A multi-axle hole connector non-contact alignment method and system

Abstract

The present application relates to the technical field of intelligent equipment, and particularly relates to a multi-shaft hole connector non-contact alignment method, which constructs a non-contact alignment framework of "image acquisition-rotational pose detection-continuous frame filtering-closed loop servo-convergence judgment", relies on basic vision and motion control logic, directly outputs a rotational bounding box with an angle through a rotational target detection network, and internally maps image angle to physical yaw error, and for a circular aviation electrical connector or the like rotational symmetric piece, can synchronously solve position deviation and angle deviation in a non-contact state, through a filtering and smoothing processing step of continuous time sequence pose error, stably processes multiple frame detection results, so that visual servo control no longer depends on instantaneous detection results of a single frame, and convergence stability of closed loop adjustment is improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent equipment technology, specifically to a non-contact alignment method and system for multi-axis hole connectors. Background Technology

[0002] Multi-axis hole assembly of industrial robots is a core precision process in the fields of high-end electronic manufacturing and intelligent equipment. Among them, circular aviation electrical connectors are widely used for signal and power connections in complex electrical systems due to their stable connection, strong anti-interference, and compact structure. However, due to their multi-pin, multi-hole, and rotationally symmetrical structural characteristics, existing vision-guided assembly solutions have obvious defects and cannot meet the requirements of high-precision automated assembly. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides a non-contact alignment method for multi-axis hole connectors, comprising: S1. Acquire assembly images of the multi-axis hole connector; S2. Obtain the rotation bounding box parameters of the connector based on the rotation target detection network, map the angle of the rotation bounding box to the physical yaw attitude error of the connector, and obtain the position deviation and rotation attitude deviation of the connector. S3. Filter and smooth the position deviation and rotation attitude deviation of the continuous time sequence to obtain the stable pose error; S4. Adjust the pose based on the stable pose error; S5. Repeat the above steps until the stable pose error meets the convergence condition.

[0004] Preferably, in S1, acquiring the assembly image of the multi-axis connector specifically includes: Images of the connector from different directions are simultaneously acquired using a dual-view image acquisition device, with the dual-view devices arranged along the orthogonal direction of the connector.

[0005] Preferably, between S1 and S2, the following is also included: S11. Perform instance segmentation processing on the acquired original assembly image, retain the foreground area of ​​the connector, and filter out background interference.

[0006] Preferably, in S2, the rotating target detection network adopts the YOLOv8s-OBB model.

[0007] Preferably, in S2, mapping the angle of the rotated bounding box to the physical yaw attitude error of the connector specifically includes: A reference rotation angle corresponding to the standard assembly state of the connector is predefined. The physical yaw attitude error is calculated based on the difference between the current rotation angle output by the rotating target detection network and the reference rotation angle.

[0008] Preferably, in S3, the filtering and smoothing process adopts a sliding window averaging strategy, which takes the average of the pose error of N consecutive frames to suppress random fluctuations in single-frame detection.

[0009] Preferably, in S4, the pose adjustment based on the stable pose error specifically includes: The stable pose error is mapped to a pose increment in a six-DOF Cartesian space, where the degree-of-freedom adjustment of the non-aligned plane is set to zero.

[0010] Preferably, the pose increment adopts a small-step progressive adjustment strategy, setting a proportional gain coefficient of less than 1 for the adjustment amount of each degree of freedom, thereby limiting the adjustment range in a single operation.

[0011] Preferably, in S5, the convergence condition includes a position error threshold and an attitude error threshold. When the position deviation is less than the position error threshold and the rotation attitude deviation is less than the attitude error threshold, the convergence condition is determined to be met.

[0012] On the other hand, the present invention also provides a non-contact alignment system for multi-axis hole connectors, for performing the non-contact alignment method for multi-axis hole connectors described in any of the above claims, characterized in that it includes: The execution unit includes a robotic arm for performing connector pose adjustment and insertion movements; A visual sensing unit, used to acquire connector assembly images; An image processing unit is used to perform rotating target detection, pose error calculation, and continuous frame error filtering. And a motion control unit, which is used to send correction instructions to the execution unit based on error feedback.

