Pedestrian abnormal gait recognition method based on human body model analysis

Abstract

The invention provides a pedestrian abnormal gait recognition method based on human body model analysis, and the method comprises the following steps: 1, building a motion capture system, wearing an IMU sensor, designing a plurality of motion scenes, and obtaining C3D data and IMU data in different motion states; 2, building an Anybody human body model, importing the C3D data into the model, carrying out kinematics and dynamics analysis, and obtaining related simulation data; 3, different motion states are analyzed through simulation data and IMU data, gait phases are divided according to analysis results, and each set of motion data is classified; 4, building a random forest model, combining the classified data training model, and performing single-foot gait phase recognition by using the model; and 5, analyzing the types of the gait combinations of the two feet, and building a K adjacent machine learning model to recognize abnormal gait combinations. Therefore, the abnormal gait recognition of the pedestrian is completed.

Description

Technical Field

[0001] The present invention relates to the problem of gait safety in the field of power-assisted exoskeletons and human-machine collaboration, and in particular to a method for identifying abnormal gait of pedestrians based on human body model analysis. Background Art

[0002] Abnormal gait recognition of pedestrians is an important part of research in the field of human-computer interaction and assisted exoskeleton. Researching a gait phase recognition method with high universality is of great significance to ensuring the working safety of human-computer interaction and exoskeleton robots.

[0003] In recent years, exoskeleton robots have become an emerging technology in the medical, industrial and military fields. Lower limb exoskeletons have important research value in the medical field. In order to stably control the movement process of lower limb exoskeletons and enhance the consistency and coordination of human-computer interaction, it is necessary to identify the gait information of human movement and judge the state changes of human lower limb gait. The accuracy of gait phase classification has a great impact on the performance and safety of lower limb exoskeletons. The source of gait data and the gait phase detection method are the key factors affecting the accuracy of gait classification. There are usually two types of data related to human movement intention signals: bioelectric signals and non-bioelectric signals. Bioelectric signals include electromyography and electroencephalography. Non-bioelectric signals mainly include joint angles, speeds, accelerations, interaction forces / torques between the wearer and the exoskeleton, and ground reaction forces during human movement.

[0004] The main calculation methods for gait phase detection are threshold method and machine learning method. Some researchers have proposed a ground contact detection algorithm based on inertial measurement unit (IMU), which uses the range and change rate of the main parameters measured by the IMU sensor to form a threshold detection criterion, and can identify the four stages of the gait phase independently of the walking speed and walking terrain. However, the threshold needs to be set manually, and the final gait phase classification result may be affected by the designer's subjective factors. This recognition algorithm based on threshold rules is too rough to handle complex situations. It is intuitive but has poor generalization performance. Some researchers have also compared the time error and goodness index of hidden Markov machine learning method and threshold method in gait phase recognition when a single IMU sensor is placed on the instep and a force-sensitive resistor is installed on the custom insole. The results show that the hidden Markov machine learning method is better than the threshold method.

[0005] Therefore, this application proposes a method for identifying abnormal gait of pedestrians based on human body model analysis. In addition, the method proposed in this application has higher universality and can adapt to the size differences of height and leg length of different people, which has very important practical significance. Summary of the invention

[0006] In view of the deficiencies in the existing technology, the technical problem to be solved by the present invention is to provide a method for identifying abnormal gait of pedestrians based on human body model analysis and applicable to gaited people.

[0007] The technical solution adopted by the present invention to solve the technical problem is: a method for identifying abnormal gait of pedestrians based on human body model analysis, characterized in that the motion information of gait phase can be analyzed in combination with human body model, extracted by machine learning model, and accurately identified, so as to complete gait identification suitable for different people, and the specific steps of the method are: step 1, build a motion capture system and wear IMU sensor, design a variety of motion scenes, and obtain C3D data and IMU data under different motion states; step 2, build Anybody human body model, import C3D data into the model, perform kinematic and dynamic analysis, and obtain relevant simulation data; step 3, analyze different motion states using simulation data and IMU data, divide gait phase according to the analysis results, and classify each group of motion data; step 4, build a random forest model, combine the data training model completed by classification, and use the model to identify the unipedal gait phase; step 5, analyze the types of bipedal gait combinations, and build a K-nearest neighbor machine learning model to identify abnormal gait combinations. In this way, abnormal gait identification of pedestrians is completed.

[0008] Furthermore, step one includes: building a motion capture acquisition system, arranging 10 high-definition motion capture cameras in a circular shape, placing 41 marking points at specific positions on the experimenter's body, and placing two 9-axis IMU sensors on the toes of hard shoes to collect three-axis acceleration and angular velocity data of the foot. The acquisition frequency of the motion capture system is 100 frames / s, and the acquisition frequency of IMU data is 100 frames / s. 18 volunteers are recruited to use the data acquisition system to complete normal walking, slipping, tripping, left falling, right falling and other movement processes, and C3D data and IMU data are obtained and saved.

