Method and system for calibrating preset position of electric bionic robot

Abstract

The invention discloses a method and a system for calibrating a preset position of an electric bionic robot. The method comprises the following steps: acquiring a reference image; acquiring a real-time image, and performing difference analysis on the real-time image and the reference image; generating a calibration control instruction according to the difference analysis result; and driving the electric bionic robot to adjust the position and the posture based on the calibration control instruction, and returning to the step of collecting the real-time image until the difference analysis result meets the preset condition so as to adjust the electric bionic robot to be aligned with the preset position. The problems of low calibration precision, poor stability, insufficient environmental adaptability and the like caused by severe weather such as metal surface reflection, fog and rain and equipment corrosion in the prior art are solved, and the overall performance of preset position calibration of the electric bionic robot is remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of robots, in particular to a preset position calibration method and system of a power bionic robot. BACKGROUND

[0002] In the process of power inspection, the preset position calibration of the power bionic robot is a key link to ensure the accurate and efficient operation of the inspection, and the purpose is to enable the robot to stably and accurately reach the preset position, so as to ensure the detection accuracy of the power equipment. At present, in the existing technology, the preset position calibration of the power inspection robot is mostly to scan the environmental point cloud by using a laser radar, and to match the point cloud with a preset map to calculate the position deviation, and then to realize the calibration. For example, some substation inspection robots will obtain the point cloud data of the edge of the control cabinet by using a laser radar, and then combine the two-dimensional code image photographed by an industrial camera to realize the accurate alignment of the robot in the x / y direction, and complete the preset position calibration.

[0003] However, this preset position calibration method relying on the laser radar has obvious limitations and cannot meet the calibration requirements of the power bionic robot in a complex on-site environment. On the one hand, there are a large number of metal components in the on-site environment, and the reflection characteristics of the metal surface will cause the point cloud obtained by the laser radar to be distorted, so that the point cloud data cannot truly reflect the environmental characteristics, and then affect the accuracy of the position deviation calculation, resulting in a decrease in the preset position calibration accuracy. On the other hand, under adverse weather conditions such as fog and rain, the detection performance of the laser radar will be seriously affected, and even failure may occur, so that the effective collection and matching of the point cloud cannot be completed, resulting in the interruption of the preset position calibration work. In addition, as the use time increases, the power equipment may rust, and the rust will change the reflection characteristics of the equipment surface, so that the stability of the point cloud data obtained by the laser radar is reduced, the consistency of the matching of the point cloud and the preset map is affected, and then the stability of the preset position calibration is deteriorated, so that the robot cannot continuously and reliably reach the preset position. The existence of these problems seriously restricts the accuracy, stability and environmental adaptability of the preset position calibration of the power bionic robot.

[0004] Therefore, there is an urgent need for a preset position calibration method of a power bionic robot to solve the above problems. SUMMARY

[0005] In view of the deficiencies in the prior art, the present application provides a preset position calibration method and system of a power bionic robot, which solves the problems of low calibration accuracy, poor stability and insufficient environmental adaptability caused by the reflection of the metal surface, adverse weather such as fog and rain, and equipment rust, and significantly improves the overall performance of the preset position calibration of the power bionic robot.

[0006] The present application discloses a preset position calibration method of a power bionic robot, comprising:

[0007] acquiring a reference image;

[0008] collecting a real-time image, and performing difference analysis on the real-time image and the reference image;

[0009] generating a calibration control instruction according to the difference analysis result;

[0010] driving the electric bionic robot to adjust the position and posture based on the calibration control instruction, and returning to the step of collecting the real-time image until the difference analysis result meets a preset condition, so as to adjust the electric bionic robot to align with the preset position.

[0011] Optionally, the acquiring of the reference image comprises:

[0012] collecting an image of the electric bionic robot at the preset position and taking the image as the reference image.

[0013] Optionally, the difference analysis comprises pixel-level difference analysis, feature-level difference analysis and semantic-level difference analysis.

