An adaptive machine vision positioning method

Abstract

The application discloses a kind of self-adapting machine vision positioning methods, it is related to machine vision technical field, according to material size, material angle variation range and material target quantity, in double-point positioning mode, single-point positioning mode and mixed positioning mode, adaptive selection optimal positioning mode, deviation correction is carried out in material grabbing process, and in-place detection is carried out when material is transported to target position, high-precision positioning of material is realized, simultaneously, based on the working mode of double-station parallel, high-efficiency batch positioning operation of material is realized, the technical problems of insufficient precision and low efficiency of existing machine vision positioning technology in multi-person collaborative scene are solved.

Description

Technical Field

[0001] This invention relates to the field of machine vision technology, and in particular to an adaptive machine vision positioning method. Background Technology

[0002] Machine vision positioning technology, as the core perception and decision-making unit of intelligent manufacturing systems, directly determines the accuracy, efficiency, and reliability of automated production. As the manufacturing industry moves towards flexibility and intelligence, multi-person collaborative work modes are becoming increasingly common in production scenarios, exhibiting significant characteristics such as product diversification, dynamic tasks, and collaborative operations. Against this backdrop, traditional single-positioning mode machine vision positioning methods are insufficient to meet the ever-increasing accuracy requirements and the adaptability needs of multi-person collaborative operations, urgently requiring the development of intelligent positioning technologies capable of adaptive adjustment. Therefore, how to solve the problems of insufficient accuracy and low efficiency faced by machine vision positioning technology in multi-person collaborative scenarios is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0003] This invention provides an adaptive machine vision localization method to address the technical problems of insufficient accuracy and low efficiency in existing machine vision localization technologies in multi-person collaborative scenarios.

[0004] In view of this, the present invention provides an adaptive machine vision localization method, comprising: This system is applied to a dual-station, four-camera collaborative working system. The first station of the system includes a first CCD camera and a second CCD camera, while the second station includes a third CCD camera and a fourth CCD camera. The first CCD camera is mounted on a first robotic arm at the first station, and the third CCD camera is mounted on a second robotic arm at the second station. The first and third CCD cameras have downward-facing fields of view, while the second and fourth CCD cameras have upward-facing fields of view. Before grasping the material, the first and third CCD cameras, based on the material size, the range of material angle changes, and the quantity of the target material, determine the target positioning mode from three options—dual-point positioning, single-point positioning, and hybrid positioning—using an adaptive decision model to locate the material. The adaptive decision model is as follows: ; Where Mode is the adaptive decision function, TwoPoint is the two-point positioning mode, SinglePoint is the single-point positioning mode, and Hybrid is the hybrid positioning mode. For material size threshold, This is the threshold value for the range of material angle variation. The target quantity threshold for materials. For material dimensions, For the range of material angle variation, The target quantity of materials; The first and second robotic arms grab the material based on the material positioning results and move it to the designated position within the shooting field of view of the second and fourth CCD cameras; The second and fourth CCD cameras photograph the material and correct its position. The first and second robotic arms adjust the material position based on the correction results and deliver the material to the target position. The first CCD camera and the fourth CCD camera detect the material at the target location. If the material is in place, the material grabbing and positioning ends. If the material is not in place, an alarm is triggered and manual correction is performed.

[0005] Optionally, the two-point positioning mode is: Obtain the pixel coordinates and shooting coordinates of two feature points of the material; Convert the pixel coordinates of two feature points to physical coordinates; Calculate the physical distance between two feature points based on their shooting coordinates; Calculate the actual physical distance between two feature points based on their physical coordinates. Calculate the reference angle between the two feature points based on their actual physical distance. The material position is located based on the physical distance between the two feature points and the reference angle of the two feature points.

[0006] Optionally, the single-point positioning mode is: Obtain the pixel coordinates and shooting coordinates of two feature points of the material; Convert the pixel coordinates of two feature points to physical coordinates; Calculate the actual physical distance between two feature points based on their physical coordinates. Calculate the reference angle between the two feature points based on their actual physical distance. The offset angle of the material is calculated based on the reference angle of the two feature points and the real-time angle of the target position of the material. Calculate the x-direction offset value and y-direction offset value of the material based on the offset angle of the material; Calculate the absolute coordinates of the two feature points based on the material's x-direction and y-direction offset values; The material location is determined by the absolute coordinates of two feature points.