[0009] Furthermore, step two includes: building an Anybody human model, which consists of 39 limb joints and 134 muscles, using the model's built-in scaling formula and the height and weight information of different individuals to obtain the geometric data and mass of each part of the limbs, importing C3D data into the human body model, and using the data points of the C3D file to pull the corresponding points in the model to drive the human body model to complete kinematic and dynamic analysis, and obtain relevant simulation data, which includes the position, velocity, acceleration, angular velocity and other information of each point of the model during movement.

[0010] Furthermore, step three includes: studying the normal walking gait of the human body, analyzing the periodic motion laws of the two feet, dividing it into a support phase and a swing phase according to the change of the Euler angle in a complete gait cycle, and dividing the normal single-foot walking phase into five periods: Supporting Phase (SP), Swing Preparation Stage (SPS), Pre Swing Stage (PSS), Late Swing Stage (LSS), and Swing Recovery Stage (SRS) according to the angles and angular velocity of the foot pitch angle and roll angle, and combining the linear velocity and acceleration of the foot, and classifying the simulation data and IMU data according to the divided phases.

[0011] Furthermore, step four includes: building a random forest model (RF), using the bootstrap sampling method to generate 50 sub-samples to improve the stability and generalization ability of the model, using the divided data set to train the model, and taking the pitch angle, roll angle, pitch angular velocity, velocity and acceleration modulus of the foot during normal walking as the input of the model, and the corresponding gait phase as the output of the model.

[0012] Furthermore, step five includes: summarizing all phase combinations of bipedal gaits, dividing them into two types: normal combinations and abnormal combinations, building a K-nearest neighbor machine learning model to combine and classify bipedal gait phases, and completing the abnormal gait recognition of pedestrians.

[0013] The beneficial effect of the present invention is that a method for identifying abnormal gait of pedestrians based on human body model analysis is proposed, the motion information of human walking is analyzed by the method of human body model mapping, the walking gait phase is classified by using data features, and the gait is monitored in real time by combining random forest and K-nearest neighbor machine learning models, thereby realizing the identification of abnormal gait of human walking. The method has high universality and convenience. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0015] Figure 1 The overall flow chart of the present invention is

[0016] Figure 2The working principle diagram of the human body model of the present invention is

[0017] Figure 3 Schematic diagram of gait phase analysis principle of the present invention

[0018] Figure 4 Schematic diagram of the random forest model constructed for the present invention

[0019] Figure 5 The result diagram of the present invention for the left foot phase recognition

[0020] Figure 6 The result diagram of the present invention for the right foot phase recognition

[0021] Figure 7 The sequence diagram of the bipedal phase combination of the present invention is DETAILED DESCRIPTION

[0022] Reference Figure 1 , a pedestrian abnormal gait recognition method based on human body model analysis, comprising the following steps: Step 1, build a motion capture system and wear an IMU sensor, design a variety of motion scenes, and obtain C3D data and IMU data under different motion states; Step 2, build an Anybody human body model, import C3D data into the model, perform kinematic and dynamic analysis, and obtain relevant simulation data; Step 3, use simulation data and IMU data to analyze different motion states, divide gait phases according to the analysis results, and classify each group of motion data; Step 4, build a random forest model, combine the classified data training model, and use the model to identify the single-leg gait phase; Step 5, analyze the types of bipedal gait combinations, and build a K-nearest neighbor machine learning model to identify abnormal gait combinations. In this way, the abnormal gait recognition of pedestrians is completed.

[0023] Reference Figure 2, Steps 1 and 2 include: building a motion capture acquisition system, arranging 10 high-definition motion capture cameras in a circular shape, placing 41 markers at specific locations on the experimenter, and placing two 9-axis IMU sensors on the toes of hard shoes to collect three-axis acceleration and angular velocity data of the foot. The acquisition frequency of the motion capture system is 100 frames / s, and the acquisition frequency of IMU data is 100 frames / s. 18 volunteers are recruited to use the data acquisition system to complete normal walking, slipping, tripping, left fall, right fall and other motion processes, and C3D data and IMU data are obtained and saved. Build an Anybody human model, which consists of 39 limb joints and 134 muscles. Use the built-in scaling formula of the model and the height and weight information of different individuals to obtain the geometric data and mass of each part of the limb. Import C3D data into the human model, use the data points of the C3D file to pull the corresponding points in the model, so as to drive the human model to complete kinematic and dynamic analysis, and obtain relevant simulation data. The simulation data includes the position, velocity, acceleration, angular velocity and other information of each point in the model during the movement process.

[0024] Reference Figure 3 Step three includes: studying the normal walking gait of the human body, analyzing the periodic motion law of the two feet, dividing it into the support phase and the swing phase according to the change of the Euler angle in a complete gait cycle, and dividing the normal walking single-foot phase into the support phase (Supporting Phase, SP), swing preparation phase (Swing Preparation Stage, SPS), pre-swing stage (Pre Swing Stage, PSS), late swing stage (Late Swing Stage, LSS), and swing recovery stage (Swinging Recovery Stage, SRS) according to the angles and angular velocity of the foot pitch angle and roll angle, and combining the linear velocity and acceleration of the foot. Five periods, the simulation data and IMU data are classified according to the divided phases.