[0013] Compared with the prior art, the present invention has the following beneficial effects: This invention provides a non-contact alignment method for multi-axis hole connectors. It constructs a non-contact alignment framework consisting of "image acquisition—rotation pose detection—continuous frame filtering—closed-loop servoing—convergence judgment." Relying on basic vision and motion control logic, it directly outputs angled rotational bounding boxes through a rotation target detection network and incorporates a mapping logic from image angles to physical yaw errors. For rotationally symmetrical components such as circular avionics connectors, it can simultaneously calculate position and angle deviations in a non-contact state. Through continuous temporal pose error filtering and smoothing steps, it stabilizes the multi-frame detection results, making visual servo control no longer dependent on the instantaneous detection results of a single frame, thus improving the convergence stability of closed-loop adjustment. Attached Figure Description

[0014] Figure 1This is a flowchart of the non-contact alignment method for multi-axis hole connectors provided by the present invention; Figure 2 This is a flowchart illustrating the connector pose perception process based on rotation target detection in the non-contact alignment method for multi-axis hole connectors provided by this invention. Figure 3 This is a flowchart of continuous frame pose error modeling and visual servo closed-loop process for the non-contact alignment method of multi-axis hole connectors provided by the present invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] like Figure 1 As shown, the present invention provides a non-contact alignment method for multi-axis hole connectors, comprising the following steps: S1. Acquire assembly images of the multi-axis hole connector; an RGB camera can be used as the image acquisition device, which captures red, green, and blue light through three independent sensors or filters. The combination of these colors can generate images of various colors, and finally output a two-dimensional pixel matrix to meet the usage requirements. Of course, in actual use, those skilled in the art can choose a suitable image acquisition device, as long as it meets the usage requirements, which will not be elaborated here.

[0017] S2. Obtain the rotation bounding box parameters of the connector based on the rotation target detection network, and map the angle of the rotation bounding box to the physical yaw attitude error of the connector to obtain the position deviation and rotation attitude deviation of the connector. Among them, the YOLOv8s-OBB rotation target detection model can be used to detect the connector target. By introducing the CBAM attention enhancement module in the backbone feature extraction stage, the network's ability to perceive the connector edge contour, shaft hole area and pin structure features is improved. Combined with a lightweight shared convolutional detection head, the number of model parameters and inference computation overhead are reduced, and the synchronous output of the connector center position and rotation angle information is realized. At the same time, by establishing the mapping relationship between the rotation target box angle and the assembly reference direction, the rotation angle of the image output by the network is converted into the physical yaw attitude error of the connector in the actual assembly process, realizing a unified representation of the two-dimensional position and rotation attitude of the connector, and providing input for subsequent pose error modeling and visual servo control.

[0018] S3. Filter and smooth the continuous temporal position and rotational attitude deviations to obtain a stable pose error. Based on the rotating target detection results, extract the connector center position offset and rotation angle deviation, and construct the connector pose error vector. Map the pixel deviations in the image space to the pose adjustment amount of the end effector, achieving a unified expression of position and attitude errors. S4. Adjust the pose based on the stable pose error; calculate the pose correction of the end effector based on the filtered pose error, and perform a small-step progressive adjustment. Simultaneously, after each adjustment, the image can be re-acquired and the error updated, forming a closed-loop visual servoing process of "detection-calculation-adjustment," achieving dynamic alignment of the connector's position and attitude in a non-contact state.

[0019] S5. Repeat the above steps until the stable pose error meets the convergence condition. Specifically, after each round of pose adjustment, the connector position error and attitude error are judged to converge. When the error is lower than the set threshold or the maximum number of iterations is reached, the non-contact alignment process ends and the insertion assembly operation is performed. Otherwise, return to the image acquisition step to continue the loop iteration.

[0020] Based on the above steps, this invention constructs a non-contact alignment framework consisting of "image acquisition—rotation pose detection—continuous frame filtering—closed-loop servoing—convergence judgment." Relying on basic vision and motion control logic, it directly outputs angled rotating bounding boxes through a rotating target detection network and incorporates a mapping logic from image angles to physical yaw errors. For rotationally symmetrical components such as circular aviation electrical connectors, it can simultaneously calculate position and angle deviations in a non-contact state. Through continuous temporal pose error filtering and smoothing steps, it stabilizes the multi-frame detection results, making visual servoing control no longer dependent on the instantaneous detection results of a single frame, thus improving the convergence stability of closed-loop adjustment. Furthermore, the equipment used in this invention has relatively low cost, meaning that hardware costs and system complexity are reduced while ensuring assembly quality.

