A weld tracking method and system based on a point tracking model

Abstract

This invention relates to a weld seam tracking method and system based on a point tracking model, belonging to the field of intelligent robotic welding technology. The method includes: acquiring a global visual image and a line laser visual image; selecting welding key points in an initial frame; obtaining the position and occlusion probability of the welding key points in consecutive frames using a point tracking model; employing a welding environment enhancement strategy to enhance the input image with specular perturbation and spatial features; adaptively adjusting the observation noise and process noise in the state estimation model according to the occlusion probability to obtain a stable weld seam trajectory; fusing the tracking results of the global vision and local line laser vision, and obtaining the welding trajectory in the robot's base coordinate system through coordinate transformation, thereby achieving closed-loop guidance control of the welding robot. This invention transforms the traditional weld seam reconstruction problem into a welding key point tracking problem, reducing complex image processing procedures and improving the real-time performance and robustness of weld seam tracking in complex welding environments.

Description

Technical Field

[0001] This invention belongs to the field of robotic intelligent welding and machine vision, specifically relating to a weld seam tracking method and system based on a point tracking model. Background Technology

[0002] Weld seam tracking is one of the key technologies for achieving automated welding in intelligent welding robots.

[0003] Existing weld seam tracking methods typically rely on multiple steps, including laser stripe extraction, weld seam edge detection, centerline fitting, and weld seam trajectory reconstruction, to locate the weld point. These methods suffer from the following problems:

[0004] 1. The processing flow is complex and the amount of calculation is large;

[0005] 2. Multi-stage image processing is prone to error accumulation;

[0006] 3. It has poor robustness in environments with strong arc light, smoke, and splash interference;

[0007] 4. Different weld types usually require different feature extraction methods, resulting in limited generalization ability.

[0008] Therefore, there is an urgent need for a weld tracking method that can directly obtain the location of key welding points, reduce the complex weld reconstruction process, and has strong robustness. Summary of the Invention

[0009] (a) Purpose of the invention

[0010] The purpose of this invention is to provide a weld seam tracking method and system based on a point tracking model, which transforms the traditional weld seam reconstruction problem into a welding key point tracking problem, thereby achieving high-precision real-time weld seam tracking in complex welding environments.

[0011] (II) Technical Solution

[0012] The present invention adopts the following technical solution:

[0013] A weld seam tracking method based on a point tracking model includes: first, acquiring an overall image of the workpiece using a global CCD vision unit and acquiring a local weld seam image using a line laser vision unit; then, selecting navigation points and welding points in the initial frame and inputting them into a point tracking model for continuous tracking; the point tracking model outputs the real-time position and occlusion probability of key points, and combines a coordinate attention mechanism to enhance the feature representation of the welding area; the Kalman filter parameters are adaptively adjusted based on the occlusion probability to optimize the welding trajectory; finally, coordinate transformation is used to unify the coordinates to the robot's base coordinate system to achieve robot welding control.

[0014] (iii) Beneficial effects.

[0015] Compared with the prior art, the present invention has the following advantages:

[0016] 1. This invention transforms the traditional weld reconstruction problem into a welding key point tracking problem, reducing the complexity of image processing procedures;

[0017] 2. A unified weld seam tracking model is achieved through point tracking, which improves the system's versatility;

[0018] 3. Employ a coordinate attention mechanism to enhance spatial feature representation and improve robustness in complex welding environments;

[0019] 4. Dynamically adjusting filtering parameters based on occlusion probability improves the stability of solder joint trajectory;

[0020] 5. By fusing global vision with line laser vision, the path guidance capability and welding accuracy of the welding robot are improved. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the overall weld seam tracking process of the present invention.

[0022] Figure 2 This is a diagram illustrating the point tracking model and state estimation structure of the present invention.

[0023] Figure 3 The diagram shows the point tracking network structure for which a coordinate attention mechanism is introduced in this invention. Detailed Implementation

[0024] Example 1: System Construction

[0025] like Figure 1 As shown, the system of the present invention includes: a CCD global vision unit, a line laser vision unit, a point tracking processing module, a coordinate transformation module, and a robot control module.

[0026] in:

[0027] The CCD global vision unit is used to acquire an overall image of the workpiece and track navigation points to realize global trajectory planning for the robot;

[0028] Line laser vision units are used to acquire local weld seam images, enabling high-precision positioning of weld points.