[0014] The difference analysis on the real-time image and the reference image comprises:

[0015] preprocessing the real-time image;

[0016] performing pixel-level difference analysis on the preprocessed real-time image and the reference image to calculate a mean square error and a structural similarity index;

[0017] performing feature-level difference analysis on the preprocessed real-time image and the reference image to calculate a geometric transformation matrix;

[0018] performing semantic-level difference analysis on the preprocessed real-time image and the reference image to calculate a local difference result;

[0019] obtaining the difference analysis result according to the mean square error, the structural similarity index, the geometric transformation matrix and the local difference result.

[0020] Optionally, the preprocessing comprises at least one of illumination normalization, noise suppression and distortion correction.

[0021] Optionally, the preset position calibration method of the electric bionic robot further comprises:

[0022] obtaining a weight coefficient based on a preset reinforcement learning model;

[0023] adjusting the weight distribution of the mean square error and the structural similarity index in the difference analysis result according to the weight coefficient.

[0024] Optionally, the generating of the calibration control instruction according to the difference analysis result comprises:

[0025] Calculate the deviation amount of the current position and posture of the electric bionic robot from the preset position based on the difference analysis result;

[0026] Obtain the posture data and the laser radar depth data, and correct the deviation amount based on the posture data and the laser radar depth data;

[0027] Generate a calibration control instruction based on the corrected deviation amount, wherein the calibration control instruction comprises a chassis movement instruction and a gimbal rotation instruction.

[0028] Optionally, drive the electric bionic robot to adjust the position and posture based on the calibration control instruction, comprising:

[0029] Convert the chassis movement instruction into a chassis motor control signal, and convert the gimbal rotation instruction into a gimbal motor control signal;

[0030] Adjust the spatial position of the electric bionic robot based on the chassis motor control signal, and adjust the shooting angle of the gimbal camera based on the gimbal motor control signal.

[0031] Optionally, the difference analysis result satisfies a preset condition, comprising:

[0032] Determine whether the difference analysis result is less than a preset threshold value;

[0033] If the difference analysis result is less than the preset threshold value, determine that the difference analysis result satisfies the preset condition, and update the real-time image to a preset standard template library.

[0034] Optionally, the real-time image is collected, comprising:

[0035] Collect the real-time image through the gimbal camera installed on the electric bionic robot based on a preset frame rate.

[0036] The application further discloses a preset position calibration system of an electric bionic robot, which is used for executing the preset position calibration method of the electric bionic robot.

[0037] The acquisition module is configured to acquire a reference image;

[0038] The collection and analysis module is configured to collect a real-time image and perform difference analysis on the real-time image and the reference image;

[0039] The instruction generation module is configured to generate a calibration control instruction according to the difference analysis result;

[0040] The calibration module is configured to drive the electric bionic robot to adjust the position and posture based on the calibration control instruction, and return to the step of collecting the real-time image until the difference analysis result satisfies the preset condition, so as to adjust the electric bionic robot to align with the preset position.

[0041] Compared with the prior art, the present application has the following beneficial effects:

[0042] 1. Through the pure visual closed-loop feedback mechanism, the problems such as point cloud distortion caused by metal surface reflection, equipment failure in bad weather, and matching stability affected by equipment rust are effectively overcome, the accuracy, stability and environmental adaptability of the power bionic robot preset position calibration are significantly improved, and it is ensured that the robot can continuously and reliably complete the preset position alignment to meet the needs of complex power inspection scenes.

[0043] 2. The acquisition of the reference image provides a unified and standard reference for subsequent difference analysis, ensures the reference consistency of the calibration process, and lays a foundation for realizing high-precision calibration. The real-time image is collected and compared with the reference image for difference analysis, the point cloud matching of the laser radar is replaced by the comparison of visual features, the interference of metal surface reflection and equipment rust on the detection result is avoided, and the stability and reliability of the calibration in complex environments are improved.