[0021] In one preferred embodiment, in S1 of the present invention, images of the connector in different directions are simultaneously acquired by a dual-view image acquisition device, and the dual-view of the image acquisition device is arranged along the orthogonal direction of the connector, which can simultaneously calculate the position deviation and rotation attitude of the connector from two mutually perpendicular directions, effectively decoupling the coupling error of position and attitude under single view, and improving the accuracy of yaw angle detection and system robustness.

[0022] In one preferred embodiment, after acquiring the assembly image of the multi-hole coupling in S1, the method further includes S11: performing instance segmentation processing on the acquired original assembly image to retain the foreground area of ​​the connector and filter out background interference. The multi-hole connector image is acquired by image acquisition devices installed on both sides of the assembly station. The original image is then segmented to retain the foreground image, reducing the impact of complex backgrounds, ambient light changes, and connector surface reflections on subsequent detection, thus providing stable input for the rotating target detection network.

[0023] In one preferred embodiment, in S2, mapping the angle of the rotated bounding box to the physical yaw attitude error of the connector specifically includes: pre-defining a reference rotation angle corresponding to the standard assembly state of the connector, and calculating the physical yaw attitude error based on the difference between the current rotation angle output by the rotating target detection network and the reference rotation angle. This setting eliminates the ambiguity in the angle representation of the rotating target detection network, establishes a unique mapping relationship from image domain angles to physical assembly domain yaw errors, and provides stable and interpretable attitude error input for continuous frame filtering and visual servo closed-loop control.

[0024] In one preferred embodiment, in S3, the filtering and smoothing process employs a sliding window averaging strategy, averaging the pose errors over N consecutive frames to suppress random fluctuations in single-frame detection. This setting addresses the issue of random fluctuations in single-frame visual detection caused by variations in illumination, metallic reflections, and image noise. By using a sliding window averaging strategy for continuous frame filtering of the pose error vector in the continuous time-series images, the impact of detection jitter on the control system can be reduced, thereby improving the stability and error convergence performance during visual servo adjustment and providing a stable input for subsequent closed-loop servo control.

[0025] In one preferred embodiment, in S4, adjusting the pose based on the stable pose error specifically includes: mapping the stable pose error to a pose increment in a six-DOF Cartesian space, wherein the adjustment of the degrees of freedom of the non-aligned plane is set to zero. This can lock motion axes independent of the planar assembly of the multi-axis hole connector, reducing the computational complexity required.

[0026] In one preferred embodiment, the pose increment employs a small-step progressive adjustment strategy, setting a proportional gain coefficient of less than 1 for the adjustment amount of each degree of freedom, thus limiting the adjustment amplitude in a single step. This can suppress control jitter caused by visual inspection noise, reduce the requirements for the dynamic performance of the servo system, and achieve high-precision, high-stability smooth convergence of multi-axis hole connectors during non-contact alignment.

[0027] In one preferred embodiment, in S5, the stable pose error satisfies the convergence condition, which includes a position error threshold and an attitude error threshold. When the position deviation is less than the position error threshold and the rotation attitude deviation is less than the attitude error threshold, the convergence condition is determined to be satisfied.

[0028] Furthermore, the present invention also provides a non-contact alignment system for multi-axis hole connectors, used to perform the above-mentioned non-contact alignment method for multi-axis hole connectors. The system includes an execution unit, a vision perception unit, an image processing unit, and a motion control unit. The execution unit has a robotic arm for performing connector pose adjustment and insertion movements; the vision perception unit is used to acquire connector assembly images; the image processing unit is used to perform rotating target detection, pose error calculation, and continuous frame error filtering; and the motion control unit is used to send correction instructions to the execution unit based on error feedback.

[0029] The non-contact alignment system for multi-axis hole connectors provided by this invention operates as follows in practical use: First, connector image acquisition and preprocessing: A multi-axis hole connector non-contact alignment system is constructed, consisting of an execution unit, a vision sensing unit, an image processing unit, and a motion control unit. The vision sensing unit comprises two RGB industrial cameras mounted along the X-axis and Y-axis of the connector, respectively, to synchronously acquire connector assembly images. The image processing unit and motion control unit establish data communication with the industrial cameras and execution unit via industrial communication interfaces, respectively, to perform image processing, pose calculation, and motion control command issuance.

