those with abnormal gait.

[0020] Test each type of control group Value, will Value less than threshold The control group was used as the target control group, and the COP trajectory information corresponding to the target control group was used as the plantar pressure feature.

[0021] Furthermore, the Gram angle field map includes the Gram difference field map and the Gram sum field map.

[0022] Furthermore, during the network model training process, the Dropout algorithm and the Adam optimizer are used to optimize the SGAN network model.

[0023] A gait recognition model construction system based on plantar pressure features includes a data acquisition unit, a COP trajectory calculation unit, a plantar feature recognition unit, and a model generation unit.

[0024] The acquisition unit is used to acquire VGRF signals from the left and right feet of the subject; the subject includes individuals with normal gait and individuals with abnormal gait.

[0025] The COP trajectory calculation unit is used to calculate independent COP trajectory information for the left and right feet using the VGRF signal;

[0026] The plantar feature recognition unit is used to perform a significant difference analysis on the COP trajectory information between people with normal gait and a control group with abnormal gait, and to obtain plantar pressure characteristics.

[0027] The model generation unit is used to convert the plantar pressure features into a Gram angle field map, and input the Gram angle field map into the SGAN network model for network model training to obtain a trained SGAN network model, which is then used as a gait recognition model.

[0028] Furthermore, the COP trajectory information of the left foot is determined according to the following formula:

[0029] ;

[0030] ;

[0031] The origin is the center of symmetry of the left and right feet, the Y-axis is the axis of symmetry of the left and right feet, and the X-axis is the straight line passing through the origin and perpendicular to the Y-axis. The X-axis coordinate of the left foot COP trajectory is... The Y-axis coordinate of the left foot COP trajectory. This is the number of the pressure sensor for the left foot. This represents the number of pressure sensors on the left foot. Indicates the first The x-coordinate value of each sensor location, For the first The ordinate values ​​of each sensor location; Indicates the left foot The VGRF signal value corresponding to each sensor.

[0032] Furthermore, the COP trajectory information of the right foot is determined according to the following formula:

[0033] ;

[0034] ;

[0035] The origin is the center of symmetry of the left and right feet, the Y-axis is the axis of symmetry of the left and right feet, and the X-axis is the straight line passing through the origin and perpendicular to the Y-axis. Let X be the X-axis coordinate of the right foot COP trajectory. The Y-axis coordinate of the right foot COP trajectory; This is the number of the right foot pressure sensor. This represents the number of pressure sensors on the right foot. Indicates the first The x-coordinate value of each sensor location, For the first The ordinate value of each sensor location, Indicates the right foot's first The VGRF signal value corresponding to each sensor.

[0036] Furthermore, a significant difference analysis was performed on the COP trajectory information between the control group with normal gait and abnormal gait to obtain plantar pressure characteristics, specifically including:

[0037] Based on the X and Y axis coordinates of the COP trajectory information of the left and right feet, the control groups were divided into 4 categories, with each control group including those with normal gait and those with abnormal gait.

[0038] Test each type of control group Value, will Value less than threshold The control group was used as the target control group, and the COP trajectory information corresponding to the target control group was used as the plantar pressure feature.

[0039] The beneficial effects of this invention are as follows: This invention discloses a gait recognition model construction method and system based on plantar pressure features. It uses VGRF (Vertical Ground Reverse Force) data to calculate the independent COP (Central Plantar Pressure) trajectories of the left and right feet of the subject. Then, it uses statistical methods to analyze and select the optimal COP trajectory features. Considering the time dependence between gait trajectories, it selects the Gram angle field method to convert the COP trajectory features into Gram angle field maps as input to the SGAN network and trains the SGAN network to obtain an accurate and efficient gait recognition model, providing technical support for gait feature extraction. Attached Figure Description

[0040] The present invention will be further described below with reference to the accompanying drawings and embodiments:

[0041] Figure 1 This is a schematic diagram illustrating the gait recognition model construction process of the present invention;

[0042] Figure 2 This is a schematic diagram showing the distribution of the plantar pressure sensors of the present invention;

[0043] Figure 3 This is a schematic diagram of the SGAN network model structure of the present invention. Detailed Implementation

[0044] The present invention will be further described below with reference to the accompanying drawings, as shown in the figures:

[0045] The gait recognition model construction method based on plantar pressure features of the present invention includes the following steps:

[0046] S1. Collect VGRF signals from the left and right feet of the subject; the subject includes those with normal gait and those with abnormal gait.