[0007] Optionally, the hybrid positioning mode is: Configure a first preset weight for the dual-point positioning mode and a second preset weight for the single-point positioning mode. Take the sum of the positioning results of the dual-point positioning mode and the single-point positioning mode after configuring the preset weights as the material positioning result, and locate the material position according to the material positioning result.

[0008] Optionally, the first preset weight is: ; in, As the first preset weight, The temperature parameter is used to adjust the sensitivity of weight allocation. Confidence score for the two-point positioning mode. Confidence score for single-point localization mode; The second preset weight is: .

[0009] Optionally, the formula for calculating the reference angle between the two feature points is: ; in, radAngle The reference angle between the two feature points. The actual physical distance between the two feature points along the y-axis. The x-axis represents the actual physical distance between two feature points.

[0010] Optionally, the formula for calculating the material's offset angle is: ; in, The offset angle of the material. This represents the real-time angle of the target material position. radAngle The reference angle between the two feature points.

[0011] Optionally, the formulas for calculating the x-direction offset and y-direction offset values ​​of the material based on its offset angle are as follows: ; ; in, This represents the material's offset value in the x-direction. This represents the material's offset value in the y-direction. The x-axis pixel compensation value for the target location. This is the y-axis pixel compensation value for the target location.

[0012] Optionally, the formula for calculating the absolute coordinates of the two feature points is: ; ; in, The x-axis coordinates are the absolute coordinates of the two feature points. The y-axis coordinates are the absolute coordinates of the two feature points. The x-axis physical compensation value for the target location. The y-axis physical compensation value for the target location. The x-axis coordinate of the position where the robot arm places the calibration plate during calibration. The y-axis coordinate is the position of the robot arm when placing the calibration plate during calibration.

[0013] Optionally, the first CCD camera and the fourth CCD camera perform on-time detection on the material at the target location. If the material is in place, the material grasping and positioning ends; if it is not in place, an alarm is triggered and manual correction is performed, including: The first and fourth CCD cameras compare the target position of the material with its actual position. If the deviation in the x-direction, y-direction, and angle are all within the deviation range, the material is determined to be in place, and the material grabbing and positioning ends. Otherwise, an alarm is triggered and manual correction is performed.

[0014] As can be seen from the above technical solutions, the adaptive machine vision localization method provided by the present invention has the following advantages: The adaptive machine vision positioning method provided by this invention adaptively selects the optimal positioning mode from two-point positioning mode, single-point positioning mode, and hybrid positioning mode based on the material size, the range of material angle variation, and the number of material targets. It performs deviation correction during the material grasping process and performs position detection when the material is transported to the target position, thus achieving high-precision positioning of the material. At the same time, based on the dual-station parallel working mode, it realizes efficient batch positioning operation of materials, solving the technical problems of insufficient accuracy and low efficiency of existing machine vision positioning technology in multi-person collaborative scenarios. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart illustrating an adaptive machine vision localization method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the workflow of the dual-station four-camera collaborative working system provided in this embodiment of the invention; Figure 3 The following are flowcharts illustrating the two-point mode and single-point mode workflows provided in this embodiment of the invention; Figure 4 This is a comparison chart of the XY direction positioning offset values ​​and angle offset values ​​of the dual-point mode and the single-point mode provided in the embodiments of the present invention. Figure 5 This is a comparison chart of the XY direction positioning deviation values ​​and angle deviation values ​​of the dual-point mode and the single-point mode provided in the embodiments of the present invention. Figure 6 This is a comparison chart of the XY direction positioning deviation values ​​between the adaptive machine vision positioning method provided in this embodiment of the invention and existing machine vision positioning methods. Detailed Implementation

[0017] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and 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.

[0018] For easier understanding, please refer to Figure 1 This invention provides an embodiment of an adaptive machine vision positioning method for a dual-station four-camera collaborative system. The first station of the dual-station four-camera collaborative system includes a first CCD camera and a second CCD camera, and the second station includes a third CCD camera and a fourth CCD camera. The first CCD camera is mounted on a first robotic arm at the first station, and the third CCD camera is mounted on a second robotic arm at the second station. The fields of view of the first and third CCD cameras are downward, and the fields of view of the second and fourth CCD cameras are upward. The method includes: Step 101: Before grasping the material, the first CCD camera and the third CCD camera determine the target positioning mode from the dual-point positioning mode, single-point positioning mode and hybrid positioning mode based on the material size, the range of material angle changes and the number of material targets, and locate the material.