[0044] 3. The calibration control instruction is generated according to the difference analysis result, so that the control instruction can accurately reflect the position deviation of the robot, and the pertinence and effectiveness of the calibration adjustment are ensured; the robot is driven to adjust the position and attitude based on the calibration control instruction, and the collection and analysis steps are repeatedly executed to form a complete closed-loop calibration mechanism, which can continuously correct the deviation, ensure the final accurate alignment of the robot to the preset position, avoid the influence of bad weather such as fog and rain on the calibration process, ensure the continuity of the calibration work, and improve the calibration accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 The flowchart of the preset position calibration method of the power bionic robot provided by the present application is shown in the figure.

[0046] Figure 2 The structure diagram of the preset position calibration system of the power bionic robot provided by the present application is shown in the figure. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be described clearly and completely in combination with the drawings of the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0048] The present application will be described in further detail below in combination with the drawings.

[0049] As Figure 1As shown, the embodiment of the present application provides a preset position calibration method of the power bionic robot, which comprises the following steps:

[0050] Step S1, acquiring a reference image.

[0051] In the embodiment of the present application, the acquisition of the reference image is the basic step in the preset position calibration method of the power bionic robot, and the purpose is to provide an accurate and stable reference standard for the subsequent difference analysis of the real-time image, so as to ensure the reliability and accuracy of the calibration process.

[0052] It should be noted that the reference image refers to the image collected when the power bionic robot is in the preset position, which contains the standard visual features of the power equipment in the target inspection scene. These features can be the color, texture, relative position of each component, etc. of the equipment. These standard visual features are the key basis for subsequent comparison with the real-time collected image to calculate the position deviation. The preset position refers to a specific position or attitude that is pre-set and stored in the robot system, which is used to quickly and accurately perform repetitive tasks. The concept of preset position is commonly used in industrial robots, service robots, security robots or monitoring devices with a pan-tilt unit. The core purpose is to improve efficiency, reduce human intervention and ensure operation consistency. The pan-tilt unit (PTU) refers to a device that supports a camera and realizes horizontal (Pan) and vertical (Tilt) rotation, which can accurately control the shooting angle of the camera. In the present application, the pan-tilt unit is installed on the power bionic robot, which is used to adjust the camera angle to obtain images of different positions.

[0053] In the embodiment of the present application, the process of acquiring the reference image mainly includes the following operations: controlling the power bionic robot to move to the preset position, which is an ideal position pre-set according to the requirements of the power inspection task. In this position, the robot can clearly and comprehensively obtain the image information of the target power equipment. After the robot is stably in the preset position, the image of the target scene is shot by the pan-tilt camera installed on the robot, and the image is determined as the reference image and stored, preparing for the subsequent calibration process.

[0054] In the embodiment of the present application, when collecting the reference image, the quality of the image needs to be ensured. For example, the image should be clear without obvious blur or motion blur, and the influence of shaking during shooting on the accuracy of the reference image should be avoided. At the same time, the image should completely contain the key area of the target power equipment, ensuring that all necessary feature points can be covered during the subsequent comparison and analysis. Such a reference image can effectively play its role as a reference standard and provide a reliable basis for accurate calibration.

[0055] The reference image is acquired by, for example, collecting an image of the power bionic robot at the preset position and taking the image as the reference image.

[0056] In summary, the acquisition of the reference image is the starting point of the entire preset position calibration method, and its quality directly affects the accuracy of subsequent difference analysis and the precision of the final calibration. By collecting the reference image containing standard visual features at the preset position, a solid foundation is laid for the precise calibration of the power bionic robot.

[0057] In step S2, real-time images are collected, and difference analysis is performed on the real-time images and the reference image.

[0058] In the embodiment of the present application, real-time high-precision calibration of the robot position is achieved through a visual closed-loop feedback mechanism. This step includes collecting real-time images, performing multi-dimensional difference analysis on the real-time images and the reference image, and calibrating the robot position according to the difference analysis results. This step does not rely on additional sensors such as laser radar or IMU, and has good environmental adaptability and real-time performance.

