CN110226932A - The plantar pressure feature extracting method of human body daily behavior movement - Google Patents

Abstract

The invention discloses a method for extracting plantar pressure features of human daily behaviors. The present invention collects the pressure signals of the first metatarsal, the second metatarsal and the heel area through the pressure insole, calculates the pressure ratio, the total pressure ratio, normalizes the pressure of each sensor and the total pressure, and extracts the first characteristic of the plantar pressure vector and the second eigensubvector. According to the various motion modes of the human body, the current value of the plantar pressure sensor is related to the past value, and the AR model of the plantar pressure signal is constructed to obtain the model coefficients. The AIC calculation of the plantar AR model is carried out through experiments on different daily behaviors, and the value and dimension of the AIC are integrated to propose a trade-off reliability, so that the order corresponding to the highest reliability of the trade-off is the most appropriate order. The AR model coefficients of the plantar pressure sensor are constructed as the third feature vector. The present invention determines the order of the plantar pressure AR model through the AIC criterion and the reliability of weighing, and has good effect.

Description

Plantar pressure feature extraction method for human daily behaviors

technical field

The invention belongs to the field of feature extraction, and relates to a plantar pressure feature extraction method of human daily behaviors.

Background technique

Plantar pressure refers to the action force and reaction force between the sole of the human foot and the ground when the sole of the human foot is in contact with the ground. Its distribution reveals the detailed information of the foot movement, which is of great significance to the analysis of human gait and is widely used In the fields of identity recognition, motion tracking, motion recognition, etc.

Plantar pressure profiles were originally studied pathologically due to the clinical relevance of pressure distribution in the foot to various foot pathologies. Scanning and analyzing the pressure distribution on the sole of the foot is crucial, which will aid in post-surgical biomechanical assessment, orthotic design, and research to improve patient balance control, identify gait deviations or postural disturbances. Chen et al. made a pressure insole that uses discrete contact force distribution signals for motion pattern recognition. The insole includes a foot wearable interface and four FSR pressure sensors. Achkar et al. later designed an instrumented shoe system containing pressure sensors and inertial sensors for the monitoring and diagnosis of daily activities and falls in the elderly. The processing of plantar pressure generally includes maximum value, minimum value, average value and standard deviation, etc. Xia Yi used his self-developed pressure measurement board to collect plantar pressure during walking, and used spatio-temporal HOG features to characterize the distribution of pressure for gait recognition. The experimental data belong to the image results of foot scans, the recognition rate is 93.5%, and the false alarm rate is 5.2%, which accurately depicts the division of different footprints through the distribution of plantar pressure. Chen recognizes gait movements through the total pressure of the discrete plantar pressure model, the correlation coefficient of several channels and the third-order autoregressive coefficient, and obtains considerable recognition results. Plantar pressure refers to a force signal generated when the sole of the human foot is in contact with the ground. Its size and distribution can reflect information such as the structure, function, and posture control of the human leg and foot. It is of great significance to the analysis of human gait. It is widely used in clinical diagnosis, disease degree measurement, postoperative curative effect evaluation and action recognition and other fields.

Contents of the invention

The present invention is aimed at the deficiencies in the prior art, proposes a kind of plantar pressure feature extraction method of human daily behavior action. First, according to the plantar pressure distribution map, the pressure signals of the first metatarsal, the second metatarsal and the heel area are collected through the pressure insole, the pressure ratio and the total pressure ratio are calculated, and the pressure of each sensor and the total pressure are normalized to extract the foot pressure. The first and second eigensubvectors of bottom pressure. According to the various motion modes of the human body, the current value of the plantar pressure sensor is related to the past value, and the AR model of the plantar pressure signal is constructed to obtain the model coefficients. The AIC calculation of the plantar AR model is carried out through experiments on different daily behaviors, and the value and dimension of AIC are integrated to propose a trade-off reliability, so that the order corresponding to the highest reliability of the trade-off is the most appropriate order. The AR model coefficients of the plantar pressure sensor are constructed as the third feature vector. The present invention determines the order of the plantar pressure AR model through the AIC criterion and the reliability of weighing, and has good effects.

In order to achieve the above object, the inventive method mainly comprises the following steps:

Step (1), respectively place plantar pressure sensors on the first metatarsal bone, second metatarsal bone, and heel area to collect plantar pressure signals of the daily behavior of the human body; the daily behavior of the human body includes: standing, sitting on a chair or sitting on a level ground, Squat, lie down; walk on flat ground, go up stairs, go down stairs, run; stand - sit on a chair, sit on a chair - stand, stand - sit on level ground, sit on level ground - stand, stand - squat, squat - stand, sit - lie down, lie down - sitting; walking - falling, going upstairs - falling, going downstairs - falling, running - falling.