[0019] It should be noted that, as Figure 2 As shown, the dual-station four-camera collaborative working system includes a first station on the left and a second station on the right. The working principles of the first and second stations are the same. In this invention, only the first station is described as an example, and the second station can be described similarly. The first CCD camera acquires material image information. Based on the material size, the range of material angle changes, and the quantity of the target material, and using an adaptive decision model, it determines the target positioning mode from the dual-point positioning mode, single-point positioning mode, and hybrid positioning mode to locate the material.

[0020] The material angle variation range refers to the maximum rotational deviation of the material's real-time attitude angle relative to the standard clamping position during actual placement, measured in angle values ​​(unit: °). It reflects the degree of fluctuation in the consistency of material placement in the production scenario. When the material angle variation range is large (exceeding the threshold), the system adaptively selects the dual-point positioning mode to obtain higher angle measurement accuracy; when the material angle variation range is small (not exceeding the threshold), the system can select the single-point positioning mode to improve processing efficiency.

[0021] The adaptive decision model is as follows: ; Where Mode is the adaptive decision function, TwoPoint is the two-point positioning mode, SinglePoint is the single-point positioning mode, and Hybrid is the hybrid positioning mode. For material size threshold, This is the threshold value for the range of material angle variation. The target quantity threshold for materials. For material dimensions, For the range of material angle variation, The target quantity of materials.

[0022] The first CCD camera acquires images of each of the N target points (materials) in each batch one by one. The number of imaging triggers is determined by the number N, meaning each target point corresponds to one trigger shot. To further shorten the production cycle, this invention adopts a hardware triggering method: after receiving an external trigger signal, the CCD camera immediately completes the exposure and writes the acquired data into an array. Subsequent positioning calculations directly call the pre-stored data in the array, eliminating the need for real-time repetitive calculations. This reduces the positioning cycle time by approximately 40% while maintaining accuracy.

[0023] Please see Figure 3 The workflow of the dual-point positioning mode is as follows: S1. Obtain the pixel coordinates and shooting coordinates of two feature points of the material.

[0024] Assume the pixel coordinates and shooting coordinates of two feature points Mark1 and Mark2 of the material are as follows: , and , .

[0025] S2. Convert the pixel coordinates of the two feature points into physical coordinates.

[0026] The matrix constructed after converting the marked pixel coordinates to physical coordinates is: ; ; in, Mat The transformation matrix is ​​obtained from the nine-point calibration process. and Transform the physical coordinates of the pixels of two feature points.

[0027] S3. Calculate the physical distance between the two feature points based on their shooting coordinates.

[0028] The distance between the physical coordinates of the two feature points at the shooting location is: ; ; in, The x-axis distance is the physical distance between the two feature points captured during the shooting process. The y-axis distance is the physical distance between two feature points captured during photography.

[0029] S4. Calculate the actual physical distance between the two feature points based on their physical coordinates.

[0030] The actual physical coordinates of the two feature points are: x 1r = x w1 ; y 1r = y w1 ; ; ; in, and Here are the actual physical coordinates of two feature points. The first feature point, Mark1, is taken as the reference point, and its actual physical coordinates are (…). x 1r , y 1r That is, the physical coordinates obtained in step S2. x w1 , y w1 ),Right now x 1r = x w1 , y 1r = y w1 Therefore, in the formula for actual physical distance below... x w1 , yw1 That is, the actual physical coordinates of the first feature point.

[0031] The actual physical distance between the two feature points is: ; ; in, The x-axis represents the actual physical distance between two feature points. The actual physical distance between the two feature points along the y-axis.

[0032] S5. Calculate the reference angle between the two feature points based on their actual physical distance.

[0033] The angle between two feature points in the physical coordinate system is used as the reference angle: ; in, radAngle The reference angle between the two feature points.

[0034] S6. Locate the material position based on the physical distance between the two feature points and the reference angle of the two feature points.

