Robot 3D visual positioning error correction method and system based on deep learning

Abstract

The invention provides a robot 3D visual positioning error correction method and system based on deep learning, and relates to the technical field of robot visual positioning, and the method comprises the steps: obtaining the initial position information of a robot in a three-dimensional space, extracting image features through a deep learning model, calculating a displacement error, comparing the displacement error with a distance threshold value, and executing the correction step. Through multi-sensor data fusion and dynamic target position adjustment, the positioning precision and reliability of the robot in a complex environment are optimized. The invention aims to improve the positioning accuracy of the robot in a three-dimensional space through deep learning and a multi-sensor fusion technology and reduce the positioning deviation caused by environment change and sensor errors, so that the more intelligent and autonomous robot navigation capability is realized.

Description

Technical Field

[0001] The present invention relates to the field of robot vision positioning technology, and in particular to a deep learning robot 3D vision positioning error correction method and system. Background Art

[0002] With the rapid development of robotics technology, intelligent robots have been widely used in industrial automation, unmanned driving, service robots and other fields. In these application scenarios, accurate positioning and navigation are the key to achieving autonomous operation of robots. However, the positioning accuracy of robots in complex environments is often affected by many factors, including sensor noise, ambient light changes, dynamic obstacles and measurement errors. These factors can cause deviations in the robot's position information, affecting its work efficiency and safety.

[0003] Traditional positioning methods mainly rely on geometric calculations, sensor fusion, and filtering techniques. However, these methods often fail to achieve ideal results when dealing with nonlinear and dynamic environments. For example, the Kalman filter may produce large errors when dealing with high-noise and rapidly changing environments. The visual positioning method based on feature extraction is limited by the sparseness of environmental features and changes in illumination, making it difficult to ensure stable performance.

[0004] In recent years, the rapid development of deep learning technology has provided new ideas for solving the above problems. Deep learning models, especially convolutional neural networks (CNNs), have shown superior performance in image recognition and feature extraction. By training on a large amount of data, deep learning models can automatically learn high-dimensional features, thereby improving the accuracy and robustness of positioning. However, existing deep learning methods still face the problem of insufficient positioning error correction in practical applications, especially in dynamic environments. Summary of the invention

[0005] In order to solve the technical problems in the prior art of insufficient positioning accuracy and low error correction efficiency of robots in complex environments, the present invention provides a deep learning robot 3D vision positioning error correction method and system.

[0006] The technical solution provided by the present invention is as follows:

[0007] First aspect:

[0008] The present invention provides a deep learning robot 3D visual positioning error correction method, comprising:

[0009] S1. Obtain the initial position information P of the robot in three-dimensional space initial , and define the scaling factor Rscale and distance threshold D threshold , the scaling factor has a value range of (0,2), indicating the amplification or reduction of the error, and the distance threshold is in meters, indicating the maximum acceptable positioning error;

[0010] S2. Use the deep learning model to extract features from the acquired image data and calculate the displacement error E of the robot relative to the target position. pos =P target -P current , where P target is the target position, P current is the current location information;

[0011] S3, compare the displacement error with the distance threshold, if ||E pos ||>D threshold , perform a calibration step;

[0012] S4, the correction step is calculated by the following formula:

[0013] P currected =P current +R scale ·E pos

[0014] Among them, P corrected is the corrected position information, and then the corrected information is smoothed.

[0015] Second aspect:

[0016] The present invention provides a deep learning robot 3D visual positioning error correction system, comprising:

[0017] processor;

[0018] A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the robot 3D visual positioning error correction method based on deep learning as described in the first aspect is implemented.

