A gesture recognition and control method for a bionic manipulator

Abstract

The present invention discloses a gesture recognition and control method for a bionic manipulator, belonging to the field of human-computer interaction technology. The method comprises the following steps: S1. acquiring a hand image of a target user in real time; S2. identifying the three-dimensional coordinates of each fingertip relative to the palm in the hand image, calculating the displacement change between the current frame and the reference frame, and generating a first radar chart based on the displacement change values ​​of the five fingers; S3. identifying the angular change of each fingertip relative to the palm, and generating a second radar chart based on the angular change values ​​of the five fingers; S4. calculating the area ratio of the first radar chart to the second radar chart, and determining a gesture complexity weight value V based on the area ratio; S5. if V ≤ a threshold value K, initiating a first processing mode for gesture recognition; if V > the threshold value K, initiating a second processing mode for gesture recognition. The present invention dynamically allocates computing power based on the gesture complexity weight value V and constructs a synchronous information difference degree D model to achieve abnormality warning and intelligent attribution.

Description

Technical Field

[0001] The present invention relates to the technical field of human-computer interaction, and in particular to a gesture recognition and control method of a bionic manipulator. Background Art

[0002] As a human-machine interactive automated operation device, a bionic manipulator can automate a variety of complex movements. Existing bionic manipulator operating systems mostly perform simple, repetitive tasks and lack the ability to adapt to complex work scenarios. Human-machine interaction based on gesture recognition can enhance the manipulator's operational flexibility and meet the needs of practical applications.

[0003] Currently, bionic manipulator control methods primarily rely on contact and non-contact devices to achieve gesture recognition. However, the high cost of contact devices makes them difficult to implement in practical applications. Non-contact devices address the high cost of contact devices. With the advent of deep learning, image processing technology is used to communicate gesture recognition results from images to the bionic manipulator, enabling control of the bionic manipulator.

[0004] Gesture recognition technology based on image processing enables dexterous control of bionic manipulators. However, existing technologies still have some problems: gesture recognition uses a fixed computing power model, which consumes excessive resources for simple gestures and lacks accuracy for complex gestures; it also lacks the ability to diagnose and predict abnormal control commands in real time, leading to the risk of uncontrolled manipulator movements. Therefore, those skilled in the art have provided a gesture recognition and control method for a bionic manipulator to address the issues raised in the background technology. Summary of the Invention

[0005] In view of the defects in the prior art, the present invention provides a gesture recognition control method for a bionic manipulator, comprising the following steps:

[0006] S1. Acquire a hand image of the target user in real time through an image acquisition device;

[0007] S2. Identify the three-dimensional coordinates of each fingertip relative to the palm in the hand image, calculate the displacement change value between the current frame and the reference frame, and generate a first radar chart based on the displacement change values ​​of the five fingers;

[0008] S3 identifies the angle change value of each fingertip relative to the palm, and generates a second radar chart based on the angle change value of the five fingers;

[0009] S4. Calculating the area ratio of the first radar chart to the second radar chart, and determining the gesture complexity weight value V according to the area ratio;

[0010] S5. If V ≤ threshold K, the first processing mode is started for gesture recognition; if V> threshold K, the second processing mode is started for gesture recognition; wherein the computing resource consumption of the first processing mode is lower than that of the second processing mode;

[0011] S6. Convert the recognition results into robot control instructions and execute them;

[0012] S7. Real-time acquisition of synchronization information of user gestures and robot gestures, including: response time of each finger, position offset of the finger joints relative to the palm, and spatial vector distribution of the line connecting the fingertips and the palm;

[0013] S8. Calculate the synchronization information difference D between the user gesture and the robot hand gesture. When D > dynamic threshold M, determine that the control command is abnormal and issue a warning;

[0014] S9. When an instruction is abnormal, based on the feature vector of the current difference degree D, the top N groups of historical abnormal records with the highest similarity are matched from the historical database;

[0015] S10. Output the predicted list of abnormal causes corresponding to the historical abnormal records and sort them from high to low in probability; after the abnormality is resolved, enter the abnormal data and solution into the database to update the model.

[0016] As a further solution of the present invention: the calculation of the displacement change value in step S2 includes:

[0017] Taking the origin of the palm coordinate system as the reference, calculate the coordinate changes of the thumb, index finger, middle finger, ring finger, and little finger on the X / Y / Z axis respectively;

[0018] The first radar chart uses pentagonal vertices to represent five fingers, and the distance from the vertex to the center is proportional to the displacement change value of the corresponding finger.

