Industrial robot optical navigation anti-occlusion tracking systems, methods, devices, and media

Abstract

The application discloses an industrial robot optical navigation anti-shielding tracking system, method, device and medium, wherein the system comprises a perception layer, a calculation layer and an execution layer, wherein the perception layer comprises a mixed vision network composed of two active pan-tilt cameras and four fixed cameras, and is used for acquiring a global workpiece image of a workpiece, wherein a mark point is arranged on the workpiece; the calculation layer comprises a pose solving module, a motion prediction module and a fusion decision module, is used for receiving image data of the perception layer and processing the image data to obtain pose data; and the execution layer comprises an industrial robot, is used for generating a control instruction based on the pose data and transmitting the control instruction to a control end of the industrial robot. Through dynamic viewpoint optimization, normal direction constraint modeling and hierarchical motion prediction algorithm, the application significantly improves tracking accuracy and real-time performance of the industrial robot under complex working conditions, reduces cost and power consumption, and has wide application prospect.

Description

Technical Field

[0001] This invention relates to the field of visual navigation for industrial robots, and more specifically, to an anti-occlusion tracking system, method, apparatus, and medium for optical navigation of industrial robots. Background Technology

[0002] In the field of intelligent manufacturing, real-time and accurate tracking of workpieces by industrial robots is a key technology for achieving automated assembly, welding, and other processes. Traditional optical navigation systems face three major technical bottlenecks:

[0003] 1. Limited Field of View: Commercial optical tracking systems (such as NDI Polaris) typically have a measurement volume (MV) of no more than 2 m³. When the workpiece moves beyond MV, tracking will be interrupted. Studies have shown that in automotive welding scenarios, the tracking failure rate due to workpiece movement is as high as 32%.

[0004] 2. Dynamic occlusion problem: Temporary occlusion caused by the robot itself, lifting equipment, etc., is common in industrial settings. While existing multi-camera solutions expand the field of view, they do not resolve the data conflict problem between multiple cameras, resulting in a 3-5 times increase in positioning error under 50% occlusion.

[0005] 3. Attitude Sensitivity: When the angle between the workpiece normal and the camera optical axis exceeds 60°, the success rate of marker recognition drops below 65%. Existing technologies only consider positional constraints, neglecting attitude constraints, leading to tracking failure during rapid rotation.

[0006] Existing solutions have significant shortcomings. Fixed multi-camera layouts cannot adapt to dynamic working conditions. The active navigation solution with cameras mounted on the robotic arm does not establish a normal direction constraint model, resulting in an error of ±2.3mm under workpiece flipping conditions. Summary of the Invention

[0007] The purpose of this invention is to provide an anti-occlusion tracking system, method, device, and medium for industrial robots, which is particularly suitable for real-time high-precision tracking of workpieces under dynamic occlusion and complex motion conditions.

[0008] The first aspect of this invention provides an optical navigation anti-occlusion tracking system for industrial robots, comprising:

[0009] The layers are: perception layer, computation layer, and execution layer.

[0010] The perception layer includes a hybrid vision network consisting of two active gimbal cameras and four fixed cameras, used to acquire a global image of the workpiece, wherein the workpiece is marked with points.

[0011] The computation layer includes a pose calculation module, a motion prediction module, and a fusion decision module, which are used to receive image data from the perception layer and process it to obtain pose data.

[0012] The execution layer includes an industrial robot, used to generate control commands based on the pose data and transmit them to the control terminal of the industrial robot.

[0013] In this solution, the pose calculation module is used to identify marker points in the workpiece image; the motion prediction module is used to correct the workpiece motion position; and the fusion decision module is used to detect and eliminate abnormal poses through Mahalanobis distance and calculate the pose data by fusing multi-camera data.

[0014] A second aspect of the present invention provides an anti-occlusion tracking method for optical navigation of an industrial robot, applied to any of the aforementioned anti-occlusion tracking systems for optical navigation of an industrial robot, comprising the following steps:

[0015] The target image is obtained by preprocessing data from multiple cameras, including two active gimbal cameras and four fixed cameras.

[0016] Pose calculation, motion prediction, and adaptive weighted fusion are performed on the target image to obtain pose data;

[0017] Based on the pose data, target angles of each joint of the robot are calculated to generate control commands, thereby controlling the robot's joint movements based on the control commands.