[0019] The beneficial effects brought about by the technical solution provided by the present invention include at least:

[0020] (1) In the present invention, the feature extraction capability of the deep learning model significantly improves the robot's recognition accuracy of displacement errors in complex environments. This innovation enables the robot to more accurately locate its position relative to the target, reducing the impact of sensor noise and environmental changes, thereby improving the reliability and stability of overall positioning. This high-precision error correction capability helps the robot significantly reduce potential safety risks when performing autonomous navigation and tasks;

[0021] (2) In the present invention, the introduction of a dynamic correction mechanism enables the robot to respond to environmental changes in real time. When the displacement error is detected to exceed the set distance threshold, the system can quickly execute the correction step, thereby effectively reducing the positioning error. This real-time adjustment capability makes the robot more flexible in complex and dynamic working environments, able to adapt to rapidly changing conditions, and improve work efficiency;

[0022] (3) In the present invention, the method of fusing multi-sensor data effectively enhances the integrity and redundancy of information. By combining the data of visual sensors, lidar and inertial measurement units, the system can maintain high positioning accuracy even when information is lost or a single sensor fails. This redundant design improves the robustness of the system and ensures that the robot can operate stably under a variety of working conditions, thereby broadening its application range. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0024] Figure 1 A schematic diagram of a process flow of a deep learning robot 3D visual positioning error correction method provided by an embodiment of the present invention;

[0025] Figure 2 A schematic diagram of the structure of a deep learning robot 3D vision positioning error correction system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0026] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0027] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either of the two.

[0028] In the embodiments of the present invention, "image" and "picture" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same. "of", "corresponding, relevant" and "corresponding" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same.

[0029] In the embodiments of the present invention, sometimes a subscript such as W1 may be mistakenly written as a non-subscript such as W1. When the difference is not emphasized, the meanings to be expressed are consistent.

[0030] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0031] Reference Manual Attached Figure 1 , shows a flow chart of a deep learning robot 3D vision positioning error correction method provided in an embodiment of the present invention.

[0032] The embodiment of the present invention provides a deep learning robot 3D visual positioning error correction method, which can be implemented by a deep learning robot 3D visual positioning error correction device, which can be a terminal or a server. The processing flow of the deep learning robot 3D visual positioning error correction method can include the following steps:

[0033] S1. Obtain the initial position information P of the robot in three-dimensional space initial , and define the scaling factor R scale and distance threshold D threshold The scaling factor has a value range of (0,2), which indicates the enlargement or reduction of the error. The distance threshold is in meters, which indicates the maximum acceptable positioning error.

[0034] It should be noted that obtaining the initial position information of the robot in three-dimensional space is the basis of the entire positioning process. The initial position is the reference point for subsequent calculations and is crucial for accurately tracking the robot's motion. In addition, the purpose of defining the scaling factor and distance threshold is to reasonably control the way errors are handled. The scaling factor is used to adjust the sensitivity of the positioning error, which can amplify or reduce the impact of the error in certain situations, making the system more adaptable in different environments. The distance threshold sets the maximum allowable positioning error to ensure that the robot can remain within a safe and effective operating range when performing tasks, thereby improving the reliability of the overall system.

[0035] S2. Use the deep learning model to extract features from the acquired image data and calculate the displacement error E of the robot relative to the target position. pos =P target -P current , where P target is the target position, P current It is the current location information.

[0036] It should be noted that using deep learning models to extract features from acquired image data is a key step in improving positioning accuracy. Deep learning algorithms can automatically identify important features in images and extract discriminative high-dimensional data, which are then used to calculate the displacement error of the robot relative to the target position. The importance of this step lies in that it not only improves the efficiency and accuracy of feature extraction, but also reduces dependence on human intervention, allowing the robot to quickly obtain the required information in complex and dynamic environments, thereby updating its position information in real time.

[0037] S3, compare the displacement error with the distance threshold, if ||E pos ||>D threshold , perform the calibration steps.

[0038] It should be noted that comparing the calculated displacement error with the set distance threshold and deciding whether to perform the correction step is a key judgment link in the error correction process. When the displacement error exceeds the distance threshold, it means that the current positioning information may have significant deviations and corrections must be made to ensure that the robot can correctly understand its actual situation relative to the target position. This comparison step can effectively prevent small errors from being over-corrected, thereby reducing system instability, while ensuring that corrective measures are taken in a timely manner when necessary to improve the overall reliability of positioning.