[0047] S2. Calculate the independent COP trajectory information for the left and right feet using the VGRF signal;

[0048] S3. A significant difference analysis of COP trajectory information between individuals with normal gait and a control group with abnormal gait was performed to obtain plantar pressure characteristics;

[0049] S4. Convert the plantar pressure features into a Gram angle field map, and input the Gram angle field map into the SGAN network model for network model training to obtain a trained SGAN network model. Use the trained SGAN network model as a gait recognition model.

[0050] This invention first preprocesses the acquired VGRF signal data to remove outliers and reduce their impact on subsequent calculations. Then, it uses the filtered VGRF signal to calculate the independent COP trajectories of the left and right feet. Next, it uses statistical methods to perform significance analysis on the X-axis and Y-axis coordinates of the COP trajectories of individuals with normal and abnormal gait to obtain important plantar pressure features. To preserve all the information of the plantar pressure features, the Gram angle field method is used for image encoding. The encoded images will be used to train the SGAN network model.

[0051] In this embodiment, in step S1, eight pressure sensors are installed on the left and right feet of each subject. The positions of the sensors are as follows: Figure 2As shown. This sensor is used to record the vertical ground reaction force (VGRF) of the foot in contact with the ground. Data collected from each subject is generated into a file containing 19 columns. The first column is time, columns 2 through 17 show the changes in VGRF values ​​from each pressure sensor, and the last two columns show the total VGRF changes for the left and right feet, respectively. VGRF signals can be collected in three different modes: flat ground walking with a dual-task test (Ga), treadmill test (Si), and auditory stimulation test (Ju).

[0052] In this embodiment, in step S2, the subjects are in three different modes, with each subject corresponding to one data point, resulting in a total of 156 data samples. The VGRF acquisition frequency is 100Hz, and the Ga mode acquisition time is approximately 2 minutes. Considering that the data at the beginning of acceleration and the end of deceleration lack rhythm, the gait data in the first 20 seconds and the last 10 seconds are removed, resulting in a gait sequence of 9000 sample points.

[0053] The data acquisition time for Si mode is approximately 2 minutes, while the acquisition time for Ju mode varies, with the acquisition frequency being 100Hz for both modes. Since subjects can quickly enter the data acquisition state under external intervention in Ju and Si modes, no data truncation is performed for these two modes. Considering that leg swinging motions can affect the sensor, VGRF values ​​less than 20N in all data are set to 0.

[0054] During walking, the sole of the foot contacts the ground, generating a perpendicular and opposing force. The distribution and magnitude of plantar pressure vary from person to person, reflecting changes in lower limb movement. The center of plantar pressure (COP) is the point of application of the resultant force of various plantar pressures during walking, describing the current position of the foot's center of gravity. Simultaneously, the COP trajectory is the changing trajectory of the COP, unaffected by individual differences such as age, weight, and height. This invention calculates the COP trajectory of subjects, and the analysis method based on the COP trajectory overcomes the problems of randomness and individual variability inherent in traditional gait feature analysis methods.

[0055] The sensors are numbered according to their coordinate positions, as shown in Table 1 and Table 2:

[0056] Table 1

[0057]

[0058] Table 2

[0059]

[0060] The COP trajectory information of the left foot is determined according to the following formula:

[0061] ;

[0062] ;

[0063] Among them, combined Figure 2 With the center of symmetry of the left and right feet as the origin, and the axis of symmetry of the left and right feet as the Y-axis ( Figure 2 The vertical axis in the diagram), and the straight line passing through the origin and perpendicular to the Y-axis is taken as the X-axis ( Figure 2 (the horizontal axis in the middle). The X-axis coordinate of the left foot COP trajectory is... The Y-axis coordinate of the left foot COP trajectory. The pressure sensors for the left foot are numbered 1 through 8. This represents the number of pressure sensors on the left foot. The value is 8. Indicates the first The x-coordinate value of each sensor location, For the first The ordinate values ​​of each sensor location; Indicates the left foot The VGRF signal value corresponding to each sensor.