[0059] Collecting real-time images includes collecting real-time images based on a preset frame rate through a gimbal camera installed on the power bionic robot. The gimbal camera can rotate in the horizontal and vertical directions to adjust the shooting angle. The preset frame rate can be set to 30 fps to ensure the real-time performance of image acquisition.

[0060] In the embodiment of the present application, the difference analysis includes pixel-level difference analysis, feature-level difference analysis, and semantic-level difference analysis. The difference analysis between the real-time image and the reference image includes: preprocessing the real-time image; performing pixel-level difference analysis on the preprocessed real-time image and the reference image to calculate the mean square error and the structural similarity index; performing feature-level difference analysis on the preprocessed real-time image and the reference image to calculate the geometric transformation matrix; performing semantic-level difference analysis on the preprocessed real-time image and the reference image to calculate the local difference result; and obtaining the difference analysis result according to the mean square error, the structural similarity index, the geometric transformation matrix, and the local difference result. The preprocessing includes at least one of illumination normalization, noise suppression, and distortion correction.

[0061] It should be noted that illumination normalization is used to solve the problem of uneven image brightness caused by strong light or shadow, and can be realized by Retinex algorithm. The normalized pixel value is obtained by calculating the difference between the logarithm of the original pixel value and the logarithm of the convolution result of the Gaussian kernel, so that the image brightness characteristics under different illumination conditions remain consistent. Noise suppression is used to reduce the noise in the image. It can be combined with bilateral filtering and non-local mean filtering to set appropriate spatial standard deviation and gray standard deviation to retain image edge details while reducing noise. Distortion correction is used to eliminate the geometric deformation of the image caused by the optical characteristics of the camera. Based on the camera calibration parameters (including intrinsic matrix and distortion coefficient), the real-time image and the reference image have comparability in geometric shape. The preprocessing is performed in the edge calculation unit, and the specific implementation manner is as follows:

[0062] (1) Illumination normalization: Retinex algorithm is used to eliminate the influence of strong light or shadow, and the formula is as follows:

[0063] I norm (x,y)=log(I(x,y))-log(G*I(x,y))

[0064] Where I(x,y) is the original pixel value, G is the Gaussian kernel, and * represents convolution operation.

[0065] (2) Noise suppression: combined with bilateral filtering and non-local mean filtering, the parameter setting is spatial standard deviation σ s =10, gray standard deviation σ r =0.1.

[0066] (4) Distortion correction: based on camera calibration parameters (intrinsic matrix K, distortion coefficients k1, k2, k3, p1, p2), realized by undistort() function of OpenCV.

[0067] It should be understood that the difference analysis between the preprocessed live image and the reference image involves pixel-level, feature-level, and semantic-level difference analysis, with the final difference analysis result synthesized. Pixel-level difference analysis quantifies the overall difference between the live image and the reference image by calculating the mean squared error (MSE) and the structural similarity index (SI). The MSE reflects the overall deviation by comparing the numerical differences between corresponding pixels in the two images. The SI assesses the similarity between the two images based on brightness, contrast, and structure. The combination of these two provides a comprehensive picture of the overall consistency of the images. Feature-level difference analysis is used to determine the geometric deviation between the two images in terms of spatial position. The process involves feature extraction, matching optimization, and calculation of the geometric transformation matrix. Feature extraction utilizes the ORB algorithm, identifying feature points in the image through FAST corner detection, and then describing them using the BRIEF descriptor. Matching optimization eliminates mismatches using a random sampling consensus algorithm to ensure the accuracy of feature point correspondence. Based on the optimized matching point pairs, a geometric transformation matrix is ​​calculated. This matrix quantifies the spatial geometric transformation relationship between the live image and the reference image, providing a basis for position deviation calculation. Semantic-level difference analysis is used to assess local differences between two images in the areas of key components of power equipment. This is achieved by deploying a lightweight CNN model (such as MobileNet). This model extracts semantic features of key equipment components (such as insulators and switches) from the preprocessed real-time image and the reference image. It then calculates the cosine similarity of the two sets of feature vectors to obtain local difference results, which specifically reflect the impact of changes in the state of key components on the image. The specific implementation of difference analysis is as follows:

[0068] (1) Pixel-level difference analysis: Calculate the mean square error (MSE) and structural similarity index (SSIM) to quickly screen image differences.

