A human body carrying motion recognition method based on a multi-layer threshold discrimination method

Abstract

The application provides a human carrying motion recognition method based on a multi-layer threshold discrimination method, comprising the following steps: measuring human motion data through an inertial sensor when a human carries out carrying work; using a human carrying motion preprocessing algorithm based on a sliding window, analyzing and calculating each data frame and a sliding data window with a given time length according to the measured human motion data to obtain a feature vector of human carrying motion at a moment; recognizing the human carrying motion to obtain a first layer recognition result of the human carrying motion at the moment; analyzing and calculating a second layer recognition result of the human carrying motion at the moment according to the calculated human motion feature vector and the first layer recognition result; and analyzing and verifying the second layer recognition result to obtain a final recognition result of the human carrying motion. The technical scheme of the application can reduce the calculation complexity of the recognition model and improve the stability and accuracy of the human motion recognition algorithm.

Description

Technical Field

[0001] This invention belongs to the field of action recognition technology, and in particular to a human movement recognition method based on a multi-layer threshold discrimination method. Background Technology

[0002] my country's manufacturing industry is highly developed, with many sectors being labor-intensive, including professions requiring heavy lifting, prolonged bending, or repetitive lumbar movements. In the industrial sector, intelligent material handling assistive exoskeleton robots have become a hot research topic in recent years, experiencing rapid development both domestically and internationally. Research areas include biomimetic structural design based on human kinematics, multi-sensor information fusion, human motion intention recognition, and human-machine collaborative control.

[0003] To enable assisted transport exoskeleton robots to achieve their purpose, they need to accurately identify human transport movements during the transport process. Therefore, designing a simple, accurate, and stable identification method for human transport movements is one of the challenging tasks that urgently needs to be solved in the research field of assisted transport exoskeleton robots.

[0004] Based on the processing methods of the raw sensor signal data, commonly used human motion recognition algorithms can be divided into threshold detection-based recognition algorithms and machine learning-based recognition algorithms. Machine learning-based recognition algorithms use various machine learning models, such as linear discriminant analysis, support vector machines, neural networks, and hidden Markov models, to identify and classify sensor information to obtain the required human motion information. This method requires a large number of samples to train the model to obtain a relatively accurate recognition model, and it is computationally complex, has poor real-time performance, and its recognition accuracy decreases on different users, requiring retraining with new samples, resulting in poor portability.

[0005] Threshold-based discrimination methods obtain human motion information by comparing the feature vector of human motion with a threshold vector. The model is simple and computationally inexpensive, ensuring good real-time performance. This method is well-suited for repetitive tasks such as material handling. However, currently used simple threshold discrimination methods still suffer from poor stability.

[0006] Chinese Patent 202210074745.7 discloses a method, system, and device for analyzing abdominal extension movements. This patent determines the user's abdominal extension movement characteristics based on the rate of change of the abdominal extension angle and the maximum lower leg tilt angle. Specifically, it obtains the user's abdominal extension movement by comparing the rate of change of the abdominal extension angle and the maximum lower leg tilt angle with a threshold. However, a single threshold discrimination method can lead to unstable and inaccurate human movement recognition results. Summary of the Invention

[0007] To address the problems of computational complexity, low accuracy, and poor stability in existing methods for recognizing human movement, this invention provides a human movement recognition method based on a multi-level threshold discrimination method. By utilizing a sliding window-based human movement preprocessing algorithm and a multi-level threshold discrimination method for human movement recognition, the stability and accuracy of the human movement recognition algorithm can be greatly improved, while reducing the computational complexity of the recognition model.

[0008] The technical solution to achieve the purpose of this invention is as follows:

[0009] A method for recognizing human carrying actions based on a multi-layer threshold discrimination approach includes:

[0010] Step 1: When the human body is being moved, the Euler angles and angular velocities in the sagittal plane of different parts of the human body are collected by inertial sensors worn on the human body.

[0011] Step 2: Using a sliding window-based human movement preprocessing algorithm, based on the measured human movement data, analyze and calculate the feature vector of the human movement at a certain moment for each data frame and a sliding data window with a given number of frames.