[0064] The COP trajectory information for the right foot is determined using the following formula:

[0065] ;

[0066] ;

[0067] in, Let X be the X-axis coordinate of the right foot COP trajectory. The Y-axis coordinate of the right foot COP trajectory; The pressure sensors for the right foot are numbered 1 through 8. This represents the number of pressure sensors on the right foot. The value is 8. Indicates the first The x-coordinate value of each sensor location, For the first The ordinate value of each sensor location, Indicates the right foot's first The VGRF signal value corresponding to each sensor.

[0068] In this embodiment, step S3 involves performing a significant difference analysis on the COP trajectory information between individuals with normal gait and a control group with abnormal gait to obtain plantar pressure characteristics, including:

[0069] Based on the X and Y axis coordinates of the COP trajectory information of the left and right feet, the control groups were divided into 4 categories, with each control group including those with normal gait and those with abnormal gait.

[0070] Test each type of control group Value, will Value less than threshold The control group was used as the target control group, and the COP trajectory information corresponding to the target control group was used as the plantar pressure feature. Among these, the threshold... The possible value is 0.001.

[0071] Specifically, independent samples t-tests were conducted on the X and Y axis coordinates of the left and right foot COP trajectories of individuals with normal gait and individuals with abnormal gait, for a total of four control groups. The test results are shown in Table 3 (T-test of plantar pressure characteristics of subjects (variance ± mean)).

[0072] Table 3

[0073]

[0074] Table 3 shows that the P-value of the Y-axis coordinate of the COP trajectory of the left foot in individuals with normal gait and those with abnormal gait is less than 0.001, indicating that it is highly significant.

[0075] The test results show that: 1) The method of calculating the COP trajectory of the left and right feet independently is effective. Statistical analysis of the COP trajectory coordinates of various left and right feet in several individuals with normal and abnormal gait postures revealed a highly significant difference in the Y-axis coordinate of the COP trajectory of the left foot between individuals with abnormal gait postures and those with normal gait postures. 2) Compared to traditional gait feature extraction methods, the calculation of the independent COP trajectory of the left and right feet is unrelated to the subject's weight, age, and height, thus avoiding the influence of individual differences on the calculation and analysis.

[0076] Based on the above analysis results, the Y-axis coordinate of the left foot COP trajectory is selected as a new and important plantar pressure feature data.

[0077] In this embodiment, step S4 uses the Gramian Angular Field (GAF) method to extract statistically validated important plantar pressure features as a time series. The Gramian Angular Field (GAF) method is an image encoding method that, through a given... The time series is then scaled to the interval [-1,1] or [0,1]. The scaled values ​​are then... or Encoded as angle cosine and time nodes Encoded as radius .

[0078] (3)

[0079] or

[0080] (4)

[0081] (5)

[0082] In the formula, It indicated a point timestamp, It is the number of all time points contained in the time series data.

[0083] Furthermore, the Gram difference field matrix is ​​obtained. and Gram and field matrix :

[0084] (6)

[0085] (7)

[0086] In the formula, For unit row vectors, for transpose, for Matrix data after scaling.

[0087] By defining the inner product and This transforms the two forms of Gram angle fields mentioned above into quasi-Gram matrices.

[0088] Through the above operation process, two types of image data of plantar pressure features, namely the Gram difference field and the Gram sum field, can be obtained after transcoding, which are used for training the SGAN network model. That is, the Gram angle field image includes the Gram difference field image and the Gram sum field image, wherein the Gram difference field image and the Gram sum field are respectively represented by the aforementioned matrix. and matrix To express.

[0089] In this embodiment, to automatically extract deep feature information from image data, a combination of unsupervised and supervised learning, SGAN (Semi-Supervised Generative Adversarial Network), is used. SGAN is a semi-supervised generative adversarial network that adds N+1 class multi-class classification functionality to GAN, and its structure is as follows. Figure 3 As shown.