[0039] S4, the correction step is calculated by the following formula:

[0040] P corrected =P current +P scale ·E pos

[0041] Among them, P currected is the corrected position information, and then the corrected information is smoothed.

[0042] It should be noted that the calculation of the correction step is implemented through a specific formula, and its main purpose is to systematically correct the positioning error to obtain the corrected position information. Through the application of this formula, the impact of the error on the robot's motion can be effectively reduced to ensure the accuracy of its movement trajectory. The corrected information also needs to be smoothed to reduce the mutations introduced by the correction and improve the continuity and consistency of the positioning data. Such smoothing not only enhances the stability of the system, but also provides a more reliable foundation for subsequent positioning and navigation, which helps the robot to perform more accurately and efficiently when performing tasks in complex environments.

[0043] In a possible implementation manner, a specific method for calculating the displacement error includes:

[0044] Get the target position P target =(x target ,y target ,z target ) and current location information y current =(x current ,y current ,z current );

[0045] The displacement error is calculated by the following formula:

[0046]

[0047] Among them, E pos is the displacement error.

[0048] It should be noted that the first step to obtain the target position and current position information is to ensure a clear reference. Only with this information can the system clearly determine the relationship between the actual position of the robot and the expected position, thereby providing the necessary data support for subsequent error calculations. This step lays the foundation for the evaluation of positioning accuracy and ensures that the system can accurately identify potential positioning problems.

[0049] The purpose of calculating the displacement error through a specific formula is to quantify the position deviation. This quantification method gives the error a clear numerical representation, which is easy to analyze and process. The calculated displacement error not only reflects the accuracy of positioning, but also provides a key basis for subsequent correction steps. This design enhances the feedback capability of the system, enabling the robot to quickly respond to positioning errors in complex environments and improve overall navigation performance and reliability.

[0050] In a possible implementation, the deep learning model further includes:

[0051] Set the training round T during training epoch , the calculation method is as follows:

[0052]

[0053] Among them, η is an empirical coefficient ranging from 1 to 5, which is used to dynamically adjust the training rounds to enhance the learning ability of the model.

[0054] It should be noted that the design of the deep learning model further includes setting the training rounds, which is set to optimize the learning effect of the model. By reasonably adjusting the training rounds, it is possible to avoid overfitting while ensuring that the model fully learns the data features. This dynamic adjustment capability enables the model to respond flexibly according to actual conditions and improve its adaptability and generalization ability.

[0055] The empirical coefficient is introduced as a tuning parameter in the calculation method to provide a quantitative way to optimize the training process. The range of the empirical coefficient is set to 1 to 5, which ensures the flexibility of training and allows appropriate adjustments in tasks of different complexity. In this way, the model can intelligently adjust the learning rounds according to the characteristics of the training data, thereby enhancing the model's ability to learn data and improving its performance in practical applications. This design not only improves training efficiency, but also helps to ultimately generate a more robust and efficient deep learning model, providing strong support for the robot's positioning tasks in complex environments.

[0056] In a possible implementation, it includes:

[0057] The loss function of the convolutional neural network in the group connection process is used to optimize the model parameters to minimize the difference between the predicted value and the true value. The loss function is as follows:

[0058]

[0059] Among them, y i is the true value, is the predicted value, N is the total number of samples, λ is the regularization coefficient, and the calculation method is as follows:

[0060]

[0061] The regularization coefficient is used to prevent overfitting through regularization.

[0062] It should be noted that the convolutional neural network loss function is used in the group connection process to optimize the model parameters and minimize the difference between the predicted value and the true value. This design is key because the effectiveness of the loss function directly affects the training effect of the model. By quantifying the deviation between the predicted result and the actual value, the loss function provides a clear optimization direction for the model, enabling it to adjust parameters more accurately, thereby improving the accuracy of the prediction.