[0030] Secondly, connector image acquisition and background segmentation are performed. After the robotic arm of the execution unit moves to the pre-alignment area, a sequence of connector images is continuously acquired using an industrial camera mounted along the X-axis: CamX. i ={c1,c2,c3,…… ,c n Simultaneously, an industrial camera mounted along the Y-axis synchronously acquires an image sequence from another perspective: CamY. i ={c 1′ ,c 2′ ,c 3′ ,…… ,c n′}, where c n With c n′ These represent the original connector images acquired at the corresponding viewpoint at time n. The acquired image sequence is input into the FastSAM instance segmentation model to segment the connector body region, removing the background region and retaining only the connector body and shaft hole region information. The connector image sequence after background segmentation is represented as follows: SegX i ={s1,s2,s3,……,s nSegY i ={s 1′ ,s 2′ ,s 3′ ,…… ,s n′} Among them, s n With s n′ This represents the connector foreground image obtained after FastSAM segmentation. Subsequently, the segmented image sequence is input into the subsequent rotating target detection module for extracting connector position and rotation angle information.

[0031] Second, such as Figure 2 As shown, the configuration includes rotating target detection and connector pose representation. First, rotating target detection and 2D pose extraction are performed, and the connector foreground image sequence SegX obtained in the previous step is used. i With SegY i The input rotating target detection network is used to detect the position and rotational attitude of the connector. This implementation adopts a rotating target detection structure based on YOLOv8s-OBB, realizing the two-dimensional pose perception of the connector by outputting rotational bounding box parameters. To improve the detection stability and real-time processing capability in industrial environments, a CBAM attention enhancement module is introduced in the backbone feature extraction stage to improve the network's feature response capability to the connector edge contour, pin area, and shaft hole structure area; at the same time, a lightweight shared convolutional structure LSCD is used in the detection head to reduce the network's computational load and improve the real-time inference speed. After rotating target detection, the connector rotation frame detection sequences under dual-view perspectives are obtained: R Xi ={R1,R2,R3,……,R n}、R Yi ={R 1′ ,R 2′ ,R 3′ ,……,R n′},in: R i =(x i ,y i ,w i ,h i ,θ i ), R i′ =(x i′ ,y i′ ,w i′ ,h i′ ,θ i′ ) In the formula, x i and y i Here are the pixel coordinates of the connector's center point; w i and h i θ represents the width and height parameters of the rotated frame.i The rotation angle output by the rotating target detection network.

[0032] Secondly, perform rotation angle physical mapping and yaw error transformation; define the angle parameter output by the rotating target detection network as the angle between the long side of the rotating box and the image coordinate system, i.e., the horizontal direction, and define its output angle range as follows: This angle is only used to describe the geometric orientation of the target frame in the image and does not directly correspond to the physical yaw attitude of Rx and Ry during the actual assembly process of the connector. Therefore, in this embodiment, the long side direction of the connector rotation frame is defined as the connector main direction axis, and the angle between this direction axis and the assembly reference direction is mapped to the connector physical yaw angle.

[0033] Specifically, during the system initialization phase, the reference rotation angle θ is first acquired when the connector is in a standard alignment state. ref Since the multi-axis connector in this embodiment uses a planar plug-in assembly method, and the main direction axis of the connector in its standard alignment state corresponds to the vertical direction of the image coordinate system, the vertical direction is defined as the standard assembly posture of the connector, i.e., the reference rotation angle: θ. ref =90°, and then, during each frame of rotating target detection, the physical yaw angle error ∆θi of the connector's current posture relative to the standard assembly posture is calculated based on the rotation angle θi output by the detection network. Since the positive and negative ranges of the rotating target detection angle have different representations, a segmented mapping method is used for unified conversion: Specifically: when the detection angle is in the range [0, π / 2], it indicates that the connector has rotated clockwise relative to the target attitude; therefore, the difference between the reference angle and the current angle is used to represent the yaw error. When the detection angle is in the range [-π / 2, 0], it indicates that the connector has rotated counterclockwise relative to the target attitude; to maintain the continuity of the physical yaw angle, the absolute value of the angle is used to represent the current yaw error. Through this method, the image rotation angle output by the rotating target detection network is converted into the physical yaw attitude error during the actual assembly process of the connector, thus achieving an explicit expression of the connector's rotation state.