[0012] Step 3: Using the first-layer threshold discrimination method of the human body carrying action recognition algorithm based on the multi-layer threshold discrimination method, the human body carrying action is identified by the first-layer threshold discrimination method according to the calculated human motion feature vector, and the first-layer recognition result of the human body carrying action at a certain moment is obtained.

[0013] Step 4: Using the second-layer threshold discrimination method of the human movement recognition algorithm based on the multi-layer threshold discrimination method, based on the calculated human motion feature vector and the first-layer recognition result obtained by the first-layer threshold discrimination method, the second-layer threshold discrimination method is used to obtain the second-layer recognition result of the human movement at a certain moment.

[0014] Step 5: Using the additional rules of the human body carrying action recognition algorithm based on the multi-layer threshold discrimination method, analyze and verify the second-layer recognition result of the human body carrying action according to the obtained second-layer recognition result, and obtain the final recognition result of the human body carrying action.

[0015] The significant advantages of this invention compared to existing technologies are:

[0016] This invention provides a human movement recognition algorithm based on a multi-layer threshold discrimination method. Based on human motion data measured by inertial sensors, it quickly and accurately identifies human movement data through a first-layer and second-layer threshold discrimination method, improving the real-time performance of the algorithm. Additional rules are then used to verify the results, ensuring the accuracy and stability of the recognition. This algorithm compares the feature values ​​contained in the feature vector with the threshold to obtain the result. The calculation is simple, and users can personalize the threshold based on actual measurement data from different individuals, greatly improving the algorithm's portability. Attached Figure Description

[0017] Figure 1 This is a flowchart of the steps of a human body carrying action recognition algorithm provided by the present invention.

[0018] Figure 2 This is a schematic diagram of the wearable part of an inertial sensor on the human body.

[0019] Figure 3 This is a flowchart illustrating the specific implementation of a human movement recognition algorithm provided by the present invention.

[0020] Figure 4 This is a schematic diagram illustrating the changing patterns of effective handling movements during human body transport, provided by the present invention. Detailed Implementation

[0021] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0022] Combination Figure 1 The present invention provides a method for recognizing human carrying actions based on a multi-layer threshold discrimination method, comprising the following steps:

[0023] Step 1: When the human body is being moved, the triaxial Euler angles and angular velocities in the sagittal plane of different parts of the human body are collected by inertial sensors worn on the human body.

[0024] Three inertial sensors are worn on the human body. One is placed at the center line of the waist, and the other two are symmetrically distributed on both sides of the hip joint. They measure the motion data of the waist and hip during human movement, and obtain the angle and angular velocity data of the waist and hip in the sagittal plane when the human body is carrying.

[0025] Step 2: Using a sliding window-based human movement preprocessing algorithm, based on the measured human motion data, analyze and calculate the feature vector of the human movement at a given moment for each data frame and a sliding data window of a given time length; the specific steps are as follows:

[0026] Step 2.1: Collect the data of the lumbar angle and angular velocity of the left and right hip joints in the sagittal plane at time i during the human body handling process, and calculate the average value Mean of the left and right hip joint angles at time i. i The absolute value of the difference between the left and right hip joint angles. i The angles and angular velocities of the three body parts at this moment form the feature vector g of human motion for data frame analysis. i =(α i ,ω i Mean i Difference i ), where α i Let ω be the angle of the lumbar region in the sagittal plane of the human body at time i. i The angular velocity of the human body in the sagittal plane at time i.

[0027]

[0028] Mean i Difference represents the average angle of the left and right hip joints at time i. i β represents the absolute value of the difference in hip joint angles between the left and right sides at time i. 1i β 2i and represent the left and right hip joint angles of the data frame at time i, respectively.

[0029] Step 2.2: During the human body handling process, collect the data of the left and right hip joint angles in the sagittal plane of the human body at time i and the preceding n-1 data frames to form a sliding window of human motion data at time i. Calculate the absolute value of the difference between the left and right hip joint angles corresponding to each frame within the human motion data window. i The average value is the human motion characteristic value analyzed by the sliding window at time i.