[0069] The MSE formula is:

[0070]

[0071] The SSIM formula is:

[0072]

[0073] Among them, I is the real-time image, T is the reference image, μ is the mean, σ is the variance, σ IT is the covariance, c1 and c2 are constants.

[0074] (2) Feature-level difference analysis: Extract ORB feature points, use FAST corner detection and BRIEF descriptor, set the number of feature points to 500, and the number of pyramid levels to 8. Optimize the matching using the RANSAC algorithm, set the maximum number of iterations to 200, and the reprojection error threshold to 3 pixels. Calculate the homography matrix H:

[0075] H = findHomography(pts T , pts I , RANSAC);

[0076] where pts T and pts I are the feature point coordinates of the reference image and real-time image, respectively.

[0077] (3) Semantic difference analysis: deploy a lightweight CNN model (such as MobileNet) to extract the semantic features of key components of the equipment (such as insulators and knife switches), and evaluate the local difference through cosine similarity:

[0078]

[0079] where f I and f T are the feature vectors of the real-time image and the reference image, respectively.

[0080] In the embodiments of the present application, the online learning mechanism is also combined in the difference analysis process to dynamically optimize the accuracy of the difference analysis result. This includes obtaining the weight coefficient based on the preset reinforcement learning model, and then adjusting the weight distribution of the mean square error and the structural similarity index in the difference analysis result according to the weight coefficient. Wherein, the state space of the reinforcement learning model is the current environmental parameters, such as the light intensity and the equipment rust degree, the action space is the weight coefficient, and the reward function is the matching success rate of the difference analysis result and the actual position deviation. Through continuous learning, the difference analysis can be self-adaptive to different environmental conditions.

[0081] To ensure that the reference image can adapt to environmental changes, a dynamic template updating mechanism is also used. When the difference analysis results of a continuous preset number of real-time images and the reference image are all less than the threshold value, these real-time images are merged into the reference image library, and the reference image is updated using the exponential weighted average method. The new reference image is calculated by weighting the original reference image and the current real-time image through the forgetting factor, so that the reference image can reflect the slow changes of the equipment state and avoid the large deviation between the reference and the actual scene caused by long-term use.

[0082] For example, when the difference of a plurality of consecutive frames (such as 5 frames) of images is less than a set threshold value, the current image is merged into the reference image library and updated using the exponential weighted average:

[0083] T new = a * I + (1-a) * T old ;

[0084] where a is the weight coefficient, I is the current image, T old and T newrespectively, are the reference images before and after updating.

[0085] In summary, by real-time image acquisition and preprocessing to improve data quality, comprehensive capture of image difference characteristics through multi-dimensional difference analysis, and combination of online learning and dynamic template updating mechanism to realize environmental self-adaptation, these steps work together to provide precise and stable difference analysis results for power bionic robot preset position calibration, and lay a solid foundation for subsequent generation of calibration control instructions.

[0086] Step S3, generating calibration control instructions according to the difference analysis results.

[0087] In the embodiment of the present application, by calculating the position and attitude deviation and fusing multi-sensor data to correct the deviation, the control instructions for driving the chassis and the gimbal movement are finally generated, realizing high-precision closed-loop position calibration.

[0088] For example, generating calibration control instructions according to the difference analysis results includes: calculating the deviation amount of the current position and attitude of the power bionic robot from the preset position based on the difference analysis results; obtaining attitude data and laser radar depth data, and correcting the deviation amount based on the attitude data and laser radar depth data; generating calibration control instructions based on the corrected deviation amount, wherein the calibration control instructions include chassis movement instructions and gimbal rotation instructions.