[0029] Example 2: Tracking of Key Welding Points

[0030] The operator selects navigation points and welding points in the initial image. Then, the key points and corresponding image sequences are input into the point tracking model to achieve continuous tracking of the welding key points.

[0031] The point tracking model includes: a Backbone feature extraction network, a temporal correlation module, and an iterative update module.

[0032] Model outputs: weld point location, occlusion probability, and tracking confidence.

[0033] Example 3: Coordinate Attention Enhancement

[0034] like Figure 3 As shown, this invention introduces a coordinate attention module in the feature extraction stage of the point tracking network.

[0035] This module aggregates feature information along both the horizontal and vertical directions:

[0036]

[0037]

[0038] Subsequently, direction-related attention weights are generated to reweight the input features, thereby improving the ability to represent the features of the welding area.

[0039] Example 4: Occlusion Adaptive State Estimation

[0040] like Figure 2 As shown, this invention uses Kalman filtering to optimize the welding trajectory.

[0041] Constructing the state vector: .

[0042] The state prediction model is as follows: .

[0043] The observation model is: .

[0044] in:

[0045] This is process noise;

[0046] To observe noise.

[0047] The noise matrix is ​​dynamically adjusted based on the occlusion probability output by the point tracking model:

[0048]

[0049]

[0050] This improves trajectory stability under occlusion conditions.

[0051] Example 5: Robot Welding Control

[0052] The transformation relationships between the global vision coordinate system, the line laser vision coordinate system, and the robot base coordinate system are obtained through calibration.

[0053] After uniformly converting the welding points to the robot's base coordinate system, a welding control trajectory is generated, enabling closed-loop welding by the welding robot. in conclusion

[0054] This invention proposes a weld seam tracking method based on a point tracking model. Through dual-layer visual fusion, coordinate attention enhancement, and occlusion adaptive state estimation, it achieves highly robust real-time weld seam tracking in complex welding environments and has good engineering application value.

Claims

1. A weld seam tracking method based on a point tracking model, characterized by, Includes the following steps: S1: Obtain the overall image sequence of the welded workpiece through the global CCD vision unit, and obtain the local weld seam image sequence through the line laser vision unit; S2: Select navigation points and welding points as key points to be tracked in the initial image; S3: Combine the image sequence with the key point input point tracking model to continuously track navigation points and welding points, and output the real-time coordinates and occlusion probability of key points; S4: Adaptively adjust the state estimation model based on the occlusion probability to optimize the welding point trajectory; S5: Transform the global vision coordinate system and the linear laser vision coordinate system into the robot's base coordinate system through coordinate transformation; S6: Generates welding control commands based on navigation points and welding point trajectories in a unified coordinate system to control the welding robot to complete the welding operation.

2. The method according to claim 1, characterized in that: The point tracking model in step S3 includes a feature extraction module, a temporal correlation module, and an iterative optimization module, which are used to realize the temporal tracking of welding key points in continuous image frames.

3. The method according to claim 1, characterized in that: In step S3, a coordinate attention mechanism is used to enhance the feature map in order to improve the model's ability to perceive the spatial location information of the welding area.

4. The method according to claim 3, characterized in that: The coordinate attention mechanism aggregates features along the horizontal and vertical directions respectively, generates direction-related attention weights, and reweights the input feature map.

5. The method according to claim 1, characterized in that: In step S3, the welding environment of the training image is enhanced, including arc light enhancement, specular disturbance, smoke disturbance, and brightness change enhancement.

6. The method according to claim 1, characterized in that: The weld point state vector is constructed in step S4:

7. Among them: for the position of the weld point; For the welding point speed.

8. The method according to claim 1, characterized in that: In step S4, the process noise matrix and observation noise matrix in the Kalman filter are dynamically adjusted based on the occlusion probability output by the point tracking model.

9. The method according to claim 1, characterized in that: The coordinate unification in step S5 includes: Transformation from the global vision coordinate system to the robot's base coordinate system; Transformation from the linear laser vision coordinate system to the robot's base coordinate system.

10. The method according to claim 1, characterized in that: When the tracking confidence of the line laser vision unit is higher than the threshold, the coordinates of the local welding point are used first; when the local vision is occluded or fails, the global vision navigation point is used for path compensation.

11. A weld seam tracking system based on a point tracking model, characterized in that, include: Global CCD vision module, line laser vision module, point tracking module, state estimation module, coordinate unification module, and welding control module; The modules work together to implement the method described in any one of claims 1-9.

Images