[0063] The formula of the loss function involves multiple parameters such as the true value, predicted value, total number of samples, and regularization coefficient. The introduction of these parameters makes the loss calculation more comprehensive and can effectively reflect the performance of the model under different conditions. At the same time, the setting of the regularization coefficient is intended to prevent the model from overfitting and ensure that the model can maintain a high generalization ability on both training data and unseen data. The regularization calculation method provides additional constraints for the optimization process, helping the model to maintain balance when learning on complex data sets and avoid overfitting on specific training sets, thereby improving the reliability and stability of the model in practical applications.

[0064] In a possible implementation manner, the smoothing process specifically includes:

[0065] Smoothing is performed using the following smoothing formula:

[0066] S smoothed =α·P corrected +(1-α)·P previous

[0067] Where α is a smoothing factor, ranging from (0,1), and is defined as:

[0068]

[0069] Where β is the adjustment speed parameter, ranging from 1 to 10.

[0070] It should be noted that the specific implementation of smoothing includes smoothing the data through a specific smoothing formula. The purpose of this processing step is to reduce the mutations introduced by the correction operation and improve the continuity and consistency of the data. By applying the smoothing formula, short-term fluctuations can be effectively eliminated, ensuring that the final output position information is more stable, thereby improving the reliability of the system in actual operation.

[0071] The setting of the smoothing factor is the key to this step, and the definition of the range allows the system to flexibly adjust the degree of smoothing as needed. The specific definition and calculation method ensure that the smoothing factor can dynamically adapt to different application scenarios and data characteristics. In addition, the introduction of the adjustment speed parameter, with a range of 1 to 10, allows users to fine-tune the response speed of the smoothing process according to specific needs. With this design, the system can quickly adapt to real-time changing data inputs while maintaining the smoothing effect, further enhancing the accuracy and usability of the positioning results. This comprehensive processing method makes the final positioning results more operational, laying a solid foundation for the application of robots in complex environments.

[0072] In a possible implementation manner, the dynamic adjustment process of the smoothing factor specifically includes:

[0073] Calculate the current displacement error ||E pos||, the smoothing factor α is updated using the following formula:

[0074]

[0075] Among them, k is the learning rate, ranging from 0.01 to 0.1.

[0076] It should be noted that the dynamic adjustment process of the smoothing factor is performed after calculating the current displacement error. This step is crucial because it ensures that the smoothing factor can reflect the latest error in real time. By updating the smoothing factor, the system can dynamically adjust the smoothness of the data according to the current displacement error, thereby more effectively responding to environmental changes and error fluctuations. This adaptability improves the stability of the positioning data, allowing the robot to better perform tasks in complex environments.

[0077] When using a specific formula to update the smoothing factor, the learning rate is introduced to control the sensitivity of the adjustment. The learning rate range is set to 0.01 to 0.1, which ensures that during the smoothing factor adjustment process, it can not only respond quickly to large error changes, but also avoid unnecessary fluctuations caused by over-adjustment. This flexible design makes the smoothing factor adjustment process not only real-time, but also balances the reaction speed and stability, thereby optimizing the final positioning results and improving the overall performance and reliability of the system.

[0078] In a possible implementation manner, the target position obtaining method specifically includes:

[0079] By fusing multi-sensor information, we can obtain:

[0080] Collecting data from a variety of sensors, including vision sensors, lidar, and inertial measurement units;

[0081] The sensor data is fused using the weighted average method, with the weight W sensor The calculation method is as follows:

[0082]

[0083] Among them, S j is the signal-to-noise ratio of the jth sensor, and M is the number of sensors.

[0084] It should be noted that the method of obtaining the target position is achieved by fusing multi-sensor information. This process is crucial because relying on data from a single sensor may lead to inaccurate or unstable positioning. Therefore, combining data from multiple sensors can significantly improve the accuracy and reliability of positioning. These sensors include visual sensors, lidars, and inertial measurement units, each of which provides information in different dimensions and complements each other, allowing the system to fully understand the state of the environment.