Step (2) Let F _i be the pressures of the three pressure sensors on the soles of the feet respectively, i=1, 2, 3, and F _st be the total pressure when standing, which can represent whether the weight of the human body is entirely borne by the feet; The pressure of the sensor and the total pressure are normalized, and the first characteristic subvector F1 and the second characteristic subvector F2 of the plantar pressure are extracted. The extraction formulas of the characteristic vectors F1 and F2 are as follows:

Step (3) According to the various motion patterns of the human body, the current value of the plantar pressure sensor is related to the past value, constructing the AR model of the plantar pressure signal, and obtaining the model coefficients η ₁ , η ₂ ,...η _P , details as follows:

Let Xi be the _plantar pressure signal sequence, then the AR model of the plantar pressure signal sequence is as follows:

X _i -η ₁ X _i-1 -η ₂ X _i-2 -...-η _p X _ip = C _i (2)

Wherein, P is the order of the AR model, _Ci is the error coefficient, and η _p is the AR coefficient;

The AR model can describe the signal relationship of each order through the error equation. The error equation of the P-order AR model is as follows:

C _p+1 ＝X _p η ₁ +X _p-1 η ₂ +...+X ₁ η _p -X _p+1

C _p+2 ＝X _p+1 η ₁ +X _p η ₂ +...+X ₂ η _p -X _p+2

...

C _N ＝X _n-1 η ₁ +X _n-1 η ₂ +...+X _np η _p -X _n (3)

Write formula (3) in matrix form:

C=Xη-Y (4)

Wherein C=[C _P+1 , C _P+2 ,...C _n ] ^T , η=[η ₁ , η ₂ ,...η _P ] ^T , Y=[X _P+1 , X _P+2 ,...X _n ] ^T ; then the least squares solution of η is: η=(X ^T X) ^-1 X ^T Y;

Step (4) AIC calculation of the plantar AR model is performed on different daily behaviors through experiments; the details are as follows:

The unbiased estimation of the residual variance of the P-order AR model is as follows:

in,

AIC of the AR model:

AIC＝NLogσ _η ² +2p (6)

According to the formula (6), the AIC calculation of the plantar AR model is performed on different daily behaviors through experiments;

Step (5) Synthesize the value of AIC and the dimension p, and propose the trade-off credibility Cre:

In the formula, AIC _nor is the normalized AIC value, p is the order of the AR model, that is, the feature dimension, and k is the dimension weight index; the higher the credibility Cre of the trade-off is, the corresponding order P _C is the most Appropriate level;

Step (6) takes the AR model coefficients of the 3 plantar pressure sensors It is constructed as a third feature vector F3=[η ₁ ,η ₂ ,η ₃ ].

Compared with the feature extraction algorithms of many existing plantar pressure signals, the present invention has the following characteristics:

Based on the feature extraction method of plantar pressure, the AR model has high frequency resolution and can still produce reasonable spectral estimates for short-segment data records. The choice of model order has a great influence on the characteristics of spectral estimation. If the order is too low, the resolution of spectral estimation will be low, and ideal results cannot be obtained. If the order is too high, it will not only increase the amount of calculation, but also increase the spectral False details are generated in the estimation, and the present invention determines the order of the AR model through the AIC criterion and the weighing credibility Cre, which has a good effect.

Description of drawings

Fig. 1 is the implementation flowchart of the present invention;

Figure 2 is a partial gait and fall plantar pressure signal diagram

Fig. 3 is partial static total pressure ratio;

Figure 4 is a grayscale map of the trade-off credibility.

Detailed ways

The embodiments of the present invention are described in detail below in conjunction with the accompanying drawings: this embodiment is implemented on the premise of the technical solution of the present invention, and detailed implementation methods and specific operating procedures are provided, but the protection scope of the present invention is not limited to the following the described embodiment.

As shown in Figure 1, this embodiment includes the following steps:

Step 1: Place plantar pressure sensors on the first metatarsal, second metatarsal, and heel respectively to collect plantar pressure signals.

Step 2, let F _i (i=1, 2, 3) be the pressure of the three pressure sensors on the sole of the foot, and F _st be the total pressure when standing, which can indicate whether the weight of the human body is entirely borne by the feet. The pressure of each sensor and the total pressure are normalized, and the first characteristic subvector F1 and the second characteristic subvector F2 of the plantar pressure are extracted.

Step 3: According to the correlation between the current value of the plantar pressure sensor and the past value in various human body movement modes, the AR model of the plantar pressure signal is constructed, and the model coefficients η ₁ , η ₂ ,...η _P are obtained.

Step 4: Calculate the AIC (Akaike Information Criteria) of the plantar AR model for different daily behaviors through experiments.

Step 5: Combining the value of AIC and the dimension p, the reliability of the trade-off is proposed. The higher the reliability of the trade-off, the corresponding order is the most appropriate order.