[0018] In this scheme, the target image is subjected to pose calculation and motion prediction, specifically including:

[0019] Extract the target image, which includes a distortion-free image and 2D pixel coordinates;

[0020] The 6D pose of the workpiece is calculated based on 2D pixel coordinates and preset 3D model coordinates.

[0021] Based on the 6D pose of the workpiece, motion prediction is performed to correct the workpiece's motion position and obtain the predicted pose, specifically through iterative correction using Newton's method.

[0022] In this scheme, the pose data is obtained through adaptive weighted fusion, specifically including:

[0023] Based on the 6D pose and the predicted pose, calculate the visibility score of each camera for the workpiece;

[0024] The pose data is obtained by adjusting the multi-camera viewpoints based on the visibility score and performing adaptive weighted fusion calculation based on the confidence scores of different cameras.

[0025] A third aspect of the present invention also provides an anti-occlusion tracking device for industrial robot optical navigation, comprising a memory and a processor. The memory includes a program for an industrial robot optical navigation anti-occlusion tracking method. When the processor executes the program for the industrial robot optical navigation anti-occlusion tracking method, it performs the following steps:

[0026] The target image is obtained by preprocessing data from multiple cameras, including two active gimbal cameras and four fixed cameras.

[0027] Pose calculation, motion prediction, and adaptive weighted fusion are performed on the target image to obtain pose data;

[0028] Based on the pose data, target angles of each joint of the robot are calculated to generate control commands, thereby controlling the robot's joint movements based on the control commands.

[0029] In this scheme, the target image is subjected to pose calculation and motion prediction, specifically including:

[0030] Extract the target image, which includes a distortion-free image and 2D pixel coordinates;

[0031] The 6D pose of the workpiece is calculated based on 2D pixel coordinates and preset 3D model coordinates.

[0032] Based on the 6D pose of the workpiece, motion prediction is performed to correct the workpiece's motion position and obtain the predicted pose, specifically through iterative correction using Newton's method.

[0033] In this scheme, the pose data is obtained through adaptive weighted fusion, specifically including:

[0034] Based on the 6D pose and the predicted pose, calculate the visibility score of each camera for the workpiece;

[0035] The pose data is obtained by adjusting the multi-camera viewpoints based on the visibility score and performing adaptive weighted fusion calculation based on the confidence scores of different cameras.

[0036] A fourth aspect of the present invention provides a computer-readable storage medium comprising a program for an industrial robot optical navigation anti-occlusion tracking method, wherein when executed by a processor, the program implements the steps of the industrial robot optical navigation anti-occlusion tracking method as described in any of the preceding claims.

[0037] This invention discloses an optical navigation anti-occlusion tracking system, method, device, and medium for industrial robots. Through dynamic viewpoint optimization, normal direction constraint modeling, and hierarchical motion prediction algorithms, it significantly improves the tracking accuracy and real-time performance of industrial robots under complex working conditions, while reducing cost and power consumption, and has broad application prospects. Attached Figure Description

[0038] Figure 1 A block diagram of an optical navigation anti-occlusion tracking system for industrial robots according to the present invention is shown.

[0039] Figure 2 A flowchart of an anti-occlusion tracking method for optical navigation of an industrial robot according to the present invention is shown;

[0040] Figure 3 This invention illustrates a schematic diagram of the normal direction constraint set model for an anti-occlusion tracking method for optical navigation of industrial robots according to the present invention.

[0041] Figure 4 A block diagram of an anti-occlusion tracking device for optical navigation of an industrial robot according to the present invention is shown. Detailed Implementation

[0042] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0043] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0044] Figure 1 A block diagram of an optical navigation anti-occlusion tracking system for industrial robots according to this application is shown.

[0045] like Figure 1 As shown, this application discloses an optical navigation anti-occlusion tracking system for industrial robots, comprising:

[0046] The layers are: perception layer, computation layer, and execution layer.

[0047] The perception layer includes a hybrid vision network consisting of two active gimbal cameras and four fixed cameras, used to acquire a global image of the workpiece, wherein the workpiece is marked with points.

[0048] The computation layer includes a pose calculation module, a motion prediction module, and a fusion decision module, which are used to receive image data from the perception layer and process it to obtain pose data.

[0049] The execution layer includes an industrial robot, used to generate control commands based on the pose data and transmit them to the control terminal of the industrial robot.