[0035] Through the angle adaptive mechanism of the dual-point positioning mode, the system can accurately capture the real-time attitude changes of the product, providing reliable geometric constraints for precise positioning. This method is particularly suitable for the handover of large-sized products in multi-person collaborative operations and assembly scenarios requiring high angular accuracy. Its angle measurement accuracy can reach 0.025°, which is significantly better than traditional single-point positioning methods.

[0036] The workflow of single-point positioning mode is as follows: T1. Obtain the pixel coordinates and shooting coordinates of two feature points of the material.

[0037] The first CCD camera acquires images of each of the N target points (materials) in each batch one by one. The number of imaging triggers is determined by the quantity N, meaning each target point corresponds to one trigger for shooting. After each shooting, the vision system extracts the image coordinates and real-time angle radTar (i.e., the real-time attitude angle of the target product in the current point image determined by the rotation amount after template matching) of the target point through template matching. These coordinates, along with the corresponding shooting coordinates and compensation values, are stored in an array for subsequent positioning calculations.

[0038] Assume the pixel coordinates and shooting coordinates of two feature points Mark1 and Mark2 of the material are as follows: , and , .

[0039] The material loading process should include a compensation value, and the physical compensation needs to be converted into pixel compensation at the target location: ; ; in, The x-axis pixel compensation value for the target location. The y-axis pixel compensation value for the target location. The x-axis physical compensation value for the target location. The y-axis physical compensation value for the target location. For the camera's single pixel precision, and These represent the directions when the target position is compensated and converted to the image coordinate system (1 indicates calibration along the same direction, -1 indicates calibration along the opposite direction).

[0040] T2. Convert the pixel coordinates of the two feature points into physical coordinates.

[0041] The matrix constructed after converting the marked pixel coordinates to physical coordinates is: ; ; in, Mat The transformation matrix is ​​obtained from the nine-point calibration process. and Transform the physical coordinates of the pixels of two feature points.

[0042] T3. Calculate the actual physical distance between the two feature points based on their physical coordinates.

[0043] The distance between the physical coordinates of the two feature points at the shooting location is: ; ; in, The x-axis distance is the physical distance between the two feature points captured during the shooting process. The y-axis distance is the physical distance between two feature points captured during photography.

[0044] The actual physical coordinates of the two feature points are: x 1r = x w1 ; y 1r = y w1 ; ; ; in, and Here are the actual physical coordinates of two feature points. The first feature point, Mark1, is taken as the reference point, and its actual physical coordinates are (…). x 1r , y 1r That is, the physical coordinates obtained in step T2. x w1 , y w1 ),Right now x 1r = x w1 , y 1r = y w1 Therefore, in the formula for actual physical distance below... x w1 , y w1 That is, the actual physical coordinates of the first feature point.

[0045] The actual physical distance between the two feature points is: ; ; in, The x-axis represents the actual physical distance between two feature points. The actual physical distance between the two feature points along the y-axis.

[0046] T4. Calculate the reference angle between the two feature points based on their actual physical distance.

[0047] The angle between two feature points in the physical coordinate system is used as the reference angle: ; in, radAngle The reference angle between the two feature points.

[0048] T5. Calculate the material offset angle based on the reference angle of the two feature points and the real-time angle of the material target position.

[0049] The formula for calculating the offset angle of the material is: ; in, The offset angle of the material. The real-time angle of the target material position is directly output by the template matching module of the vision software during the image processing stage. radAngle The reference angle between the two feature points.

[0050] T6. Calculate the x-direction offset value and y-direction offset value of the material based on the offset angle of the material.

[0051] The x-direction offset and y-direction offset values ​​of the material are: ; ; in, This represents the material's offset value in the x-direction. This represents the material's offset value in the y-direction. The x-axis pixel compensation value for the target location. This is the y-axis pixel compensation value for the target location.

[0052] T7. Calculate the absolute coordinates of the two feature points based on the x-direction offset and y-direction offset values ​​of the material.

[0053] The formula for calculating the absolute coordinates of two feature points is: ; ; in, The x-axis coordinates are the absolute coordinates of the two feature points. The y-axis coordinates are the absolute coordinates of the two feature points. The x-axis physical compensation value for the target location. The y-axis physical compensation value for the target location. The x-axis coordinate of the position where the robot arm places the calibration plate during calibration. The y-axis coordinate is the position of the robot arm when placing the calibration plate during calibration.