[0019] As a further solution of the present invention, the calculation of the angle change value in step S3 includes:

[0020] In the palm plane projection, calculate the angle change between the fingertip vector of each finger and the palm reference axis;

[0021] The second radar chart uses pentagonal vertices to represent five fingers, and the distance from the vertex to the center is proportional to the angle change value of the corresponding finger.

[0022] As a further solution of the present invention: the calculation of the gesture complexity weight value V in step S4 satisfies:

[0023] V = α·(S1 / S2)+β·|S1-S2|, where S1 is the area of ​​the first radar chart, S2 is the area of ​​the second radar chart, and α and β are preset normalization coefficients.

[0024] As a further solution of the present invention: the difference between the first processing mode and the second processing mode in step S5 includes:

[0025] The first processing mode uses low-resolution image feature extraction and linear classifier;

[0026] The second processing mode uses high-resolution image feature extraction, convolutional neural network and three-dimensional posture optimization algorithm.

[0027] As a further solution of the present invention: the calculation of the difference D in step S8 includes:

[0028] D=γ·Δt+δ·ΔP+ε·ΔV, where:

[0029] Δt is the average response time difference of the five fingers;

[0030] ΔP is the sum of the Euclidean distances of the position offsets of each finger joint;

[0031] ΔV is the cosine similarity deviation of the angle between the fingertip and palm line vectors;

[0032] γ, δ, ε are weighting coefficients.

[0033] As a further solution of the present invention: the matching method of the historical database in step S9 includes:

[0034] Extract the following feature vectors of the current abnormal state: difference D value, displacement change value distribution entropy, angle change value variance, and abnormal duration;

[0035] Retrieve the top-N matching records in the historical database using the cosine similarity algorithm.

[0036] As a further solution of the present invention: the generation of the abnormality cause prediction list in step S10 includes:

[0037] Perform weight analysis on the matched historical anomaly records. The weight calculation formula is: Wi = Ai*Bi*Ci, where Wi is the weight of the historical anomaly record, Ai is the similarity, Bi is the resolution efficiency, Ci is the occurrence frequency, and i = 1···n, where n is a positive integer;

[0038] Arrange the output exception reason labels in descending order of Wi values.

[0039] As a further solution of the present invention: when the database is updated in step S10, a mapping relationship table between abnormal solutions and difference feature vectors is established, and the dynamic threshold M is optimized:

[0040] M new =η·M old +(1-η)·Dmax , where D max is the peak difference within this abnormal period, η is the forgetting factor, M new is the updated dynamic threshold, M old is the dynamic threshold before updating.

[0041] As a further solution of the present invention: when synchronously collecting information in step S7, an inertial measurement unit (IMU) and a fingertip pressure sensor installed on the mechanical finger joint are fused to generate a spatial vector distribution map.

[0042] The beneficial effects of the present invention are embodied in:

[0043] 1. This application dynamically switches processing modes through the V value, effectively reducing power consumption in complex gesture scenarios and enhancing security. It achieves millisecond-level anomaly detection through the difference D model, thereby significantly reducing the false alarm rate. In addition, this application has a high accuracy rate in anomaly attribution based on historical data, and the database continuously iterates and optimizes the dynamic threshold M, thereby achieving intelligent evolution. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly describes the drawings required for the specific embodiments or the description of the prior art. Similar elements or parts are generally identified by similar reference numerals throughout the drawings. Elements or parts in the drawings are not necessarily drawn to scale.

[0045] Figure 1 is a flow chart of a gesture recognition and control method for a bionic manipulator;

[0046] Figure 2 The figure is a flow chart for implementing a gesture recognition and control method for a bionic manipulator. DETAILED DESCRIPTION

[0047] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0048] As mentioned in the background technology of this application, research has found that the existing gesture recognition adopts a fixed computing power mode, simple gestures consume excessive resources, and the recognition accuracy of complex gestures is insufficient; and abnormal control instructions lack real-time diagnosis and prediction capabilities, resulting in the risk of uncontrolled robot movements, which has certain defects.

[0049] In order to address the above-mentioned defects, the present application discloses a gesture recognition control method for a bionic manipulator, which can optimize energy efficiency, dynamically switch processing modes through the V value, effectively reduce power consumption in complex gesture scenarios, and enhance security. It realizes millisecond-level anomaly detection through the difference D model, thereby significantly reducing the false alarm rate. In addition, the present application has a high accuracy rate in anomaly attribution based on historical data, and the database continuously iterates and optimizes the dynamic threshold M, thereby realizing intelligent evolution.