[0050] It should be noted that, in this embodiment, as Figure 1 As shown, the perception layer includes a gimbal camera and a fixed camera, specifically a hybrid vision network consisting of two active gimbal cameras and four fixed cameras, used to acquire a global image of the workpiece. The workpiece is marked with points. In one embodiment of the invention, the frame rate of the active gimbal camera is 200Hz and the focal length is 12mm, the frame rate of the fixed camera is 50Hz, and the coverage area of ​​the hybrid vision network includes 8 m³.

[0051] Furthermore, in this embodiment, the computing layer includes a pose calculation module, a motion prediction module, and a fusion decision module, which are used to receive image data from the perception layer and process it to obtain pose data. The pose calculation module is used to identify marker points in the workpiece image, the motion prediction module is used to correct the workpiece motion position, and the fusion decision module is used to detect and remove abnormal poses through Mahalanobis distance and calculate the pose data by fusing multi-camera data.

[0052] Furthermore, in this embodiment, the execution layer includes an industrial robot, used to generate control commands based on the pose data and transmit them to the control terminal of the industrial robot. As shown in Figure 1, the computing layer sends control commands to the robot control terminal via the EtherCAT bus, and the robot feeds back status information (such as joint angles and end-effector poses) to the computing layer, forming a closed-loop control.

[0053] Figure 2 A flowchart of an anti-occlusion tracking method for optical navigation of an industrial robot according to this application is shown.

[0054] like Figure 2 As shown, this application discloses an anti-occlusion tracking method for industrial robot optical navigation, applied to any of the industrial robot optical navigation anti-occlusion tracking systems described in this application. The method includes the following steps:

[0055] S202, collects data from multiple cameras and performs preprocessing to obtain the target image. The multiple cameras include two active gimbal cameras and four fixed cameras.

[0056] S204, Perform pose calculation, motion prediction and adaptive weighted fusion on the target image to obtain pose data;

[0057] S206, based on the pose data, calculate the target angles of each joint of the robot and generate control commands, thereby controlling the robot joint movements based on the control commands.

[0058] It should be noted that, in this embodiment, the workpiece is marked with markers. Therefore, during the workpiece's movement, multi-camera data is collected and preprocessed to obtain the target image. Kalman filtering is used for image preprocessing. The multi-camera setup includes two active gimbal cameras and four fixed cameras. Position and attitude constraints are defined to ensure that the visibility of the markers on the workpiece is improved to 98% in any attitude. Specifically, as shown... Figure 3 As shown, the formula for calculating the position constraint is as follows:

[0059] ;

[0060] in, The coordinates of the workpiece marking points. To determine the equation of the j-th boundary plane of the measurement volume (MV), Let be the radius of the envelope of the marked point.

[0061] The formula for calculating attitude constraints is as follows:

[0062] ;

[0063] in, Let be the direction vector of the camera's optical axis. The normal vector of the marker point.

[0064] Furthermore, pose calculation, motion prediction, and adaptive weighted fusion are performed on the target image to obtain pose data, which will be described in detail in the subsequent specification. Then, based on the pose data, the target angles of each joint of the robot are calculated to generate control commands, thereby controlling the robot joint movements based on the control commands. Real-time control communication is performed based on the EtherCAT bus, and the repeatability is ±0.03mm.

[0065] According to an embodiment of the present invention, the pose calculation and motion prediction of the target image specifically include:

[0066] Extract the target image, which includes a distortion-free image and 2D pixel coordinates;

[0067] The 6D pose of the workpiece is calculated based on 2D pixel coordinates and preset 3D model coordinates.

[0068] Based on the 6D pose of the workpiece, motion prediction is performed to correct the workpiece's motion position and obtain the predicted pose, specifically through iterative correction using Newton's method.

[0069] It should be noted that, in this embodiment, the target image is extracted, which includes a distorted image and 2D pixel coordinates. The 6D pose of the workpiece is calculated based on the 2D pixel coordinates and a preset 3D model coordinate. Then, motion prediction is performed based on the 6D pose of the workpiece to correct the workpiece's motion position and obtain the predicted pose. Specifically, KF-Newton composite prediction is used.