[0054] T8. Locate the material position based on the absolute coordinates of two feature points.

[0055] The single-point mode reduces positioning cycle time by approximately 40% through hardware triggering and pre-calculation strategies while maintaining a high accuracy of 0.013mm. This method is particularly suitable for multi-person collaborative, multi-target batch production scenarios, and its processing efficiency is significantly better than traditional point-by-point calculation methods, effectively supporting multi-station parallel operations and flexible personnel scheduling.

[0056] To achieve complementary advantages of the two positioning modes in a multi-person collaborative environment, this invention designs an adaptive decision-making mechanism based on the characteristics of the collaborative scenario, namely, constructing an adaptive decision-making model: ; Where Mode is the adaptive decision function, TwoPoint is the two-point positioning mode, SinglePoint is the single-point positioning mode, and Hybrid is the hybrid positioning mode. For material size threshold, This is the threshold value for the range of material angle variation. The target quantity threshold for materials. For material dimensions, For the range of material angle variation, The target quantity of materials.

[0057] The logic of the adaptive decision-making model is as follows: when the product size exceeds the threshold or the angle change range is large, the dual-point positioning mode is selected to ensure the angle measurement accuracy; when the number of targets is large and the angle change is within a controllable range, the single-point positioning mode is selected to improve processing efficiency; in other cases, a hybrid mode is adopted to combine the advantages of the two strategies.

[0058] The hybrid mode combines the calculation results of the two modes through weighted fusion, thus balancing accuracy and efficiency. ; in, For the location results of the mixed mode, This is the output result of the two-point positioning mode. This is the output of the single-point positioning mode. As the first preset weight, This is the second preset weight.

[0059] ; in, As the first preset weight, The temperature parameter is used to adjust the sensitivity of weight allocation. Confidence score for the two-point positioning mode. Confidence score for single-point localization mode; The second preset weight is: .

[0060] For example, confidence score conf two , conf single The temperature parameter is determined by normalizing the template matching similarity of the corresponding positioning pattern to the interval [0,1]. α Used to adjust the steepness of weight allocation. α The larger the value, the more the weight is tilted towards the higher confidence level of the mode. In this embodiment, α=5. When the confidence level of the dual-point positioning mode in a certain positioning operation... conf two =0.9, Confidence level of single-point positioning mode conf single When = 0.6, the first preset weight w two=e 5×0.9 / (e 5×0.9 +e 5×0.6 = 90.017 / (90.017+20.086)≈0.818, second preset weight w single =1- w two ≈0.182; at this point, the hybrid positioning result is 0.818 × dual-point positioning result + 0.182 × single-point positioning result.

[0061] This adaptive decision-making mechanism enables the system to maximize processing efficiency while ensuring accuracy, based on the work rhythm and task allocation of different operators in a multi-person collaborative work mode, providing a flexible and reliable positioning solution for collaborative production scenarios.

[0062] Step 102: The first and second robotic arms grab the material according to the material positioning results and move it to the designated position in the field of view of the second and fourth CCD cameras.

[0063] It should be noted that after the first CCD camera completes the positioning before material grabbing, the robotic arm grabs the material based on the positioning information and places the material above the second CCD camera, with the second CCD's field of view facing upwards.

[0064] Step 103: The second CCD camera and the fourth CCD camera take pictures of the material and correct its position.

[0065] It should be noted that after the robotic arm places the material above the second CCD camera, the second CCD camera quickly takes a picture of the material from bottom to top, obtaining the real-time pixel coordinates of the material's feature points. These real-time pixel coordinates are compared with the target position coordinates stored in the array during the pre-acquisition stage to calculate the real-time deviations Δx and Δy of the material's feature points. In step 103, the second and fourth CCD cameras only perform correction imaging, and the target position coordinates used are derived from the data stored in the array during the pre-acquisition stage in step 101.

[0066] Step 104: The first and second robotic arms adjust the material position according to the correction results and deliver the material to the target position.

[0067] It should be noted that the real-time deviation is fed back to the motion control module to control the robot arm to complete dynamic compensation before reaching the bonding position, thereby realizing real-time correction during the movement process and ensuring that the material falls accurately into the target position.