[0090] The supervised learning process of the SGAN model is as follows: a small amount of labeled real data is input into the discriminator network, and the latent features of the labeled data are captured through operations such as convolution and normalization. Finally, the Softmax function is added to output an N-dimensional probability vector value.

[0091] The unsupervised learning process is as follows: Unlabeled real images, random noise, and images generated by the generator are input into the discriminator network. A sigmoid function is added to the softmax output probability vector to distinguish between real and fake generated data. The discriminator and generator networks are trained alternately, optimizing and updating their parameters. By minimizing the loss functions of both the generator and discriminator, the generation capability of the generator and the classification accuracy of the discriminator can be improved simultaneously.

[0092] The loss function of the generator is:

[0093] (8)

[0094] The loss function of the discriminator is:

[0095] (9)

[0096] in, This indicates that the discriminator will recognize the real samples. The probability of classifying it as true. This indicates that the discriminator will generate samples. The probability of classifying it as true. It is the random noise vector input to the generator.

[0097] To address the overfitting problem and improve the robustness of the SGAN network model, the Dropout algorithm is introduced. By randomly dropping neurons during the training phase, Dropout effectively reduces the SGAN network model's dependence on specific samples, thereby improving generalization ability and anti-interference capability. For optimizing model parameters, the Adam optimizer is chosen. Compared to the traditional gradient descent algorithm, the Adam optimizer has lower memory consumption and faster convergence speed, making it better suited to the training requirements of deep learning models.

[0098] The present invention also relates to a gait recognition model construction system based on plantar pressure features. The system corresponds to the above-mentioned gait recognition model construction method based on plantar pressure features and can be understood as a system for implementing the above method. The system includes a data acquisition unit, a COP trajectory calculation unit, a plantar feature recognition unit, and a model generation unit.

[0099] The acquisition unit is used to acquire VGRF signals from the left and right feet of the subject; the subject includes individuals with normal gait and individuals with abnormal gait.

[0100] The COP trajectory calculation unit is used to calculate independent COP trajectory information for the left and right feet using the VGRF signal;

[0101] The plantar feature recognition unit is used to perform a significant difference analysis on the COP trajectory information between people with normal gait and a control group with abnormal gait, and to obtain plantar pressure characteristics.

[0102] The model generation unit is used to convert the plantar pressure features into a Gram angle field map, and input the Gram angle field map into the SGAN network model for network model training to obtain a trained SGAN network model, which is then used as a gait recognition model.

[0103] This invention provides an objective and scientific method and system for constructing a gait recognition model. It calculates the pressure center trajectories of the subject's left and right feet, which can more accurately capture potential gait abnormalities and lay the foundation for improving the classification accuracy of the gait recognition model. By converting the VGRF signal into a Gram angular field map, it can more effectively characterize gait abnormalities, thereby improving the accuracy and robustness of the gait recognition model.

[0104] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for constructing a gait recognition model based on plantar pressure features, characterized in that: Includes the following steps: S1. Collect VGRF signals from the left and right feet of the subject; the subject includes those with normal gait and those with abnormal gait. S2. Calculate the independent COP trajectory information for the left and right feet using the VGRF signal; S3. A significant difference analysis of COP trajectory information between individuals with normal gait and a control group with abnormal gait was performed to obtain plantar pressure characteristics; S4. Convert the plantar pressure features into a Gram angle field map, and input the Gram angle field map into the SGAN network model for network model training to obtain a trained SGAN network model. Use the trained SGAN network model as a gait recognition model. The SGAN network model is a semi-supervised generative adversarial network.

2. The method for constructing a gait recognition model based on plantar pressure features according to claim 1, characterized in that: The COP trajectory information of the left foot is determined according to the following formula: ； ； The origin is the center of symmetry of the left and right feet, the Y-axis is the axis of symmetry of the left and right feet, and the X-axis is the straight line passing through the origin and perpendicular to the Y-axis. The X-axis coordinate of the left foot COP trajectory is... The Y-axis coordinate of the left foot COP trajectory. This is the number of the pressure sensor for the left foot. This represents the number of pressure sensors on the left foot. Indicates the first The x-coordinate value of each sensor location, For the first The ordinate values ​​of each sensor location; Indicates the left foot The VGRF signal value corresponding to each sensor.