[0030]

[0031] Mean_Difference i Difference represents the absolute value of the difference in the left and right hip joint angles corresponding to each frame within the human motion data window. i The average value is n, where n is the number of data frames contained in the selected data window.

[0032] Step 2.3: Combine the human motion feature vector from the data frame analysis at time i and the human motion feature value from the sliding window analysis at time i to obtain the human carrying action feature vector f at time i. i =(g i Mean_Difference i )=(αi ,ω i Mean i Difference i Mean_Difference i ).

[0033] Step 3: Using the first-layer threshold discrimination method of the human carrying action recognition algorithm based on the multi-layer threshold discrimination method, the human carrying action is identified according to the calculated human motion feature vector, thus obtaining the first-layer recognition result of the human carrying action at a certain moment. The specific steps are as follows:

[0034] Step 3.1: Take the feature vector f i The angle α of the human waist at time i. i Set three waist angle thresholds A1, A2, A3 (A3 > A2 > A1), and set α i By comparing with A1, A2, and A3 respectively, we can obtain the possible carrying actions that the human body can perform.

[0035] When A1≤α i If the value is ≤A2, then this action is first classified as a possible bending action.

[0036] When A2 < α i If the value is ≤A3, then this action is first classified as a possible walking action.

[0037] When α i When the value is other than 0, the first layer of recognition results in other actions.

[0038] Step 3.2: Given that this action may involve bending over, take the feature vector f. i The angular velocity ω of the human waist at time i. i , will ω i The result of the first layer of recognition is obtained by comparing it with 0.

[0039] If the waist angular velocity ω i If the value is ≤0, then the first layer of recognition results is a possible placement action.

[0040] If the waist angular velocity ω i If the result is greater than 0, then the first layer of recognition results is a possible action to improve performance.

[0041] Step 3.3: Under the condition that this action is a possible walking action, set the first threshold C1 corresponding to the absolute value of the difference in the angles of the left and right hip joints, and set the Difference... i The first layer recognition result is obtained by comparing it with C1.

[0042] If Difference i If C1 ≤ C1, the first layer recognition result is a possible walking action.

[0043] If Difference i If the value is any other, the first layer of recognition will result in another action.

[0044] Step 4: Using the second-layer threshold discrimination method of the human body carrying action recognition algorithm based on the multi-layer threshold discrimination method, the second-layer recognition result of the human body carrying action at a certain moment is analyzed and calculated based on the calculated human motion feature vector and the first-layer recognition result.

[0045] The second-level result of the human transport action recognition algorithm based on the multi-level threshold discrimination method is defined as five typical actions that occur during human transport: lifting, placing, walking, standing, and others. Among them, the first four actions are valid actions, and the other actions are invalid actions.

[0046] The specific steps of the second-layer human movement recognition algorithm are as follows: Take the feature vector f i Mean of the average left and right hip joint angles at time i. i The absolute value of the difference between the left and right hip joint angles. i .

[0047] Step 4.1: If the first layer recognition result indicates a possible placement action, set the first threshold B1 corresponding to the average left and right hip joint angles and the second threshold C2 corresponding to the absolute value of the difference between the left and right hip joint angles. Then, set the Mean... i Difference i The results are compared with B1 and C2 respectively to obtain the second layer recognition results.

[0048] When Mean i ≤B1 and Difference i When C2 is less than or equal to 2, the second layer recognition result is a placement action.

[0049] When Mean i and Difference i When the value is other than 0, the second layer recognition result is other actions.

[0050] Step 4.2: If the first-layer recognition result indicates a possible improvement action:

[0051] When Mean i ≤B1 and Difference i When C2 is less than or equal to 2, the second layer recognition result is an lifting action.

[0052] When Mean i and Difference i When the value is other than 0, the second layer recognition result is other actions.

[0053] Step 4.3: If the first layer recognition result is a possible walking action, set the second threshold B2 corresponding to the average value of the left and right hip joint angles and the third threshold C3 corresponding to the absolute value of the difference between the left and right hip joint angles.

[0054] When Mean i ≥B2 and Difference i When C3 is less than or equal to 3, the second layer recognition result is a walking action. When Mean i and Difference i When the value is other than 0, the second layer recognition result is other actions.