[0034] Finally, dual-view pose sequence construction and feature representation are performed. Based on the rotating target detection results, connector center position sequence and rotation attitude sequence are established respectively: P Xi ={(x1,y1),(x2,y2),……,(x n ,y n )} P Yi ={(x 1′ ,y 1′),(x 2′ ,y 2′ ),…… ,(x n′ ,y n′ )} Θ Xi ={Δθ 1, Δθ2,…… ,Δθ n} Θ Yi ={Δθ 1′ ,Δθ 2′ ,…… ,Δθ n′} Among them, P Xi With P Yi These represent the sequence of changes in the connector center position under dual perspectives, Θ Xi With Θ Yi These represent the yaw angle change sequence of the connector relative to the target assembly posture. The above position and posture sequences will be used together as inputs for subsequent shaft hole pose error modeling and visual servo control modules.

[0035] Third, such as Figure 3 As shown, continuous frame pose error modeling and closed-loop servo control are configured. First, a dual-view pose acquisition and error vector calculation module is constructed, based on the dual-view connector center position sequence P obtained in the previous step. Xi P Yi and rotational attitude sequence Θ Xi Θ Yi Calculate the positional and attitude deviations of the connector relative to the target assembly state at the current moment. The positional offsets of the connector under both viewpoints are expressed as follows: , ; In the formula, ( , The coordinates represent the center positions of the female and male connectors from the corresponding viewpoint when the connector is in a standard assembly state.

[0036] Meanwhile, based on the physical yaw angle error obtained in the preceding sequence: Δθ i , Δθ i′ Together, construct the connector's pose error vector at the current moment: ,in: ; ; In the formula, p and q are defined as the proportionality coefficients between the image pixel coordinates and the actual spatial displacement. These coefficients are determined empirically by the camera device's installation distance at the end effector, with p = 0.01 and q = 0.005, and are used for the mapping and conversion between the image rotation angle and the robotic arm's posture adjustment. Through this method, a unified mapping is achieved between the spatial errors of the dual-view images and the assembly pose errors of the robotic arm.

[0037] Secondly, a continuous frame error filtering and stabilization module is constructed (construction window). ,calculate Due to variations in lighting conditions, connector surface reflections, and visual inspection noise in industrial environments, single-frame detection results can exhibit random fluctuations. Therefore, a sliding window filter is applied to the continuous temporal error vector to improve the stability of pose error estimation. The pose errors of N consecutive frames are constructed as an error window: W i ={e i−N+1 ,e i−N+2 ,…… ,e i} And the error within the window is calculated using the formula: The smoothed pose error is obtained by averaging: The sliding window method suppresses single-frame visual detection jitter through continuous frame error smoothing, avoiding frequent oscillations or overcorrection during the adjustment process of the robotic arm, thereby improving the stability and convergence performance of the closed-loop control process.

[0038] Furthermore, the visual servoing pose incremental control module based on error feedback will smooth the pose error. Input the visual servo control module, and combine it with the six-DOF Cartesian space motion form of the robot's end effector. Set the Z-direction translation and Z-axis rotation of the components not involved in the current plane alignment task to 0, and construct the target pose increment at the current moment: To reduce robotic arm oscillations caused by sudden error changes during visual servo adjustments, a proportional gain coefficient pos_gain=[0.1, 0.1, 0, 0.1, 0.1, 0] is set for the pose increment of each degree of freedom, limiting the proportional coefficient of each degree of freedom's pose increment to the range [0,1]. Simultaneously, to ensure real-time synchronization between visual detection and robotic arm control, the robot servo command transmission cycle is set to 0.008 s to make the fine-tuning motion of the robotic arm's end effector more continuous and smooth. This is achieved by executing the following function... ServoCart(desc_pos, pos_gain=[0.1, 0.1, 0, 0.1, 0.1, 0] ,cmdT=0.008) The robot arm's end-effector pose is updated. During the movement, the robot's execution unit simultaneously acquires images of the connector and repeatedly performs rotational target detection, error calculation, and continuous frame filtering processes, updating the target pose increment des_pos in real time. This forms a dynamic closed-loop control process of "image acquisition - rotational pose detection - error modeling - visual servo adjustment".

[0039] Finally, after each round of visual servo adjustment, a convergence condition is determined for the current pose error. When the following conditions are met: , When the current position and orientation of the connector meet the alignment requirements of the shaft hole, the non-contact alignment process ends, and the robot is controlled to perform subsequent insertion and assembly operations. Wherein: ε p ε is the position error threshold. r The alignment accuracy is set to ε when the alignment task begins, serving as the attitude error threshold. p = 0.3mm, ε r =0.3°. If the current error does not meet the convergence condition, continue to execute the next round of vision servo closed-loop adjustment until the non-contact alignment of the multi-axis hole connector is completed.