[0050] The following will describe in detail how the solution of this application solves the above technical problems with reference to the accompanying drawings.

[0051] See also Figure 1In an embodiment of the present invention, a gesture recognition control method for a bionic manipulator includes the following steps: S1. acquiring a hand image of a target user in real time through an image acquisition device; S2. identifying the three-dimensional coordinates of the fingertip of each finger relative to the palm in the hand image, calculating the displacement change value between the current frame and the reference frame, and generating a first radar map based on the displacement change values ​​of the five fingers; S3. identifying the angle change value of the fingertip of each finger relative to the palm, and generating a second radar map based on the angle change values ​​of the five fingers; S4. calculating the area ratio of the first radar map to the second radar map, and determining a gesture complexity weight value V according to the area ratio; S5. if V ≤ a threshold value K, starting a first processing mode for gesture recognition; if V > the threshold value K, starting a second processing mode for gesture recognition; wherein the calculation information of the first processing mode is The resource consumption is lower than in the second processing mode. S6. The recognition results are converted into robot control instructions and executed. S7. Synchronous information between the user's gesture and the robot's gesture is collected in real time. This information includes: the response time of each finger, the position offset of the knuckles relative to the palm, and the spatial vector distribution of the line connecting the fingertips and the palm. S8. The difference D between the synchronization information of the user's gesture and the robot's gesture is calculated. When D exceeds the dynamic threshold M, the control instruction is determined to be abnormal and an alert is issued. S9. When an instruction is abnormal, the top N groups of historical abnormal records with the highest similarity are matched from the historical database based on the feature vector of the current difference D. S10. A list of predicted abnormal causes corresponding to the historical abnormal records is output, sorted from high to low likelihood. After the abnormality is resolved, the abnormal data and solution are entered into the database to update the model. This setup covers a complete closed loop from gesture image acquisition, complexity analysis (V value calculation), dynamic processing mode selection, control instruction generation and execution, execution synchronization monitoring (D value calculation), abnormality detection and alerting, intelligent diagnosis (history matching), and system self-learning (database update). The core function of this process is to optimize energy consumption and performance through dynamic resource allocation, and to ensure control safety and reliability through real-time monitoring and intelligent diagnosis. This invention uses the V value to dynamically switch processing modes, effectively reducing power consumption in complex gesture scenarios and enhancing security. The difference D model enables millisecond-level anomaly detection, significantly reducing false alarm rates. Furthermore, this application achieves high accuracy in anomaly attribution based on historical data, and the database continuously iterates and optimizes the dynamic threshold M, thereby achieving intelligent evolution.

[0052] In this embodiment, the calculation of the displacement change value in step S2 includes: taking the origin of the palm coordinate system as the reference, respectively calculating the coordinate changes of the thumb, index finger, middle finger, ring finger, and little finger on the X / Y / Z axis; the first radar chart uses pentagonal vertices to represent the five fingers, and the distance from the vertex to the center is proportional to the displacement change value of the corresponding finger. This setting details the calculation method of the displacement change value and the generation rules of the first radar chart. Its function is to quantify the absolute position movement amplitude of the finger in three-dimensional space and visualize the displacement information of the five fingers as a geometric figure (first radar chart), providing basic spatial motion feature data for the subsequent calculation of the gesture complexity weight value V. In order to solve the problem of calculating the area of ​​the first radar chart, normalization processing is first performed, and then calculated using the polygon area formula. The normalization of the displacement change value is specifically as follows:

[0053] (i=1,2,···5);

[0054] in, is the displacement change value of the finger, Change the value for the preset displacement.

[0055] In this embodiment, the calculation of the angle change value in step S3 includes: calculating the angle change between the fingertip vector of each finger and the palm reference axis within the palm plane projection; the second radar chart uses pentagonal vertices to represent the five fingers, and the distance from the vertex to the center is proportional to the angle change value of the corresponding finger. This setting details the calculation method of the angle change value and the generation rules of the second radar chart. Its function is to quantify the bending or pointing angle change of the fingers relative to the palm plane, and visualize the angle information of the five fingers as a geometric figure (second radar chart), providing basic posture change feature data for the subsequent calculation of the gesture complexity weight value V. In order to solve the area calculation problem of the second radar chart, normalization processing is first performed, and then calculated using the polygon area formula. The angle change value normalization is specifically as follows:

[0056] (i=1,2,···5);

[0057] in, Change the value for the angle of the finger, Change the value for the preset angle.