[0089] It should be noted that the deviation amount of the current position and attitude of the power bionic robot from the preset position is calculated based on the difference analysis results. The deviation amount is a quantity that integrates translation and rotation information, which is used to quantify the difference between the current state and the target state. The deviation amount is calculated from the results of the aforementioned multi-dimensional difference analysis (including pixel-level, feature-level and semantic-level analysis), for example, the offset and rotation angle in the image plane can be derived from the homography matrix H, and then mapped to the actual motion space of the robot.

[0090] Obtaining attitude data and laser radar depth data, and correcting the deviation amount based on the attitude data and laser radar depth data. Attitude data can be provided by an inertial measurement unit (IMU), containing information such as pitch angle, roll angle and yaw angle; laser radar depth data provides point cloud distance information of the surrounding environment. Since pure visual calculation may produce errors or transient instability under certain conditions, such as fast motion and texture missing area, introducing multi-sensor fusion strategy can effectively correct the deviation amount and improve the robustness and accuracy of the system.

[0091] For example, the visual positioning error can be corrected by Kalman filtering algorithm. The state equation can be expressed as:

[0092] x k =Ax k-1 +Bu k +wk ;

[0093] The observation equation can be expressed as:

[0094] z k =Hx k +v k ;

[0095] where x k is the state vector (may include position, velocity, attitude), u k is the control input, z k is the observation vector (fusion of IMU and lidar data), w k and v k are process noise and observation noise respectively. Through filtering, more reliable position and attitude estimation is obtained, and the corrected deviation is output.

[0096] Finally, based on the corrected deviation, calibration control instructions are generated, including chassis movement instructions and gimbal rotation instructions. The chassis movement instructions are used to control the motor of the robot chassis to realize forward, backward or rotation; the gimbal rotation instructions are used to control the horizontal or vertical rotation of the gimbal motor. The specific parameters (such as PWM duty ratio, pulse number or target angle) of the control instructions are proportional to the corrected deviation or generated through PID controller and other control algorithms to ensure smooth and accurate approximation to the preset position.

[0097] In summary, the process of generating calibration control instructions based on the difference analysis results converts abstract image differences into specific action parameters, and combines multi-source data correction and optimization to ensure the accuracy and reliability of the control instructions, providing clear and effective guidance for subsequent position and attitude adjustment of the robot, and is an important bridge to realize accurate calibration of the preset position.

[0098] Step S4, based on the calibration control instructions, drive the electric bionic robot to adjust the position and attitude, and return to step S2 until the difference analysis result meets the preset condition, to adjust the electric bionic robot to align with the preset position.

[0099] In the embodiment of the application, the robot is driven to adjust according to the calibration control instructions, and precise alignment with the preset position is realized through a closed-loop feedback cycle. This method continuously collects images, analyzes differences and drives adjustment until the preset condition is met, completing the calibration process.

[0100] Illustratively, based on the calibration control instructions, the electric bionic robot adjusts the position and attitude, including: converting the chassis movement instructions into chassis motor control signals, and converting the gimbal rotation instructions into gimbal motor control signals; based on the chassis motor control signals, adjusting the spatial position of the electric bionic robot, and based on the gimbal motor control signals, adjusting the shooting angle of the gimbal camera.

[0101] In the embodiments of the present application, the chassis movement instruction is converted into a chassis motor control signal, and the gimbal rotation instruction is converted into a gimbal motor control signal. The chassis movement instruction and the gimbal rotation instruction are digital instructions generated by the upper control unit according to the corrected deviation. The conversion process is usually completed by a motor driver or a special motion control card, which converts the digital instruction into an analog voltage signal or a pulse signal (such as a PWM signal) executable by the motor.