[0068] Step 105: The first CCD camera and the fourth CCD camera perform on-time detection on the material at the target location. If the material is in place, the material grabbing and positioning ends. If the material is not in place, an alarm is triggered and manual correction is performed.

[0069] It should be noted that after the robotic arm completes the material bonding action, it returns to a safe position, and the first CCD camera moves above the material to take a second picture of the bonded material. A template matching combined with precise center positioning method is used to extract the actual coordinates of the product's feature points (Mark points and center), compare them with the target specification coordinates, and calculate the x-direction deviation |Δx|, y-direction deviation |Δy|, and angle deviation |Δθ| respectively. The pass / fail criteria are |Δx|≤0.02 mm, |Δy|≤0.02 mm, and |Δθ|≤0.05°. If all three indicators are met simultaneously, a pass signal is output and the next production cycle is triggered; otherwise, the system triggers an alarm and initiates a corrective or manual intervention process, forming a complete quality closed-loop control.

[0070] The adaptive machine vision positioning method provided by this invention adaptively selects the optimal positioning mode from two-point positioning mode, single-point positioning mode, and hybrid positioning mode based on the material size, the range of material angle variation, and the number of material targets. It performs deviation correction during the material grasping process and performs position detection when the material is transported to the target position, thus achieving high-precision positioning of the material. At the same time, based on the dual-station parallel working mode, it realizes efficient batch positioning operation of materials, solving the technical problems of insufficient accuracy and low efficiency of existing machine vision positioning technology in multi-person collaborative scenarios.

[0071] To verify the effectiveness of the adaptive machine vision positioning method provided by this invention, a systematic experimental evaluation was conducted in a real industrial production environment. The experiment employed an intelligent manufacturing system equipped with four CCD cameras, a system already successfully applied in fields such as electronics manufacturing, precision assembly, and display assembly. The experimental environment temperature was controlled at 20±3°C, and the relative humidity was maintained within the range of 45-65% to ensure the stability and repeatability of the positioning results. Positioning accuracy was evaluated using international standards for visual measurement, with key indicators including x-direction deviation, y-direction deviation, and angular deviation. Positional deviation was calculated by comparing the Euclidean distance between the actual assembly position and the ideal target position, while angular deviation was measured by the absolute difference between the assembly angle and the target angle. To ensure the statistical significance of the experimental results, each experiment was repeated seven times, and the mean and standard deviation were calculated.

[0072] The experiment first systematically tested two positioning modes, and the results showed that the method of the present invention exhibits excellent accuracy and stability. In seven repeated tests, the dual-point positioning mode showed a consistent positioning accuracy with an x-direction offset range of 0.687-1.140 mm, a y-direction offset range of 3.981-4.244 mm, and an angle offset range of -5.132° to 2.402°. The single-point positioning mode was tested against seven different target points, with an x-direction offset range of -2.606 to 5.647 mm, a y-direction offset range of -5.369 to 5.212 mm, and an angle offset range of -61.772° to 119.625°. These data indicate that the system can accurately position the product even when there are significant positional and angular deviations. It should be noted that the offset values ​​of each mode described in this invention (including those mentioned above)... x direction, y The directional offset range and angular offset range refer to the initial deviation of the actual placement position of the product relative to the taught standard position during loading. This is the input condition that the system needs to compensate for, not the residual error after positioning compensation. In the dual-point positioning mode, the same target point is repeatedly positioned, resulting in a smaller offset range. In the single-point positioning mode, multiple different target points are targeted, resulting in a larger offset range. After positioning compensation by this invention, the residual deviation (see...) Figure 5 The angular deviation is no more than 0.05° and the displacement deviation is no more than 0.02mm, therefore, even if the initial angular offset range is large, the system can still achieve accurate positioning. Figure 4 As shown, during multi-person collaborative production, the maximum product offset along the X, Y, and R directions reached 4.2 mm and 5°, respectively. Product handover and workstation switching between different operators may introduce significant angular errors. To verify the system's performance over a wide angle range in multi-person collaborative work mode, a wide-angle verification test was conducted. Regarding angle measurement, tests were performed on products at seven different positions. The average angle error in the dual-point mode was 0.025°, and in the single-point mode, it was 0.024°, both demonstrating excellent angle measurement capabilities. In terms of position measurement, through template matching and precise center positioning, the average accuracy of the dual-point positioning mode reached 0.016 mm (x-direction) and 0.015 mm (y-direction), while the average accuracy of the single-point positioning mode was 0.013 mm (x-direction) and 0.015 mm (y-direction), both achieving sub-millimeter-level positioning levels, significantly superior to traditional methods.