[0085] The design of using weighted average method to fuse sensor data aims to integrate the advantages and disadvantages of each sensor to improve the credibility of the data. The weight calculation method takes the signal-to-noise ratio into consideration to ensure that sensors with high signal-to-noise ratio contribute more to the final result, thereby improving the accuracy of overall positioning. As an important indicator for evaluating sensor performance, signal-to-noise ratio enables the system to intelligently select more reliable information sources when processing data. This weighted fusion mechanism not only enhances the accuracy of target position acquisition, but also improves the system's ability to adapt to environmental changes, providing a solid data foundation for subsequent error correction.

[0086] In a possible implementation manner, the fusion calculation of the target position specifically includes:

[0087] For each sensor j, calculate the effectiveness factor E j , the formula is:

[0088]

[0089] Among them, S j is the signal-to-noise ratio of the jth sensor, D j is the jth measurement error of sensor, D max The maximum allowed measurement error ensures that the validity factor reflects the relationship between signal quality and measurement reliability;

[0090] The target position is calculated using the following weighted fusion formula:

[0091]

[0092] Among them, E j is the target position measured by the jth sensor.

[0093] It should be noted that the fusion calculation of the target position is achieved by calculating the effectiveness factor for each sensor j. The importance of this process lies in that the effectiveness factor can reflect the relationship between the quality of the sensor signal and the reliability of the measurement result. The signal-to-noise ratio, measurement error and maximum allowable measurement error of the jth sensor are introduced into the formula to ensure that the effectiveness factor not only takes into account the performance of the sensor, but also takes into account the uncertainty of the measurement. This design can help the system more accurately evaluate the contribution of each sensor in the overall positioning process.

[0094] After calculating the validity factor, the weighted fusion formula is used to calculate the target position in order to integrate the data from different sensors and improve the positioning accuracy. In this way, the system can make full use of the information from each sensor in the process of determining the final target position, while reducing the positioning inaccuracy caused by the error of individual sensors. This weighted fusion method makes the calculation result of the target position more stable and reliable, providing a solid foundation for the robot's positioning in complex environments, thereby improving its overall navigation and operation performance.

[0095] In a possible implementation manner, the dynamic adjustment process of the target position specifically includes:

[0096] Calculate the fused target position P target After that, a dynamic feedback mechanism is set to adjust the priority of the target position in real time according to the current motion state of the robot, and the priority adjustment parameter P is used. priority , and its calculation formula is:

[0097]

[0098] Among them, γ is the adjustment speed, δ is the reference error value;

[0099] The final position correction is performed using the following formula based on the target position after priority adjustment:

[0100] P final =P target +P priority ·( corrected -P target )

[0101] Among them, P final is the final adjusted target position.

[0102] It should be noted that the dynamic adjustment process of the target position is achieved by setting a dynamic feedback mechanism, which is designed to adjust the priority of the target position in real time according to the current motion state of the robot. In this way, the system can respond quickly to changes in the environment and improve the flexibility and accuracy of navigation. The calculation formula of the priority adjustment parameter introduces the adjustment speed and the reference error value, ensuring that the adjustment of the target position can be more timely and reasonable under different motion conditions.

[0103] The design of the dynamic feedback mechanism enables the system to make decisions based on real-time information instead of relying solely on static data. This real-time nature is the key to achieving efficient positioning. Through dynamic adjustment of priorities, the system can effectively balance the importance of different target locations in complex environments, ensuring that the robot can prioritize more urgent positioning needs when performing tasks.

[0104] Finally, according to the target position after priority adjustment, the system uses a specific formula to correct the position, thereby obtaining the final adjusted target position. This process not only improves the accuracy of positioning, but also enhances the robot's adaptability in dynamic environments, ensuring that it can complete various complex tasks more smoothly. This comprehensive adjustment mechanism provides the robot with more efficient navigation capabilities, helping to improve the safety and reliability of its operations.