[0058] In this embodiment, the gesture complexity weight V in step S4 is calculated according to the following formula: V = α·(S1 / S2)+β·|S1-S2|, where S1 is the area of ​​the first radar chart, S2 is the area of ​​the second radar chart, and α and β are preset normalization coefficients. This calculation is performed by nonlinearly combining the first and second radar chart areas S1 and S2, with a larger value indicating a more complex gesture spatial transformation. This setting defines the formula for calculating the gesture complexity weight V. Its function is to integrate the characteristics of two dimensions, displacement information and angle information, and comprehensively assess the overall spatial transformation complexity of the gesture through a nonlinear combination formula (including proportion and difference). The V value serves as a key decision-making indicator, directly determining the subsequent resource allocation strategy (selection of processing mode).

[0059] In this embodiment, the differences between the first and second processing modes in step S5 include: the first processing mode utilizes low-resolution image feature extraction and a linear classifier, resulting in a computational latency of less than 50ms and a 40% reduction in power consumption; the second processing mode utilizes high-resolution image feature extraction, a convolutional neural network, and a 3D pose optimization algorithm, improving recognition accuracy to 99.2%. This setup compares the technical features and performance differences between the two processing modes. Its function is to implement a dynamic resource allocation strategy: the first processing mode (low complexity) prioritizes speed and energy efficiency, using a lightweight algorithm to quickly recognize simple gestures; the second processing mode (high complexity) prioritizes accuracy, devoting more computing resources (high resolution, deep networks, and complex optimization) to accurately recognize complex gestures. The comparison of the V value with the threshold K is the decision-making mechanism that triggers this strategy switch.

[0060] In this embodiment, the calculation of the difference degree D in step S8 includes the following: D = γ·Δt + δ·ΔP + ε·ΔV, where Δt is the average response time difference of the five fingers; ΔP is the sum of the Euclidean distances of the position offsets of each finger joint; ΔV is the cosine similarity deviation of the angles between the fingertip-palm line vectors; and γ, δ, and ε are weighting coefficients. This setting defines the formula for calculating the synchronization information difference degree D. Its function is to quantitatively evaluate the comprehensive deviation between the user's intended gesture and the actual execution state of the robot hand. This formula integrates synchronization information from three key dimensions: time delay (Δt), spatial position offset (ΔP), and posture direction deviation (ΔV). Through weighted summation, it generates a single, comparable difference degree indicator D, providing an objective basis for anomaly determination.

[0061] In this embodiment, the historical database matching method in step S9 includes extracting the following feature vectors of the current abnormal state: difference D, displacement change distribution entropy, angle change variance, and abnormal duration; and searching the historical database for the top-N matching records using the cosine similarity algorithm. This setting describes the historical database matching method. When an abnormality is detected (D > M), it extracts the feature vector of the current abnormal state and uses similarity calculation (cosine algorithm) to intelligently retrieve the top N historical abnormality records in the historical database that are most similar to the current situation. This is the first step in intelligent diagnosis, aiming to identify reference historical cases.

[0062] In this embodiment, generating the predicted abnormality cause list in step S10 includes performing a weighted analysis on the matched historical abnormality records, using the weight calculation formula: Wi = Ai * Bi * Ci, where Wi represents the weight of the historical abnormality record, Ai represents the similarity, Bi represents the resolution efficiency, and Ci represents the occurrence frequency, with i = 1...n and n representing positive integers. The abnormality cause labels are then sorted in descending order by Wi value and output. This setting defines the method for generating the predicted abnormality cause list. Its function is to prioritize the historical records matched in step S9. A weight calculation formula (Wi = Ai * Bi * Ci) comprehensively considers each record's matching degree (Ai), resolution efficiency (Bi refers to whether the corresponding solution to the cause was effective and fast in the past), and occurrence frequency (Ci). The predicted abnormality causes are sorted in descending order by the calculated weight Wi, outputting a list of predicted abnormality causes ranked by likelihood, providing diagnostic recommendations to the user or maintenance system.