[0073] like Figure 5As shown, the angular deviation remains within 0.05°, and the displacement deviation remains within 0.02mm, fully demonstrating that the method of this invention can meet the accuracy requirements. The method of this invention shows a significant advantage in operational efficiency compared to traditional methods. Traditional teaching methods require 32.5 minutes to complete a seven-point positioning sequence, with 28 minutes requiring continuous operator supervision. This severely impacts overall efficiency and personnel scheduling flexibility in multi-person collaborative work environments. In contrast, the automated positioning method completes the entire sequence in 8.5 minutes in two-point mode and 6.8 minutes in single-point mode, reducing operator intervention time to 2.3 and 1.8 minutes respectively, representing approximately 75% reduction in operation time and approximately 85% reduction in operator intervention. This significantly reduces communication costs and waiting time in multi-person collaborative work. This efficiency improvement is achieved simultaneously with accuracy improvement; the method of this invention achieves an accuracy of 0.013-0.016mm, superior to the 0.043-0.698mm of existing methods. The system maintains its performance within an extended calibration stabilization period of 72-96 hours, which is a significant improvement over the 24-36 hour stabilization period of traditional methods. This is especially important for multi-shift, multi-person collaborative production modes, as it reduces production interruptions and personnel coordination difficulties caused by frequent calibrations.

[0074] To comprehensively evaluate the technical advantages of the method of the present invention, several representative existing methods were selected for comparative experiments. Figure 6 This shows a comparison of positioning deviation values ​​using different methods. Figure 6 Figures (a) and (b) show that the deviations in both the single-point and double-point modes of this invention remain within 0.02 mm in the X and Y directions, demonstrating excellent stability. In contrast, the method of Wang F, Liang C, Ru C, Cheng H. An Improved Point Cloud Descriptor for VisionBased Robotic Grasping System. Sensors (Basel) 2019, 19(10): 2225 achieves a deviation of 0.043-0.698 mm in the X direction. Figure 6 (Figure (c)), Parviziomran I, Cao S, Yang H, Park S, Won D. Optimization of Passive Chip Components Placement with Self-Alignment Effect for Advanced Surface Mounting Technology. Procedia Manufacturing 2019, 39: 202-209. The method deviation range is 0.2-0.8mm. Figure 6 (d) Figure), Zhou K, Meng Z, He M, Hou J, Li T. Design and Test of a Sorting Device Based on Machine Vision. IEEE Access 2020, PP(99): 1. The method has deviations of 0.030mm and 0.017mm on the X and Y axes, respectively. Figure 6 (e) Figure); Fu T, Li F, Zheng Y, Quan W, Li Y. Dynamically Grasping with Incomplete Information Workpiece Based on Machine Vision. In: 2019 IEEE International Conference on Unmanned Systems (ICUS). The method deviation range is 0.2-1.0 mm. Figure 6 (See Figure (f)). These comparative results clearly demonstrate that the method of the present invention is significantly superior to the prior art in terms of positioning accuracy and stability, especially in handling complex scenes and multi-target positioning tasks, where it exhibits obvious technical advantages.

[0075] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An adaptive machine vision localization method, characterized in that, This system is applied to a dual-station, four-camera collaborative working system. The first station of the system includes a first CCD camera and a second CCD camera, while the second station includes a third CCD camera and a fourth CCD camera. The first CCD camera is mounted on a first robotic arm at the first station, and the third CCD camera is mounted on a second robotic arm at the second station. The first and third CCD cameras have downward-facing fields of view, while the second and fourth CCD cameras have upward-facing fields of view. Before grasping the material, the first and third CCD cameras, based on the material size, the range of material angle changes, and the quantity of the target material, determine the target positioning mode from three options—dual-point positioning, single-point positioning, and hybrid positioning—using an adaptive decision model to locate the material. The adaptive decision model is as follows: ； Where Mode is the adaptive decision function, TwoPoint is the two-point positioning mode, SinglePoint is the single-point positioning mode, and Hybrid is the hybrid positioning mode. For material size threshold, This is the threshold value for the range of material angle variation. The target quantity threshold for materials. For material dimensions, For the range of material angle variation, The target quantity of materials; The first and second robotic arms grab the material based on the material positioning results and move it to the designated position within the shooting field of view of the second and fourth CCD cameras; The second and fourth CCD cameras photograph the material and correct its position. The first and second robotic arms adjust the material position based on the correction results and deliver the material to the target position. The first CCD camera and the fourth CCD camera detect the material at the target location. If the material is in place, the material grabbing and positioning ends. If the material is not in place, an alarm is triggered and manual correction is performed.