Step 6, put the AR model coefficients of the 3 plantar pressure sensors It is constructed as a third feature vector F3=[η ₁ ,η ₂ ,η ₃ ].

As shown in Figure 2, the sub-vector F2 formed by the total plantar pressure ratio can intuitively identify the contact between the footsteps and the ground, and then distinguish static and gait movements. Figure 3 shows 10 groups of standing, sitting, and lying down The schematic diagram of the total pressure ratio of , according to the size of the respective values, it is easy to use the threshold method to identify whether the static action has support behavior.

When AIC(min) takes the minimum value, the most suitable order is reached. Generally, the appropriate order for AIC calculation is not too high, but in order to measure the relationship between the order and the AIC value and obtain the most suitable order value, the present invention limits the AR order between 1st and 12th order. For different ADLs including falling movements, the AIC of the plantar AR model was calculated separately, and the respective AIC values were recorded in Table 1. The values in bold in the table are the minimum AIC values in this movement.

Table 1 AIC values of ADLs and falls under different AR orders

Although the smaller the value of AIC, the determined order achieves better feature extraction effect, but the higher the selected order, it will cause the dimension of the input signal to increase, which brings a certain degree of complexity to the subsequent processing. , it is obvious that the larger the Cre is, the more appropriate the order is. Figure 4 shows the trade-off reliability of each data calculated by the above formula, where the ordinates are the various ADLs numbered in abbreviated form, and the abscissa indicates the order of AR. For each row of data, the lighter the color, the higher the reliability of the trade-off, and the better the feature extraction effect achieved by the order represented by the abscissa. From each row, the light-colored areas are mainly concentrated in the order 1-5, so this embodiment of the present invention selects the order 3 as the parameter of the AR model.

Claims (2)

1. the plantar pressure feature extraction method of human daily behavior action, it is characterized in that, this method comprises the steps: Step (1), respectively place plantar pressure sensors on the first metatarsal, second metatarsal, and heel area to collect plantar pressure signals of human daily behavior; Step (2) Let F _i be the pressures of the three pressure sensors on the soles of the feet respectively, i=1, 2, 3, and F _st be the total pressure when standing, which can represent whether the weight of the human body is entirely borne by the feet; The pressure of the sensor and the total pressure are normalized, and the first characteristic subvector F1 and the second characteristic subvector F2 of the plantar pressure are extracted. The extraction formulas of the characteristic vectors F1 and F2 are as follows: Step (3) According to the various motion patterns of the human body, the current value of the plantar pressure sensor is related to the past value, constructing the AR model of the plantar pressure signal, and obtaining the model coefficients η ₁ , η ₂ ,...η _P , details as follows: Let Xi be the _plantar pressure signal sequence, then the AR model of the plantar pressure signal sequence is as follows: X _i -η ₁ X _i-1 -η ₂ X _i-2 -...-η _p X _ip = C _i (2) Wherein, P is the order of the AR model, _Ci is the error coefficient, and η _p is the AR coefficient; The AR model can describe the signal relationship of each order through the error equation. The error equation of the P-order AR model is as follows: Write formula (3) in matrix form: C=Xη-Y (4) Wherein C=[C _P+1 , C _P+2 ,...C _n ] ^T , η=[η ₁ , η ₂ ,...η _P ] ^T , Y=[X _P+1 , X _P+2 ,...X _n ] ^T ; then the least squares solution of η is: η=(X ^T X) ^-1 X ^T Y; Step (4) Perform the AIC calculation of the plantar AR model through experiments on different daily behaviors; the details are as follows: The unbiased estimation of the residual variance of the P-order AR model is as follows: in, AIC of the AR model: AIC＝NLogσ _η ² +2p (6) According to the formula (6), the AIC calculation of the plantar AR model is carried out through experiments on different daily behaviors; Step (5) Synthesize the value of AIC and the dimension p, and propose the trade-off credibility Cre: In the formula, AIC _nor is the normalized AIC value, p is the order of the AR model, that is, the feature dimension, and k is the dimension weight index; the higher the credibility Cre of the trade-off is, the corresponding order P _C is the most Appropriate level; Step (6) takes the AR model coefficients of the 3 plantar pressure sensors It is constructed as a third feature vector F3=[η ₁ ,η ₂ ,η ₃ ]. 2. the plantar pressure feature extraction method of human body's daily behavior according to claim 1, is characterized in that: described human body's daily behavior comprises: standing, sitting on a chair or sitting on the level ground, squatting, lying down; flat ground Walk, go up stairs, go down stairs, run; stand - sit on a chair, sit on a chair - stand, stand - sit on a level ground, sit on a level ground - stand, stand - squat, squat - stand, sit - lie down, lie down - sit; walk - Fall, go upstairs-fall, go downstairs-fall, run-fall.