[0070] ;

[0071] in, Here is the state transition matrix. For state vectors, To control the input matrix, To control the input vector, For process noise, This is the state correction amount. The Hessian matrix converges after three iterations, where the state vector... Used to describe the pose, velocity, and other states of the system at time k, and as a control input matrix. Used to map control inputs to the state space, control input vector This corresponds to external control signals, such as robot joint commands and process noise. State corrections used to describe uncertainties in a system model. The increment used for iterative optimization of pose estimation is further used to construct the QP problem:

[0072] ;

[0073] in, This is the weight matrix. This indicates that the optimization is performed by minimizing the variable. This represents the velocity increment, typically used for velocity or position corrections in motion prediction. It is the speed increment transpose, is a positive definite weight matrix used to weight the velocity increments in different dimensions, controlling the importance of each component during the optimization process; c is the constraint matrix. G is the transpose of constraint matrix c, G is the upper bound vector of constraints corresponding to constraint matrix c, and h is the velocity increment. The upper bound of the absolute value constraint is used to limit the velocity increment. The maximum range of variation, where the constraints include the limits of the robotic arm joints, are used to correct the workpiece's motion position and obtain the predicted pose.

[0074] According to an embodiment of the present invention, adaptive weighted fusion is used to obtain pose data, specifically including:

[0075] Based on the 6D pose and the predicted pose, calculate the visibility score of each camera for the workpiece;

[0076] The pose data is obtained by adjusting the multi-camera viewpoints based on the visibility score and performing adaptive weighted fusion calculation based on the confidence scores of different cameras.

[0077] It should be noted that, in this embodiment, the visibility score of each camera on the workpiece is calculated. The visibility score includes scores based on distance, normal direction, and occlusion. The multi-camera viewpoint is adjusted based on the visibility score, and the optimal viewpoint camera is selected to control the rotation of the gimbal to ensure that at least one camera is always visible. If multiple cameras are visible, the pose data is obtained by adaptive weighted fusion calculation based on the confidence of different cameras. The confidence weight is based on the premise of eliminating abnormal poses based on Mahalanobis distance.

[0078] Figure 4 A block diagram of an anti-occlusion tracking system for optical navigation of an industrial robot according to the present invention is shown.

[0079] like Figure 4 As shown, this invention discloses an industrial robot optical navigation anti-occlusion tracking system 40, including a memory 41 and a processor 42. The memory 41 includes an industrial robot optical navigation anti-occlusion tracking method program. When the industrial robot optical navigation anti-occlusion tracking method program is executed by the processor 42, it performs the following steps:

[0080] The target image is obtained by preprocessing data from multiple cameras, including two active gimbal cameras and four fixed cameras.

[0081] Pose calculation, motion prediction, and adaptive weighted fusion are performed on the target image to obtain pose data;

[0082] Based on the pose data, target angles of each joint of the robot are calculated to generate control commands, thereby controlling the robot's joint movements based on the control commands.

[0083] It should be noted that, in this embodiment, the workpiece is marked with markers. Therefore, during the workpiece's movement, multi-camera data is collected and preprocessed to obtain the target image. Kalman filtering is used for image preprocessing. The multi-camera setup includes two active gimbal cameras and four fixed cameras. Position and attitude constraints are defined to ensure that the visibility of the markers on the workpiece is improved to 98% in any attitude. Specifically, as shown... Figure 3 As shown, the formula for calculating the position constraint is as follows:

[0084] ;

[0085] in, The coordinates of the workpiece marking points. To determine the equation of the j-th boundary plane of the measurement volume (MV), Let be the radius of the envelope of the marked point.

[0086] The formula for calculating attitude constraints is as follows:

[0087] ;

[0088] in, Let be the direction vector of the camera's optical axis. The normal vector of the marker point.

[0089] Furthermore, pose calculation, motion prediction, and adaptive weighted fusion are performed on the target image to obtain pose data, which will be described in detail in the subsequent specification. Then, based on the pose data, the target angles of each joint of the robot are calculated to generate control commands, thereby controlling the robot joint movements based on the control commands. Real-time control communication is performed based on the EtherCAT bus, and the repeatability is ±0.03mm.

[0090] According to an embodiment of the present invention, the pose calculation and motion prediction of the target image specifically include:

[0091] Extract the target image, which includes a distortion-free image and 2D pixel coordinates;

[0092] The 6D pose of the workpiece is calculated based on 2D pixel coordinates and preset 3D model coordinates.