[0063] In this embodiment, when the database is updated in step S10, a mapping relationship table between abnormal solutions and difference feature vectors is established, and the dynamic threshold M is optimized: M new =η·M old +(1-η)·D max , where D max is the peak difference within this abnormal period, η is the forgetting factor, M new is the updated dynamic threshold, M old is the dynamic threshold before update. This setting describes the update mechanism of the database and dynamic threshold, which is used to achieve self-learning and continuous optimization of the system. The detailed features (feature vectors) of this anomaly and its final confirmed solution are mapped and stored in the database to enrich the historical knowledge base for future matching diagnosis. max ), using the smooth update formula (M new =η·M old +(1-η)·D max) Adjust the dynamic threshold M. The forgetting factor η controls the weight of the impact of new data on the historical threshold, so that the threshold can adapt to changes in the operating state of the system and improve the accuracy of anomaly detection.

[0064] In this embodiment, during synchronization information collection in step S7, the inertial measurement unit (IMU) installed at the manipulator's finger joints and the fingertip pressure sensors are fused to generate a spatial vector distribution map. This setup illustrates how synchronization information (particularly the spatial vector distribution map) is collected. Its function is to provide high-precision manipulator posture and state data. Utilizing multi-sensor fusion technology (the IMU provides joint angle and acceleration information, and the pressure sensors provide contact and force information), a comprehensive calculation is performed to generate a spatial vector distribution map reflecting the precise orientation and distance of each fingertip relative to the palm. This provides key input data for calculating the synchronization information difference D (particularly its ΔV component), ensuring comprehensive and accurate monitoring.

[0065] In order to further illustrate the present invention, a gesture recognition and control method for a bionic manipulator provided by the present invention is described in detail below with reference to embodiments.

[0066] Example 1, Objective: Demonstrate V value calculation and first processing mode triggering process

[0067] Step S1: Acquire the user's hand image through the camera.

[0068] Step S2: Calculate the displacement change of the current frame relative to the reference frame (unit: cm) based on the origin of the palm coordinate system:

[0069] thumb: =[0.1,0.2,0.05] → Euclidean distance 0.23cm;

[0070] index finger: [0.08,0.15,0.03]→0.17cm;

[0071] Middle finger: =[0.05,0.1,0.02]→0.11cm;

[0072] Ring finger: =[0.04,0.08,0.01]→0.09cm;

[0073] Little finger: =[0.03,0.05,0.01]→0.06cm;

[0074] set up =10cm, according to Calculate the normalized displacement values: thumb 0.023, index finger 0.017, middle finger 0.011, ring finger 0.009, little finger 0.006;

[0075] A first radar chart is generated based on the normalized displacement value, and the area of ​​the first radar chart is calculated using the polygon area formula: S1 = 0.000413.

[0076] Step S3: Calculate the angle change value on the palm projection plane (unit: °):

[0077] thumb: =3.2°;

[0078] index finger: =2.5°;

[0079] Middle finger: =1.8°;

[0080] Ring finger: =1.2°;

[0081] Little finger: =0.9°;

[0082] set up =90°, according to Calculate the normalized angle values: thumb 0.0356, index finger 0.0278, middle finger 0.02, ring finger 0.0133, little finger 0.01;

[0083] A second radar chart is generated based on the angle normalization value, and the area of ​​the second radar chart is calculated using the polygon area formula: S2 = 0.001094.

[0084] Step S4: Calculate the complexity weight V (α=0.6, β=0.4):

[0085] V=0.6*(S1 / S2)+0.4*|S1-S2|=0.6*(0.000413 / 0.001094)+0.4*|0.0004 13-0.001094|=0.6*0.3775+0.4*0.000681=0.2265+0.0002724=0.2268;

[0086] Step S5: Preset threshold K=0.5, because V=0.2268≤K, start the first processing mode.

[0087] The present invention dynamically switches processing modes through the V value, effectively reducing power consumption in complex gesture scenarios and enhancing security. It achieves millisecond-level anomaly detection through the difference D model, thereby significantly reducing the false alarm rate. In addition, the application has a high accuracy rate in anomaly attribution based on historical data, and the database continuously iterates and optimizes the dynamic threshold M, thereby achieving intelligent evolution.

[0088] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some or all of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and description of the present invention.