[0105] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0106] (1) In the present invention, the feature extraction capability of the deep learning model significantly improves the robot's recognition accuracy of displacement errors in complex environments. This innovation enables the robot to more accurately locate its position relative to the target, reducing the impact of sensor noise and environmental changes, thereby improving the reliability and stability of overall positioning. This high-precision error correction capability helps the robot significantly reduce potential safety risks when performing autonomous navigation and tasks;

[0107] (2) In the present invention, the introduction of a dynamic correction mechanism enables the robot to respond to environmental changes in real time. When the displacement error is detected to exceed the set distance threshold, the system can quickly execute the correction step, thereby effectively reducing the positioning error. This real-time adjustment capability makes the robot more flexible in complex and dynamic working environments, able to adapt to rapidly changing conditions, and improve work efficiency;

[0108] (3) In the present invention, the method of fusing multi-sensor data effectively enhances the integrity and redundancy of information. By combining the data of visual sensors, lidar and inertial measurement units, the system can maintain high positioning accuracy even when information is lost or a single sensor fails. This redundant design improves the robustness of the system and ensures that the robot can operate stably under a variety of working conditions, thereby broadening its application range.

[0109] Reference Manual Attached Figure 2 , shows a structural schematic diagram of a deep learning robot 3D vision positioning error correction system provided by an embodiment of the present invention.

[0110] The present invention also provides a deep learning robot 3D visual positioning error correction system, which is applied to the deep learning robot 3D visual positioning error correction method, comprising:

[0111] processor;

[0112] A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the robot 3D visual positioning error correction method based on deep learning as described in the method embodiment is implemented.

[0113] The deep learning robot 3D visual positioning error correction system provided by the present invention can execute the deep learning robot 3D visual positioning error correction method described above and achieve the same or similar technical effects. To avoid repetition, the present invention will not go into details.

[0114] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0115] (1) In the present invention, the feature extraction capability of the deep learning model significantly improves the robot's recognition accuracy of displacement errors in complex environments. This innovation enables the robot to more accurately locate its position relative to the target, reducing the impact of sensor noise and environmental changes, thereby improving the reliability and stability of overall positioning. This high-precision error correction capability helps the robot significantly reduce potential safety risks when performing autonomous navigation and tasks;

[0116] (2) In the present invention, the introduction of a dynamic correction mechanism enables the robot to respond to environmental changes in real time. When the displacement error is detected to exceed the set distance threshold, the system can quickly execute the correction step, thereby effectively reducing the positioning error. This real-time adjustment capability makes the robot more flexible in complex and dynamic working environments, able to adapt to rapidly changing conditions, and improve work efficiency;

[0117] (3) In the present invention, the method of fusing multi-sensor data effectively enhances the integrity and redundancy of information. By combining the data of visual sensors, lidar and inertial measurement units, the system can maintain high positioning accuracy even when information is lost or a single sensor fails. This redundant design improves the robustness of the system and ensures that the robot can operate stably under a variety of working conditions, thereby broadening its application range.

[0118] The above contents are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art who is familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

[0119] There are a few points to note:

[0120] (1) The drawings of the embodiments of the present invention only relate to the structures related to the embodiments of the present invention, and other structures may refer to the general design.

[0121] (2) For the sake of clarity, in the drawings used to describe the embodiments of the present invention, the thickness of the layers or regions is exaggerated or reduced, that is, these drawings are not drawn according to the actual scale. It is understood that when an element such as a layer, film, region or substrate is referred to as being "on" or "under" another element, the element may be "directly" "on" or "under" the other element or there may be intermediate elements.

[0122] (3) In the absence of conflict, the embodiments of the present invention and the features therein may be combined with each other to obtain new embodiments.