[0025] Reference Figure 4 , Figure 5 , Figure 6 ,Step 4 includes: building a random forest model (RF), using bootstrap sampling method to generate 50 subsamples to improve the stability and generalization ability of the model, using the divided data set to train the model, and taking the pitch angle, roll angle, pitch angular velocity, velocity and acceleration modulus of the foot during normal walking as the input of the model, and the corresponding gait phase as the output of the model. Figure 5 and Figure 6 The results of the random forest model for left and right footstep recognition are shown in the figure, corresponding to the phase recognition confusion matrix of the left and right foot respectively.

[0026] Reference Figure 7 ,Step 5 includes: summarizing all the phase combinations of bipedal gaits, ,classifying them into two types: normal combination and abnormal ,combination, building a K-nearest neighbor machine learning model to ,combine and classify the bipedal gait phases, and completing the ,abnormal gait recognition of pedestrians. Figure 7 It is the main basis for the K-nearest neighbor model to identify abnormal sequence combinations. The green sequence in the figure is a normal gait combination, and the blue sequence is an abnormal gait combination.

[0027] The above is only a specific embodiment of the present invention, and cannot be used to limit the scope of the invention. Therefore, the replacement of equivalent components, or equivalent changes and modifications made according to the scope of protection of the patent of the present invention, should still fall within the scope of this patent. In addition, the technical features of the present invention can be freely combined with each other, with each other, and with each other.

Claims

1. A method for identifying abnormal gait of pedestrians based on human body model analysis, characterized in that: The method includes the following steps: Step 1: Build a motion capture system and wear an IMU sensor, design a variety of motion scenes, and obtain C3D data and IMU data under different motion states; Step 2: Build an Anybody human model, import C3D data into the model, perform kinematic and dynamic analysis, and obtain relevant simulation data; Step 3: Use simulation data and IMU data to analyze different motion states, divide gait phases according to the analysis results, and classify each group of motion data; Step 4: Build a random forest model, combine the classified data training model, and use the model to identify the unipedal gait phase; Step 5: Analyze the types of bipedal gait combinations, and build a K-nearest neighbor machine learning model to identify abnormal gait combinations. In this way, the abnormal gait recognition of pedestrians is completed.

2. The method for identifying abnormal gait of pedestrians based on human body model analysis according to claim 1, characterized in that: Step one includes: building a motion capture acquisition system, arranging 10 high-definition motion capture cameras in a circle, placing 41 marking points at specific positions on the experimenter's body, and placing two 9-axis IMU sensors on the toes of hard shoes to collect three-axis acceleration and angular velocity data of the foot. The acquisition frequency of the motion capture system is 100 frames / s, and the acquisition frequency of IMU data is 100 frames / s. 18 volunteers are recruited to use the data acquisition system to complete normal walking, slipping, tripping, left falling, right falling and other movement processes, and C3D data and IMU data are obtained and saved.

3. The method for identifying abnormal gait of pedestrians based on human body model analysis according to claim 1, characterized in that: Step 2 includes: building an Anybody human model, which consists of 39 limb joints and 134 muscles, using the model's built-in scaling formula and the height and weight information of different individuals to obtain the geometric data and mass of each limb part, importing C3D data into the human body model, and using the data points of the C3D file to pull the corresponding points in the model to drive the human body model to complete kinematic and dynamic analysis and obtain relevant simulation data. The simulation data includes the position, velocity, acceleration, angular velocity and other information of each point of the model during movement.

4. The method for identifying abnormal gait of pedestrians based on human body model analysis according to claim 1, characterized in that: Step three includes: studying the normal walking gait of the human body, analyzing the periodic motion laws of the two feet, dividing it into the support phase and the swing phase according to the changes in the Euler angle in a complete gait cycle, and dividing the normal walking single-foot phase into five periods: the support phase (Supporting Phase, SP), the swing preparation phase (Swing Preparation Stage, SPS), the pre-swinging stage (PreSwing Stage, PSS), the late swinging stage (Late Swing Stage, LSS), and the swinging recovery stage (SRS) according to the angles and angular velocities of the foot pitch angle and roll angle, and combined with the linear velocity and acceleration of the foot, and classifying the simulation data and IMU data according to the divided phases.

5. The method for identifying abnormal gait of pedestrians based on human body model analysis according to claim 1, characterized in that: Step 4 includes: building a random forest model (RF), using the bootstrap sampling method to generate 50 sub-samples to improve the stability and generalization ability of the model, using the divided data set to train the model, and taking the pitch angle, roll angle, pitch angular velocity, velocity and acceleration modulus of the normal walking foot as the input of the model, and the corresponding gait phase as the output of the model.

6. The method for identifying abnormal gait of pedestrians based on human body model analysis according to claim 1, characterized in that: Step five includes: summarizing all bipedal gait phase combinations, dividing them into two types: normal combinations and abnormal combinations, building a K-nearest neighbor machine learning model to combine and classify bipedal gait phases, and completing the abnormal gait recognition of pedestrians.

Images