[0093] Based on the 6D pose of the workpiece, motion prediction is performed to correct the workpiece's motion position and obtain the predicted pose, specifically through iterative correction using Newton's method.

[0094] It should be noted that, in this embodiment, the target image is extracted, which includes a distorted image and 2D pixel coordinates. The 6D pose of the workpiece is calculated based on the 2D pixel coordinates and a preset 3D model coordinate. Then, motion prediction is performed based on the 6D pose of the workpiece to correct the workpiece's motion position and obtain the predicted pose. Specifically, KF-Newton composite prediction is used.

[0095] ;

[0096] in, Here is the state transition matrix. For state vectors, To control the input matrix, To control the input vector, For process noise, State correction amount The Hessian matrix converges after three iterations, where the state vector... Used to describe the pose, velocity, and other states of the system at time k, and as a control input matrix. Used to map control inputs to the state space, control input vector This corresponds to external control signals, such as robot joint commands and process noise. State corrections used to describe uncertainties in a system model. The increment used for iterative optimization of pose estimation is further used to construct the QP problem:

[0097] ;

[0098] in, This is the weight matrix. This indicates that the optimization is performed by minimizing the variable. This represents the velocity increment, typically used for velocity or position corrections in motion prediction. It is the speed increment transpose, is a positive definite weight matrix used to weight the velocity increments in different dimensions, controlling the importance of each component during the optimization process; c is the constraint matrix. G is the transpose of constraint matrix c, G is the upper bound vector of constraints corresponding to constraint matrix c, and h is the velocity increment. The upper bound of the absolute value constraint is used to limit the velocity increment. The maximum range of variation is determined, where constraints include the limits of the robotic arm joints. This is used to correct the workpiece's motion position and obtain the predicted pose.

[0099] According to an embodiment of the present invention, adaptive weighted fusion is used to obtain pose data, specifically including:

[0100] Based on the 6D pose and the predicted pose, calculate the visibility score of each camera for the workpiece;

[0101] The pose data is obtained by adjusting the multi-camera viewpoints based on the visibility score and performing adaptive weighted fusion calculation based on the confidence scores of different cameras.

[0102] It should be noted that, in this embodiment, the visibility score of each camera on the workpiece is calculated. The visibility score includes scores based on distance, normal direction, and occlusion. The multi-camera viewpoint is adjusted based on the visibility score, and the optimal viewpoint camera is selected to control the rotation of the gimbal to ensure that at least one camera is always visible. If multiple cameras are visible, the pose data is obtained by adaptive weighted fusion calculation based on the confidence of different cameras. The confidence weight is based on the premise of eliminating abnormal poses based on Mahalanobis distance.

[0103] The fourth aspect of the present invention provides a computer-readable storage medium including a program for an industrial robot optical navigation anti-occlusion tracking method. When the program is executed by a processor, it implements the steps of the industrial robot optical navigation anti-occlusion tracking method as described in any of the preceding claims.

[0104] This invention discloses an anti-occlusion tracking system, method, device, and medium for optical navigation of industrial robots. Its core innovations include: a normal direction constraint model: establishing dual constraints including position and attitude, defining a composite optimization objective function of the normal direction angle α and the distance angle β, improving the visibility of marker points on the workpiece in any attitude to 98%; a hierarchical motion prediction algorithm: combining Kalman filtering (KF) and Newton's iteration method, reducing the 6D pose prediction delay from 120ms to 18ms and increasing the speed tolerance to 2.5m / s; and an adaptive field-of-view fusion strategy: through Mahalanobis distance testing and confidence weighting, maintaining a positioning accuracy of 0.8mm even with 50% occlusion (compared to 2.1mm using traditional methods).

[0105] In the automotive door frame welding scenario, an anti-occlusion test was conducted. The working condition involved two moving workpieces with the robot body obstructing the view. The test results showed that the traditional method had an error of 2.2 mm and a failure rate of 30% when obstructed, while the maximum error of this application was 0.7 mm and the failure rate was 3.4%. The experiment showed that the tracking failure rate of this system in the automotive welding scenario was reduced to 3.2%, which is 10 times better than the traditional solution; power consumption was reduced by 40%, and the cost was only 1 / 4 of that of commercial systems.