[0102] For example, the chassis motor control signal can be a PWM wave with a certain duty cycle, used to control the speed and direction of a DC motor; the gimbal motor control signal can be a pulse sequence with adjustable pulse number and frequency, used to control the precise angle of a stepper motor. Subsequently, the spatial position of the electric bionic robot is adjusted based on the chassis motor control signal, and the shooting angle of the gimbal camera is adjusted based on the gimbal motor control signal. The chassis motor generates torque after receiving the control signal, drives the wheels to move, and thus changes the global pose of the robot (X, Y coordinates and heading angle). The gimbal motor generates rotational movement after receiving the control signal, and drives the camera on it to rotate horizontally or vertically to accurately adjust the field of view angle.

[0103] It should be noted that after the robot is driven to adjust the position and attitude, the step of collecting real-time images is returned, and the difference analysis is repeatedly performed until the difference analysis result meets the preset condition. Whether the preset condition is met is determined by judging whether the difference analysis result is less than the preset threshold. The difference analysis result is a comprehensive evaluation value, which can be obtained by weighting and fusing pixel-level MSE / SSIM, feature-level matching error, and semantic-level similarity. The comprehensive value is compared with a threshold calibrated in advance through experiments.

[0104] In the embodiments of the present application, the process of determining whether the difference analysis result meets the preset condition includes: judging whether the difference analysis result is less than the preset threshold; if the difference analysis result is less than the preset threshold, it is determined that the difference analysis result meets the preset condition, and the real-time image is updated to the preset standard template library.

[0105] It should be understood that if the difference analysis result is less than the preset threshold, it is determined that the difference analysis result meets the preset condition, at this time, it indicates that the real-time image and the reference image have been sufficiently aligned, and the electric bionic robot has reached the target preset position. Subsequently, the real-time image is updated to the preset standard template library. The updating operation aims to enable the reference template to adapt to the slow changes of the environment, such as slight rust of the equipment and seasonal changes of the light features, and improve the long-term robustness of the system.

[0106] In summary, by converting the calibration control instruction into precise mechanical action, combining closed-loop adjustment strategy and differentiated action priority, and continuously verifying and dynamically updating the reference based on the preset threshold, this link realizes the stable alignment of the robot from the deviation state to the preset position, completes the closed-loop process of the entire preset position calibration, and ensures that the power bionic robot can continuously and reliably perform the inspection task in a complex environment.

[0107] As shown in Figure 2 The embodiment of the application also provides a preset position calibration system of a power bionic robot for executing the preset position calibration method of the power bionic robot. The preset position calibration system of the power bionic robot comprises an acquisition module 201, an acquisition and analysis module 202, an instruction generation module 203 and a calibration module 204.

[0108] The acquisition module 201 is configured to acquire a reference image.

[0109] The acquisition and analysis module 202 is configured to acquire a real-time image and perform difference analysis on the real-time image and the reference image.

[0110] The instruction generation module 203 is configured to generate a calibration control instruction according to the difference analysis result.

[0111] The calibration module 204 is configured to drive the power bionic robot to adjust the position and the attitude based on the calibration control instruction, and return to the step of acquiring the real-time image until the difference analysis result meets the preset condition, so as to adjust the power bionic robot to align with the preset position.

[0112] From the above technical solutions, the preset position calibration method and system of the power bionic robot are disclosed. Through multi-dimensional visual difference analysis, dynamic template updating, lightweight edge computing and anti-interference design, the problems such as point cloud distortion caused by metal surface reflection, equipment failure in bad weather and influence of matching stability caused by equipment rust are effectively overcome. The accuracy, stability and environmental adaptability of the preset position calibration of the power bionic robot are significantly improved, so that the robot can continuously and reliably complete the preset position alignment, meet the needs of complex power inspection scenes, and ensure the continuity of the calibration work of the power bionic robot.

[0113] In summary, the preset position calibration method and system of the power bionic robot of the application are outstanding in improving technical performance, optimizing economic benefits, creating social benefits and the like, and provide a feasible solution for the precision and intelligent development of the power inspection field.