Claims

1. A gesture recognition and control method for a bionic manipulator, characterized in that: The following steps are involved: S1. Acquire a hand image of the target user in real time through an image acquisition device; S2. Identify the three-dimensional coordinates of each fingertip relative to the palm in the hand image, calculate the displacement change value between the current frame and the reference frame, and generate a first radar chart based on the displacement change values ​​of the five fingers; S3 identifies the angle change value of each fingertip relative to the palm, and generates a second radar chart based on the angle change value of the five fingers; S4. Calculating the area ratio of the first radar chart to the second radar chart, and determining the gesture complexity weight value V according to the area ratio; S5. If V ≤ threshold K, the first processing mode is started for gesture recognition; if V> threshold K, the second processing mode is started for gesture recognition; wherein the computing resource consumption of the first processing mode is lower than that of the second processing mode; S6. Convert the recognition results into robot control instructions and execute them; S7. Real-time acquisition of synchronization information of user gestures and robot gestures, including: response time of each finger, position offset of the finger joints relative to the palm, and spatial vector distribution of the line connecting the fingertips and the palm; S8. Calculate the synchronization information difference D between the user gesture and the robot hand gesture. When D > dynamic threshold M, determine that the control command is abnormal and issue a warning; S9. When an instruction is abnormal, based on the feature vector of the current difference degree D, the top N groups of historical abnormal records with the highest similarity are matched from the historical database; S10. Output the predicted list of abnormal causes corresponding to the historical abnormal records and sort them from high to low in terms of likelihood; after the abnormality is resolved, enter the abnormal data and solution into the database to update the model.

2. The gesture recognition control method of a bionic manipulator according to claim 1, characterized in that: The calculation of the displacement change value in step S2 includes: Taking the origin of the palm coordinate system as the reference, calculate the coordinate changes of the thumb, index finger, middle finger, ring finger, and little finger on the X / Y / Z axis respectively; The first radar chart uses pentagonal vertices to represent five fingers, and the distance from the vertex to the center is proportional to the displacement change value of the corresponding finger.

3. The gesture recognition control method of a bionic manipulator according to claim 2, characterized in that: The calculation of the angle change value in step S3 includes: In the palm plane projection, calculate the angle change between the fingertip vector of each finger and the palm reference axis; The second radar chart uses pentagonal vertices to represent five fingers, and the distance from the vertex to the center is proportional to the angle change value of the corresponding finger.

4. The gesture recognition and control method of a bionic manipulator according to claim 3, characterized in that: The calculation of the gesture complexity weight value V in step S4 satisfies: V = α·(S1 / S2)+β·|S1-S2|, where S1 is the area of ​​the first radar chart, S2 is the area of ​​the second radar chart, and α and β are preset normalization coefficients.

5. The gesture recognition and control method of a bionic manipulator according to claim 4, characterized in that: The differences between the first processing mode and the second processing mode in step S5 include: The first processing mode uses low-resolution image feature extraction and linear classifier; The second processing mode uses high-resolution image feature extraction, convolutional neural network and three-dimensional posture optimization algorithm.

6. The gesture recognition and control method of a bionic manipulator according to claim 5, characterized in that: The calculation of the difference D in step S8 includes: D=γ·Δt+δ·ΔP+ε·ΔV, where: Δt is the average response time difference of the five fingers; ΔP is the sum of the Euclidean distances of the position offsets of each finger joint; ΔV is the cosine similarity deviation of the angle between the fingertip and palm line vectors; γ, δ, ε are weighting coefficients.

7. The gesture recognition control method of a bionic manipulator according to claim 6, characterized in that: The matching method of the historical database in step S9 includes: Extract the following feature vectors of the current abnormal state: difference D value, displacement change value distribution entropy, angle change value variance, and abnormal duration; Retrieve the top-N matching records in the historical database using the cosine similarity algorithm.

8. The gesture recognition and control method of a bionic manipulator according to claim 7, characterized in that: The generation of the abnormality cause prediction list in step S10 includes: Perform weight analysis on the matched historical anomaly records. The weight calculation formula is: Wi = Ai*Bi*Ci, where Wi is the weight of the historical anomaly record, Ai is the similarity, Bi is the resolution efficiency, Ci is the occurrence frequency, and i = 1···n, where n is a positive integer; Arrange the output exception reason labels in descending order of Wi values.

9. The gesture recognition and control method of a bionic manipulator according to claim 8, characterized in that: When the database is updated in step S10, a mapping relationship table between abnormal solutions and difference feature vectors is established, and the dynamic threshold M is optimized: M new =η·M old +(1-η)·D max , where D max is the peak difference within this abnormal period, η is the forgetting factor, M new is the updated dynamic threshold, M old is the dynamic threshold before updating.

10. The gesture recognition and control method of a bionic manipulator according to claim 9, characterized in that: During the synchronous information collection in step S7, the inertial measurement unit and the fingertip pressure sensor installed at the mechanical finger joint are integrated to generate a spatial vector distribution map.

Images