[0040] Certain terms are used in the specification and claims to refer to specific components. Those skilled in the art will understand that hardware manufacturers may use different names to refer to the same component. This specification and claims do not distinguish components based on differences in name, but rather on differences in function. The terms "comprising" and "including" used throughout the specification and claims are open-ended and should be interpreted as "comprising / including but not limited to". "Approximately" means that within an acceptable margin of error, those skilled in the art can solve the technical problem and substantially achieve the technical effect within a certain margin of error. The following descriptions in the specification describe preferred embodiments of this application; however, these descriptions are intended to illustrate the general principles of this application. Although the invention has been described in detail with reference to embodiments, those skilled in the art should understand that modifications or equivalent substitutions to the technical solutions of this invention do not depart from the spirit and scope of the invention and should be covered by the claims.

Claims

1. A non-contact alignment method for a multi-axis hole connector, characterized in that, include: S1. Acquire assembly images of the multi-axis hole connector; S2. Obtain the rotation bounding box parameters of the connector based on the rotating target detection network, map the angle of the rotation bounding box to the physical yaw attitude error of the connector, and obtain the position deviation and rotation attitude deviation of the connector. S3. Filter and smooth the position deviation and rotation attitude deviation of the continuous time sequence to obtain the stable pose error; S4. Adjust the pose based on the stable pose error; S5. Repeat the above steps until the stable pose error meets the convergence condition.

2. The non-contact alignment method for multi-axis hole connectors according to claim 1, characterized in that, In S1, acquiring the assembly image of the multi-axis hole connector specifically includes: Images of the connector from different directions are simultaneously acquired using a dual-view image acquisition device, with the dual-view devices arranged along the orthogonal direction of the connector.

3. The non-contact alignment method for multi-axis hole connectors according to claim 1, characterized in that, Between S1 and S2, it also includes: S11. Perform instance segmentation processing on the acquired original assembly image, retain the foreground area of ​​the connector, and filter out background interference.

4. The non-contact alignment method for multi-axis hole connectors according to claim 1, characterized in that, In S2, the rotating target detection network adopts the YOLOv8s-OBB model.

5. The non-contact alignment method for multi-axis hole connectors according to claim 1, characterized in that, In S2, mapping the angle of the rotated bounding box to the physical yaw attitude error of the connector specifically includes: A reference rotation angle corresponding to the standard assembly state of the connector is predefined. The physical yaw attitude error is calculated based on the difference between the current rotation angle output by the rotating target detection network and the reference rotation angle.

6. The non-contact alignment method for multi-axis hole connectors according to claim 1, characterized in that, In S3, the filtering and smoothing process adopts a sliding window averaging strategy, which takes the average of the pose error of N consecutive frames to suppress random fluctuations in single-frame detection.

7. The non-contact alignment method for multi-axis hole connectors according to claim 1, characterized in that, In S4, pose adjustment based on stable pose error specifically includes: The stable pose error is mapped to the pose increment in the six-DOF Cartesian space of the robotic arm end effector, where the degree-of-freedom adjustment of the non-aligned plane is set to zero.

8. The non-contact alignment method for multi-axis hole connectors according to claim 7, characterized in that, The pose increment adopts a small step-by-step gradual adjustment strategy, setting a proportional gain coefficient of less than 1 for the adjustment amount of each degree of freedom, thus limiting the adjustment range in a single step.

9. The non-contact alignment method for multi-axis hole connectors according to claim 1, characterized in that, In S5, the convergence condition includes a position error threshold and an attitude error threshold. When the position deviation is less than the position error threshold and the rotation attitude deviation is less than the attitude error threshold, the convergence condition is determined to be met.

10. A non-contact alignment system for a multi-axis hole connector, used to perform the non-contact alignment method for a multi-axis hole connector according to any one of claims 1 to 9, characterized in that, include: The execution unit includes a robotic arm for performing connector pose adjustment and insertion movements; A visual sensing unit, used to acquire connector assembly images; An image processing unit is used to perform rotating target detection, pose error calculation, and continuous frame error filtering. And a motion control unit, which is used to send correction instructions to the execution unit based on error feedback.

Images