[0114] The above merely describes the preferred embodiments of the present application, and is not used to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for calibrating a preset position of an electric bionic robot, characterized in that: include: Acquire a reference image; Acquiring a real-time image, and performing a difference analysis between the real-time image and the reference image; Generate calibration control instructions based on the difference analysis results; Based on the calibration control instruction, the electric bionic robot is driven to adjust its position and posture, and the step of acquiring real-time images is returned to until the difference analysis result meets the preset condition, so as to adjust the electric bionic robot to align with the preset position.

2. The preset position calibration method of the electric bionic robot according to claim 1, characterized in that: The obtaining of the reference image comprises: An image of the electric bionic robot at a preset position is collected and used as a reference image.

3. The preset position calibration method of the electric bionic robot according to claim 1, characterized in that: The difference analysis includes pixel-level difference analysis, feature-level difference analysis and semantic-level difference analysis; The performing difference analysis between the real-time image and the reference image includes: Preprocessing the real-time image; Performing pixel-level difference analysis on the preprocessed real-time image and the reference image to calculate a mean square error and a structural similarity index; Performing the feature-level difference analysis on the pre-processed real-time image and the reference image to calculate a geometric transformation matrix; Performing the semantic level difference analysis on the preprocessed real-time image and the reference image to calculate a local difference result; The difference analysis result is obtained according to the mean square error, the structural similarity index, the geometric transformation matrix and the local difference result.

4. The preset position calibration method of the electric bionic robot according to claim 3, characterized in that: The preprocessing includes at least one of illumination normalization, noise suppression and distortion correction.

5. The preset position calibration method of the electric bionic robot according to claim 3, characterized in that: Also includes: Obtain weight coefficients based on a preset reinforcement learning model; The weight distribution of the mean square error and the structural similarity index in the difference analysis result is adjusted according to the weight coefficient.

6. The preset position calibration method of the electric bionic robot according to claim 1, characterized in that: Generating a calibration control instruction according to the difference analysis result includes: Calculating the deviation between the current position and posture of the electric bionic robot and the preset position based on the difference analysis result; Acquire posture data and laser radar depth data, and correct the deviation based on the posture data and the laser radar depth data; The calibration control instruction is generated based on the corrected deviation, wherein the calibration control instruction includes a chassis movement instruction and a pan / tilt rotation instruction.

7. The preset position calibration method of the electric bionic robot according to claim 6, characterized in that: The step of driving the electric bionic robot to adjust its position and posture based on the calibration control instruction includes: Converting the chassis movement instruction into a chassis motor control signal, and converting the pan / tilt rotation instruction into a pan / tilt motor control signal; The spatial position of the electric bionic robot is adjusted based on the chassis motor control signal, and the shooting angle of the pan-tilt camera is adjusted based on the pan-tilt motor control signal.

8. The preset position calibration method of the electric bionic robot according to claim 1, characterized in that: The difference analysis results meet the preset conditions, including: Determining whether the difference analysis result is less than a preset threshold; If the difference analysis result is less than a preset threshold, it is determined that the difference analysis result meets a preset condition, and the real-time image is updated to a preset standard template library.

9. The preset position calibration method of the electric bionic robot according to claim 1, characterized in that: The real-time image acquisition includes: Based on a preset frame rate, the real-time image is collected by a pan-tilt camera installed on the electric bionic robot.

10. A preset position calibration system for an electric bionic robot, used to execute the preset position calibration method for an electric bionic robot according to any one of claims 1 to 9, characterized in that: include: The acquisition module is configured to: acquire a reference image; The acquisition and analysis module is configured to: acquire real-time images and perform difference analysis between the real-time images and the reference images; The instruction generation module is configured to: generate a calibration control instruction according to the difference analysis result; The calibration module is configured to: drive the electric bionic robot to adjust its position and posture based on the calibration control instruction, and return to the step of collecting real-time images until the difference analysis result meets the preset conditions, so as to adjust the electric bionic robot to align with the preset position.

Images