[0055] Step 5: Using the additional rules of the human body carrying action recognition algorithm based on the multi-layer threshold discrimination method, analyze and verify the second-layer recognition result of the human body carrying action according to the obtained second-layer recognition result, and obtain the final recognition result of the human body carrying action.

[0056] If the second-layer recognition result is a placement action or other actions, no additional rules are needed; the second-layer recognition result is the final recognition result. If the second-layer recognition result is a lifting action or walking action, additional rules are needed to verify the second-layer recognition result to obtain the final recognition result. The specific steps for adding rules are as follows:

[0057] Step 5.1: Boosting Additional Rules: Take the final recognition result at time i-1. If the second-layer recognition result at time i is a boosting action:

[0058] If the final recognition result at time i-1 is either a lifting action or a placing action, then the verification is successful, and the final recognition result at time i is still a lifting action.

[0059] If the final recognition result at time i-1 is walking, standing, or other actions, the verification fails, and the final recognition result at time i is changed to other actions.

[0060] Step 5.2: Additional rules for walking: Set a threshold D1 corresponding to the human motion feature values ​​analyzed by the sliding window. If the second layer recognition result at time i is a walking action:

[0061] When Mean_Difference i When ≥D1, the verification is successful, and the final recognition result at time i is the walking action.

[0062] When Mean_Difference i When <D1, the verification fails, and the final recognition result at time i is changed to an upright action.

[0063] Example

[0064] Reference Figure 1 The flowchart of a human carrying action recognition algorithm provided by the present invention is shown, which specifically includes the following steps:

[0065] Step 1: When the human body is being moved, the triaxial Euler angles and angular velocities in the sagittal plane of different parts of the human body are collected by inertial sensors worn on the human body.

[0066] In step 1, three inertial sensors are worn on the human body. One is placed at the center of the waist, and the other two are symmetrically distributed on both sides of the hip joint. The motion data of the waist and hip are measured when the human body moves, and the triaxial Euler angles and angular velocities in the sagittal plane of the waist and hip are obtained when the human body is performing a carrying action.

[0067] In the embodiments of the present invention, the sampling frequency of the inertial sensor used is 100Hz, and the specific time is represented by the frame number i.

[0068] Reference Figure 1 Step 2: Using a sliding window-based human movement preprocessing algorithm, based on the measured human movement data, analyze and calculate the feature vector of human movement at a certain moment for each data frame and a sliding data window with a given number of frames.

[0069] In step 2, the specific steps of the human body handling motion preprocessing algorithm based on the sliding window are as follows:

[0070] Step 2.1: During the human body handling process, collect the data of the lumbar angle and angular velocity of the left and right hip joints in the sagittal plane at time i, and calculate the average value Mean of the left and right hip joint angles at this time. i Difference between left and right hip joint angles i The absolute value of , together with the angle and angular velocity of the three parts at this moment, forms the feature vector of human motion for data frame analysis.

[0071]

[0072] Where β 1i ,β 2i These represent the left and right joint angles in the i-th data frame, respectively.

[0073] Step 2.2: During the human body handling process, collect the data of the left and right hip joint angles in the sagittal plane of the human body at time i and the preceding n data frames to form a sliding window of human motion data at time i. Calculate the absolute value of the difference between the left and right hip joint angles corresponding to each frame within the human motion data window. i The average value is the human motion characteristic value analyzed by the sliding window at time i.

[0074]

[0075] n is the number of data frames contained in the data window being retrieved.

[0076] In embodiments of the present invention, the number n of data frames contained in the data window can be selected according to the actual situation, generally between 20 and 40.

[0077] Step 2.3: Combine the human motion feature vector from the data frame analysis at time i with the human motion feature values ​​from the sliding window analysis to obtain the human carrying action feature vector f at time i. i =(α i ,ω i Mean i Difference i Mean_Difference i ).

[0078] Reference Figure 1 Step 3: Using the first-layer threshold discrimination method of the multi-layer human movement recognition algorithm, the human movement is identified based on the calculated human movement feature vector to obtain the possible human movement at a certain moment, which is the first-layer recognition result.