2. The adaptive machine vision localization method according to claim 1, characterized in that, The dual-point positioning mode is: Obtain the pixel coordinates and shooting coordinates of two feature points of the material; Convert the pixel coordinates of two feature points to physical coordinates; Calculate the physical distance between two feature points based on their shooting coordinates; Calculate the actual physical distance between two feature points based on their physical coordinates. Calculate the reference angle between the two feature points based on their actual physical distance. The material position is located based on the physical distance between the two feature points and the reference angle of the two feature points.

3. The adaptive machine vision localization method according to claim 1, characterized in that, The single-point positioning mode is: Obtain the pixel coordinates and shooting coordinates of two feature points of the material; Convert the pixel coordinates of two feature points to physical coordinates; Calculate the actual physical distance between two feature points based on their physical coordinates. Calculate the reference angle between the two feature points based on their actual physical distance. The offset angle of the material is calculated based on the reference angle of the two feature points and the real-time angle of the target position of the material. Calculate the x-direction offset value and y-direction offset value of the material based on the offset angle of the material; Calculate the absolute coordinates of the two feature points based on the material's x-direction and y-direction offset values; The material location is determined by the absolute coordinates of two feature points.

4. The adaptive machine vision localization method according to claim 1, characterized in that, The hybrid positioning mode is: Configure a first preset weight for the dual-point positioning mode and a second preset weight for the single-point positioning mode. Take the sum of the positioning results of the dual-point positioning mode and the single-point positioning mode after configuring the preset weights as the material positioning result, and locate the material position according to the material positioning result.

5. The adaptive machine vision localization method according to claim 4, characterized in that, The first preset weight is: ； in, As the first preset weight, The temperature parameter is used to adjust the sensitivity of weight allocation. Confidence score for the two-point positioning mode. Confidence score for single-point localization mode; The second preset weight is: 。 6. The adaptive machine vision localization method according to claim 2, characterized in that, The formula for calculating the reference angle between two feature points is: ； in, radAngle The reference angle between the two feature points. The actual physical distance between the two feature points along the y-axis. The x-axis represents the actual physical distance between the two feature points.

7. The adaptive machine vision localization method according to claim 3, characterized in that, The formula for calculating the offset angle of the material is: ； in, The offset angle of the material. This represents the real-time angle of the target material position. radAngle The reference angle between the two feature points.

8. The adaptive machine vision localization method according to claim 7, characterized in that, The formulas for calculating the x-direction offset and y-direction offset of the material based on its offset angle are as follows: ； ； in, This represents the material's offset value in the x-direction. This represents the material's offset value in the y-direction. The x-axis pixel compensation value for the target location. This is the y-axis pixel compensation value for the target location.

9. The adaptive machine vision localization method according to claim 8, characterized in that, The formula for calculating the absolute coordinates of two feature points is: ； ； in, The x-axis coordinates are the absolute coordinates of the two feature points. The y-axis coordinates are the absolute coordinates of the two feature points. The x-axis physical compensation value for the target location. The y-axis physical compensation value for the target location. The x-axis coordinate of the position where the robot arm places the calibration plate during calibration. The y-axis coordinate is the position of the robot arm when placing the calibration plate during calibration.

10. The adaptive machine vision localization method according to claim 1, characterized in that, The first and fourth CCD cameras detect the material's arrival at the target location. If the material is in place, the material grabbing and positioning process ends; if it is not in place, an alarm is triggered and manual correction is performed, including: The first and fourth CCD cameras compare the target position of the material with its actual position. If the deviation in the x-direction, y-direction, and angle are all within the deviation range, the material is determined to be in place, and the material grabbing and positioning ends. Otherwise, an alarm is triggered and manual correction is performed.

Images