3. The method for constructing a gait recognition model based on plantar pressure features according to claim 1, characterized in that: The COP trajectory information for the right foot is determined using the following formula: ； ； in, Let X be the X-axis coordinate of the right foot COP trajectory. The Y-axis coordinate of the right foot COP trajectory; This is the number of the right foot pressure sensor. This represents the number of pressure sensors on the right foot. Indicates the first The x-coordinate value of each sensor location, For the first The ordinate value of each sensor location, Indicates the right foot's first The VGRF signal value corresponding to each sensor.

4. The method for constructing a gait recognition model based on plantar pressure features according to claim 1, characterized in that: Step S3 specifically includes: Based on the X and Y axis coordinates of the COP trajectory information of the left and right feet, the control groups were divided into 4 categories, with each control group including those with normal gait and those with abnormal gait. Test each type of control group Value, will Value less than threshold The control group was used as the target control group, and the COP trajectory information corresponding to the target control group was used as the plantar pressure feature.

5. The method for constructing a gait recognition model based on plantar pressure features according to claim 1, characterized in that: The Gram angle field map includes the Gram difference field map and the Gram sum field map.

6. The method for constructing a gait recognition model based on plantar pressure features according to claim 1, characterized in that: During the network model training process, the Dropout algorithm and Adam optimizer are used to optimize the SGAN network model.

7. A gait recognition model construction system based on plantar pressure characteristics, characterized in that: It includes a data acquisition unit, a COP trajectory calculation unit, a plantar feature recognition unit, and a model generation unit; The acquisition unit is used to acquire VGRF signals from the left and right feet of the subject; the subject includes individuals with normal gait and individuals with abnormal gait. The COP trajectory calculation unit is used to calculate independent COP trajectory information for the left and right feet using the VGRF signal; The plantar feature recognition unit is used to perform a significant difference analysis on the COP trajectory information between people with normal gait and a control group with abnormal gait, and to obtain plantar pressure characteristics. The model generation unit is used to convert the plantar pressure features into a Gram angle field map, and input the Gram angle field map into the SGAN network model for network model training to obtain a trained SGAN network model. The trained SGAN network model is used as a gait recognition model. The SGAN network model is a semi-supervised generative adversarial network.

8. The gait recognition model construction system based on plantar pressure features according to claim 7, characterized in that: The COP trajectory information of the left foot is determined according to the following formula: ； ； The origin is the center of symmetry of the left and right feet, the Y-axis is the axis of symmetry of the left and right feet, and the X-axis is the straight line passing through the origin and perpendicular to the Y-axis. The X-axis coordinate of the left foot COP trajectory is... The Y-axis coordinate of the left foot COP trajectory. This is the number of the pressure sensor for the left foot. This represents the number of pressure sensors on the left foot. Indicates the first The x-coordinate value of each sensor location, For the first The ordinate values ​​of each sensor location; Indicates the left foot The VGRF signal value corresponding to each sensor.

9. The gait recognition model construction system based on plantar pressure features according to claim 7, characterized in that: The COP trajectory information for the right foot is determined using the following formula: ； ； in, Let X be the X-axis coordinate of the right foot COP trajectory. The Y-axis coordinate of the right foot COP trajectory; This is the number of the right foot pressure sensor. This represents the number of pressure sensors on the right foot. Indicates the first The x-coordinate value of each sensor location, For the first The ordinate value of each sensor location, Indicates the right foot's first The VGRF signal value corresponding to each sensor.

10. The gait recognition model construction system based on plantar pressure features according to claim 7, characterized in that: A significant difference analysis of COP trajectory information was performed between individuals with normal gait and a control group with abnormal gait to obtain plantar pressure characteristics, specifically including: Based on the X and Y axis coordinates of the COP trajectory information of the left and right feet, the control groups were divided into 4 categories, with each control group including those with normal gait and those with abnormal gait. Test each type of control group Value, will Value less than threshold The control group was used as the target control group, and the COP trajectory information corresponding to the target control group was used as the plantar pressure feature.

Images