[0079] In embodiments of the present invention, the results identified by the multi-layer human transport action recognition algorithm are defined as five typical actions occurring during human transport: lifting, placing, walking, standing, and others. Only one action can occur at any given time. The first four actions are considered valid actions during transport, while the others are considered invalid.

[0080] Reference Figure 4 This illustrates the changing patterns of effective movements during typical human transport processes.

[0081] refer to Figure 3 This is a flowchart illustrating the implementation of a human movement recognition algorithm. The specific steps of its first-layer threshold discrimination method are as follows:

[0082] In an embodiment of the present invention, the possible human carrying actions obtained by the first-layer threshold discrimination method may include bending over and walking. The possible bending over actions are divided into two categories: possible lifting and possible placing. When the result obtained by the first-layer threshold discrimination method is another action, subsequent algorithms are not required, and the result obtained by the human carrying action recognition algorithm is the other action.

[0083] In an embodiment of the invention, human motion data from 3 to 5 complete lifting processes of the user are first measured as samples. Combined with the user's body characteristics and the actual installation position of the inertial sensor, three thresholds A1, A2, and A3 (A3 > A2 > A1) corresponding to the waist angle are set, along with a first threshold C1 corresponding to the absolute value of the difference between the left and right hip joint angles. The feature vector f is then taken. i The human waist angle α i and angular velocity ω i , will α i The possible carrying actions that the human body may perform are compared with A1, A2, and A3, which is the first layer of recognition results.

[0084] Step 3.1: First, calculate the angle α of the lumbar region in the sagittal plane of the human body at time i. i Compare with waist angle thresholds A1, A2, A3 (A3 > A2 > A1):

[0085] When A1≤α i If the value is ≤A2, then this action is first classified as a possible bending action.

[0086] When A2 < α i If the value is ≤A3, then this action is first classified as a possible walking action.

[0087] When α i When the value is other than 0, the first layer of recognition results in other actions.

[0088] Step 3.2: Under the condition that this action is a possible bending motion, if the angular velocity of the waist ω i If ≤0, the first layer recognition result is a possible placement action; if the waist angular velocity ω i If the result is greater than 0, then the first layer of recognition results is a possible action to improve performance.

[0089] Step 3.3: Given that this action is a possible walking action, if Difference i If C1 ≤ C1, the first layer recognition result is a possible walking action; if Difference i If the value is any other, the first layer of recognition will identify other actions.

[0090] Reference Figure 1 Step 4: Using the second-layer threshold discrimination method of the multi-layer human movement recognition algorithm, based on the calculated human motion feature vector and the first-layer recognition result, the second-layer recognition result of the human movement at a certain moment is obtained.

[0091] refer to Figure 3 This is a flowchart illustrating the implementation of a human movement recognition algorithm. The specific steps of its second-layer threshold discrimination method are as follows:

[0092] Take the eigenvector fi Mean of the average left and right hip joint angles at time i. i The absolute value of the difference between the left and right hip joint angles. i .

[0093] Step 4.1: If the first layer recognition result indicates a possible placement action:

[0094] Based on human movement data samples, combined with the user's body characteristics and the actual installation location of the inertial sensors, a first threshold B1 corresponding to the average left and right hip joint angles and a second threshold C2 corresponding to the absolute value of the difference between the left and right hip joint angles are set. i Difference i The results are compared with B1 and C2 respectively to obtain the second layer recognition results.

[0095] When Mean i ≤B1 and Difference i When C2 is less than or equal to 2, the second layer recognition result is a placement action; when Mean i and Difference i When the value is other than 0, the second layer recognition result is other actions.

[0096] Step 4.2: If the first-layer recognition result indicates a possible improvement action:

[0097] When Mean i ≤B1 and Difference i When C2 is less than or equal to 2, the second layer recognition result is a lifting action; when Mean i and Difference i When the value is other than 0, the second layer recognition result is other actions.

[0098] Step 4.3: If the first layer recognition result indicates a possible walking action:

[0099] If the first layer of recognition results indicates a possible walking action, set a second threshold B2 corresponding to the average angle of the left and right hip joints and a third threshold C3 corresponding to the absolute value of the difference between the angles of the left and right hip joints.