[0106] Specifically, this application significantly improves the tracking accuracy and real-time performance of industrial robots under complex working conditions through dynamic viewpoint optimization, normal direction constraint modeling, and hierarchical motion prediction algorithms, while reducing costs and power consumption, and has broad application prospects.

[0107] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0108] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0109] In addition, in the various embodiments of the present invention, each functional unit can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0110] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0111] Alternatively, if the integrated units of this invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

Claims

1. An optical navigation anti-occlusion tracking method for industrial robots, characterized in that, This is applied to an optical navigation anti-occlusion tracking system for industrial robots, the system comprising: The layers are: perception layer, computation layer, and execution layer. The perception layer includes a hybrid vision network consisting of two active gimbal cameras and four fixed cameras, used to acquire a global image of the workpiece, wherein the workpiece is marked with points. The computation layer includes a pose calculation module, a motion prediction module, and a fusion decision module, which are used to receive image data from the perception layer and process it to obtain pose data. The execution layer includes an industrial robot, used to generate control commands based on the pose data and transmit them to the control terminal of the industrial robot. The method includes the following steps: The target image is obtained by preprocessing data from multiple cameras, including two active gimbal cameras and four fixed cameras. Pose calculation, motion prediction, and adaptive weighted fusion are performed on the target image to obtain pose data; Based on the pose data, the target angles of each joint of the robot are calculated to generate control commands, thereby controlling the joint movements of the robot based on the control commands. Performing pose calculation and motion prediction on the target image specifically includes: Extract the target image, which includes a distortion-free image and 2D pixel coordinates; The 6D pose of the workpiece is calculated based on 2D pixel coordinates and preset 3D model coordinates. Based on the 6D pose of the workpiece, motion prediction is performed to correct the workpiece's motion position and obtain the predicted pose, specifically through iterative correction using Newton's method. Adaptive weighted fusion yields pose data, specifically including: Based on the 6D pose and the predicted pose, calculate the visibility score of each camera for the workpiece; The visibility score includes scores based on distance, normal direction, and occlusion. The multi-camera viewpoints are adjusted based on the visibility score, and the optimal viewpoint camera is selected to control the gimbal rotation to ensure that at least one camera is always visible. If multiple cameras are visible, the pose data is obtained by adaptive weighted fusion calculation based on the confidence of different cameras.

2. The anti-occlusion tracking method for optical navigation of industrial robots according to claim 1, characterized in that, The pose calculation module is used to identify marker points in the workpiece image; the motion prediction module is used to correct the workpiece motion position; the fusion decision module is used to eliminate abnormal poses by detecting Mahalanobis distance and to calculate the pose data by fusing multi-camera data.

3. An anti-occlusion tracking device for optical navigation of industrial robots, characterized in that, The system includes a memory and a processor. The memory contains a program for an industrial robot optical navigation anti-occlusion tracking method. When the processor executes the program, the industrial robot optical navigation anti-occlusion tracking method performs the following steps: The target image is obtained by preprocessing data from multiple cameras, including two active gimbal cameras and four fixed cameras. Pose calculation, motion prediction, and adaptive weighted fusion are performed on the target image to obtain pose data; Based on the pose data, target angles of each joint of the robot are calculated to generate control commands, thereby controlling the robot's joint movements based on the control commands; Performing pose calculation and motion prediction on the target image specifically includes: Extract the target image, which includes a distortion-free image and 2D pixel coordinates; The 6D pose of the workpiece is calculated based on 2D pixel coordinates and preset 3D model coordinates. Based on the 6D pose of the workpiece, motion prediction is performed to correct the workpiece's motion position and obtain the predicted pose. Specifically, this correction is achieved through iterative correction using Newton's method. Adaptive weighted fusion yields pose data, specifically including: Based on the 6D pose and the predicted pose, calculate the visibility score of each camera for the workpiece; The visibility score includes scores based on distance, normal direction, and occlusion. The multi-camera viewpoints are adjusted based on the visibility score, and the optimal viewpoint camera is selected to control the gimbal rotation to ensure that at least one camera is always visible. If multiple cameras are visible, the pose data is obtained by adaptive weighted fusion calculation based on the confidence of different cameras.

4. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a program for an industrial robot optical navigation anti-occlusion tracking method. When the program is executed by a processor, it implements the steps of the industrial robot optical navigation anti-occlusion tracking method as described in claim 1.

Images