[0123] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. The protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A deep learning robot 3D visual positioning error correction method, characterized in that: include: S1. Obtain the initial position information P of the robot in three-dimensional space initial , and define the scaling factor R scale and distance threshold D threshold , the scaling factor has a value range of (0, 2), indicating the amplification or reduction of the error, and the distance threshold is in meters, indicating the maximum acceptable positioning error; S2. Use the deep learning model to extract features from the acquired image data and calculate the displacement error E of the robot relative to the target position. pos =P target -P current , where P target is the target position, P current is the current location information; S3, compare the displacement error with the distance threshold, if ||E pos ||>D threshold , perform a calibration step; S4, the correction step is calculated by the following formula: P corrected =P current +R scale ·E pos Among them, P corrected is the corrected position information, and then the corrected information is smoothed.

2. The method for correcting robot 3D visual positioning errors based on deep learning according to claim 1, characterized in that: The specific calculation method of the displacement error includes: Get the target position P target =(x target ,y target , z target ) and current location information P current =(x current ,y current , z current ); The displacement error is calculated by the following formula: Among them, E pos is the displacement error.

3. The method for correcting robot 3D visual positioning errors based on deep learning according to claim 1, characterized in that: The deep learning model further comprises: Set the training round T during training epoch , the calculation method is as follows: Among them, η is an empirical coefficient ranging from 1 to 5, which is used to dynamically adjust the training rounds to enhance the learning ability of the model.

4. The method for correcting robot 3D visual positioning errors based on deep learning according to claim 3, characterized in that: include: The loss function of the convolutional neural network in the group connection process is used to optimize the model parameters to minimize the difference between the predicted value and the true value. The loss function is as follows: Among them, y i is the true value, is the predicted value, N is the total number of samples, λ is the regularization coefficient, and the calculation method is as follows: The regularization coefficient is used to prevent overfitting through regularization.

5. The method for correcting robot 3D visual positioning errors based on deep learning according to claim 1, characterized in that: The smoothing process specifically includes: Smoothing is performed using the following smoothing formula: S smoothed =α·P corrected +(1-a)·P previous Among them, α is the smoothing factor, the range is (0, 1), and it is defined as: Where β is the adjustment speed parameter, ranging from 1 to 10.

6. The method for correcting robot 3D visual positioning errors based on deep learning according to claim 5, characterized in that: The dynamic adjustment process of the smoothing factor specifically includes: Calculate the current displacement error ||E pos ||, the smoothing factor α is updated using the following formula: Among them, k is the learning rate, ranging from 0.01 to 0.

1.

7. The method for correcting robot 3D visual positioning errors based on deep learning according to claim 1, characterized in that: The target position obtaining method specifically includes: By fusing multi-sensor information, we can obtain: Collecting data from a variety of sensors, including vision sensors, lidar, and inertial measurement units; The sensor data is fused using the weighted average method, with the weight W sensor The calculation method is as follows: Among them, S j is the signal-to-noise ratio of the jth sensor, and M is the number of sensors.

8. The method for correcting robot 3D visual positioning errors based on deep learning according to claim 7, characterized in that: The fusion calculation of the target position specifically includes: For each sensor j, calculate the effectiveness factor E j , the formula is: Among them, S j is the signal-to-noise ratio of the jth sensor, D j is the jth measurement error of sensor, D max The maximum allowed measurement error ensures that the validity factor reflects the relationship between signal quality and measurement reliability; The target position is calculated using the following weighted fusion formula: Among them, E j is the target position measured by the th sensor.

9. The method for correcting robot 3D visual positioning errors based on deep learning according to claim 8, characterized in that: The dynamic adjustment process of the target position specifically includes: Calculate the fused target position P target After that, a dynamic feedback mechanism is set to adjust the priority of the target position in real time according to the current motion state of the robot, and the priority adjustment parameter P is used. priority， The calculation formula is: Among them, γ is the adjustment speed, δ is the reference error value; The final position correction is performed using the following formula based on the target position after priority adjustment: P final =P target +P priority ·(P corrected -P target ) Among them, P final is the final adjusted target position.

10. A deep learning robot 3D visual positioning error correction system, characterized in that: include: processor; A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the robot 3D visual positioning error correction method based on deep learning as described in any one of claims 1 to 9 is implemented.

Images