[0100] When Mean i ≥B2 and Difference i When C3 is less than or equal to 3, the second layer recognition result is a walking action; when Mean i and Difference i When the value is other than 0, the second layer recognition result is other actions.

[0101] Reference Figure 1Step 5: Using the additional rules of the multi-layer human movement recognition algorithm, analyze and verify the second-layer recognition results to obtain the final recognition result of human movement.

[0102] If the second-layer recognition result is a placement action or other actions, no additional rules are needed, and the second-layer recognition result is the final result. If the second-layer recognition result is a lifting action or walking action, additional rules are needed to verify the second-layer recognition result to obtain the final result.

[0103] By analyzing the movement patterns during human transport, specific relationships between human movements are identified: placement always precedes lifting; and uprighting occurs between walking and placement, and between lifting. Based on this, additional rules are proposed to verify the second-layer recognition results and ensure their accuracy. These additional rules are divided into lifting rules and walking rules.

[0104] refer to Figure 3 The diagram illustrates the implementation flowchart of a human movement recognition algorithm. The specific implementation steps for its additional rules are as follows: Step 5.1: Take the final recognition result at time i-1. If the second-layer recognition result at time i is a lifting action, then apply the lifting additional rules:

[0105] If the final recognition result at time i-1 is a lifting or placing action, the verification passes, and the final recognition result at time i remains a lifting action; if the final recognition result at time i-1 is walking, standing, or other actions, the verification fails, and the final recognition result at time i is changed to another action.

[0106] Step 5.2: Take the feature vector f i Mean_Difference of human motion features at time i in the sliding window analysis i Set the Mean_Difference within the window at time i. i If the threshold D1 is a walking action in the second layer at time i, then the walking additional rule is used:

[0107] When Mean_Difference i When <D1, the verification passes, and the second-layer recognition result at time i is still a walking action; when Mean_Difference i When <D1, the verification fails, and the second layer recognition result at time i is changed to an upright action.

[0108] In the human body handling action recognition algorithm provided by this invention, the multi-layer threshold discrimination method can quickly determine the specific action performed by the human body at a certain moment and ensure its stability. The calculation is simple, and the result is verified by additional rules to ensure its accuracy.

[0109] In embodiments of the present invention, all threshold data should be set according to the specific physical characteristics of the user. No adjustment may be needed when users have similar physical characteristics, but if high accuracy is required, the thresholds can be finely adjusted.

Claims

1. A method for recognizing human carrying actions based on a multi-layer threshold discrimination method, characterized in that, include: Step 1: When the human body is being moved, the Euler angles and angular velocities in the sagittal plane of different parts of the human body are collected by inertial sensors worn on the human body. Step 2: Using a sliding window-based preprocessing algorithm for human movement, based on the measured human motion data, analyze and calculate the feature vector of the human movement at a given moment for each data frame and a sliding data window of a given number of frames; the feature vector of the human movement at a given moment is represented as: Indicates the first The feature vector of human transport action at any given time. For the first The angle of the lumbar region in the sagittal plane of the human body at any given time. No. The angular velocity of the human body in the sagittal plane at any given moment. Indicates the first The average angle of the left and right hip joints at any given time. Indicates the first The absolute difference in the angle between the left and right hip joints at all times. , They represent the first time. The left and right hip joint angles in the time-series data frame. This represents the average of the absolute values ​​of the difference in the left and right hip joint angles corresponding to each frame within the human motion data window. The number of data frames contained in the captured data window; Step 3: Using the first-layer threshold discrimination method of the human body carrying action recognition algorithm based on the multi-layer threshold discrimination method, the human body carrying action is identified by the first-layer threshold discrimination method according to the calculated human motion feature vector, and the first-layer recognition result of the human body carrying action at a certain moment is obtained. Step 4: Using the second-layer threshold discrimination method of the human movement recognition algorithm based on the multi-layer threshold discrimination method, based on the calculated human motion feature vector and the first-layer recognition result obtained by the first-layer threshold discrimination method, the second-layer threshold discrimination method is used to obtain the second-layer recognition result of the human movement at a certain moment; specifically including: Step 4.1: If the first layer recognition result indicates a possible placement action, set the first threshold corresponding to the average angle of the left and right hip joints. The second threshold corresponding to the absolute value of the difference in angles between the left and right hip joints. , will the Average angle of left and right hip joints at any given time , No. The absolute difference in the angle between the left and right hip joints at any given moment Separately and The comparison yields the second-layer recognition result: when and At that time, the second layer recognition result is a placement action; when and When the value is other than 0, the second layer recognition result is other actions; Step 4.2: If the first-layer recognition result indicates a possible improvement action: when and At that time, the second layer recognition result is an lifting action; when and When the value is other than 0, the second layer recognition result is other actions; Step 4.3: If the first layer recognition result indicates a possible walking motion, set a second threshold corresponding to the average angle of the left and right hip joints. The third threshold corresponding to the absolute value of the difference in angles between the left and right hip joints. : when and At that time, the second layer recognition result is a walking action; when and When the value is other than 0, the second layer recognition result is other actions; Step 5: Using the additional rules of the human body carrying action recognition algorithm based on the multi-layer threshold discrimination method, analyze and verify the second-layer recognition result of the human body carrying action according to the obtained second-layer recognition result, and obtain the final recognition result of the human body carrying action. The additional rules specifically include: Step 5.1: Improve additional rules: Take the first... The final recognition result at time 1, if the first The second layer of recognition at a given moment is the lifting action: When the If the final recognition result at any given moment is either a lift or a place action, then the verification is successful. The final identification result at any given moment is still an elevation action; When the If the final recognition result at any given moment is walking, standing, or other actions, then the verification fails. The final recognition result of the moment is changed to other actions; Step 5.2: Additional walking rules: Set thresholds for the human motion feature values ​​analyzed by the sliding window. If the first The second layer of recognition at any given moment is the walking action: when At that time, the verification passed, the first The final recognition result at any given moment is a walking motion; when At that time, the verification failed. The final recognition result at that moment was changed to an upright posture; in This represents the average absolute value of the difference in the left and right hip joint angles corresponding to each frame within the human motion data window.

2. The human body handling action recognition method based on multi-layer threshold discrimination according to claim 1, characterized in that, Step 3 specifically includes: Step 3.1: Take the first eigenvector from the eigenvectors. Human waist angle at all times Set three waist angle thresholds in ,Will Separately and A comparison reveals the possible carrying actions that the human body can perform: when If so, first classify this action as a possible bending motion; when If so, first classify this action as a possible walking action; when When the value is other than 0, the first layer of recognition results in other actions; Step 3.2: Given that this action is a possible bending-over action, take the first element from the feature vector. Human waist angular velocity at all times ,Will The first layer of recognition result is obtained by comparing it with 0: If the waist angular velocity If so, the first layer of identification results is the possible placement action; If the waist angular velocity If so, the first layer of recognition results is a possible action to improve performance; Step 3.3: Under the condition that this action is a possible walking action, set the first threshold corresponding to the absolute value of the difference in the left and right hip joint angles. ,Will and The comparison yields the first-layer recognition result: like The first layer of recognition results then represents possible walking actions; like If the value is any other, the first layer recognition result will be "other action".

3. The human body handling action recognition method based on multi-layer threshold discrimination according to claim 1, characterized in that, The second layer of the human movement recognition algorithm based on the multi-level threshold discrimination method is defined as five typical movements that occur during human movement: lifting, placing, walking, standing, and others.

4. The human body handling action recognition method based on multi-layer threshold discrimination according to claim 1, characterized in that, If the second-layer recognition result is a placement action or other action, no additional rules are needed, and the second-layer recognition result is the final recognition result. If the second-layer recognition result is a lifting action or walking action, additional rules are needed to verify the second-layer recognition result to obtain the final recognition result.

5. The human body handling action recognition method based on multi-layer threshold discrimination according to claim 1, characterized in that, Three inertial sensors are worn on the human body. One is placed at the center line of the waist, and the other two are symmetrically distributed on both sides of the hip joint. They measure the motion data of the waist and hip during human movement, and obtain the angle and angular velocity data of the waist and hip in the sagittal plane when the human body is carrying.

Images