Spool stacking method based on stereoscopic vision and applied to AGV forklift

Abstract

The invention discloses a stereoscopic vision-based spool stacking method applied to an AGV forklift. The stereoscopic vision-based spool stacking method comprises the following steps of S1, acquiring three-dimensional point cloud data; s2, optimizing a matching error by a whale optimization algorithm; s3, calculating an adjustment amount and an initialized firefly population; s4, optimizing a pose adjustment scheme through a firefly algorithm; and S5, adjustment is executed, and stacking is completed. According to the invention, the stacking precision, the pose adjustment precision and the operation efficiency of the spools are obviously improved, the alignment error and the stacking failure risk are reduced, and the efficient and automatic operation of the AGV forklift is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of machine vision, and particularly to a method for stacking spools applied to an AGV forklift based on stereo vision. Background Art

[0002] In modern industrial production and warehousing logistics management, the spool, as a common load-bearing structure, is widely used in the storage and transportation of materials such as metal wires, optical fibers, and cables. In the actual operation process, the handling and stacking of spools usually rely on forklifts or automated guided vehicles (AGVs) to complete. The traditional manual forklift operation mode requires a forklift to be driven manually to complete the grasping, handling, and stacking of spools, with manual judgment of the alignment accuracy, adjustment of the placement angle, and correction through continuous attempts. Due to the large mass of the spools and the reliance on visual observation for alignment by humans, problems such as deviation, tipping, or misaligned stacking are likely to occur, thus affecting the safety of warehousing and the operation efficiency. At the same time, due to the manual operation being restricted by the experience level, it is difficult to guarantee the stacking efficiency and accuracy of the spools, and the long-term manual handling work intensity is relatively large, which is not conducive to the automation development of enterprises.

[0003] With the development of intelligent manufacturing, AGV forklifts are widely used in automated warehousing systems to replace manual forklifts in performing the handling and stacking tasks of spools. However, the existing methods for AGV forklifts to stack spools still have many limitations. Most AGV forklifts use simple two-dimensional vision systems or inertial navigation systems for positioning, mainly relying on sensors such as QR code recognition and lidar to complete navigation and alignment. However, the two-dimensional vision system has great limitations in processing three-dimensional targets, and can only provide planar information, making it difficult to accurately identify the three-dimensional relative position between the spool and the base. In addition, since AGV forklifts usually rely on fixed motion path planning for operation, it is difficult to perform dynamic adjustment in a complex industrial environment. If there is a slight displacement deviation of the spool during handling, it may lead to the failure of the final stacking, and manual intervention is required for correction, which goes against the original intention of AGV automated operation.

[0004] Regarding the stacking problem of spools, some existing solutions use the method of combining lidar and depth cameras for recognition and alignment, but these methods usually rely on specific marker points or QR codes for alignment and are difficult to adapt to markerless scenarios. At the same time, due to the limited accuracy of lidar, the error in the three-dimensional pose estimation of the spool is relatively large, and it is easily affected by factors such as environmental light and object reflectivity, resulting in a decrease in the alignment accuracy. In addition, although conventional depth cameras can provide three-dimensional point cloud data, their ranging accuracy is limited and they are greatly affected by the environment, making it difficult to meet the high-precision alignment requirements of spools. Therefore, how to achieve high-precision and high-stability automatic stacking of spools without manual intervention remains an important challenge in the current technical field.

[0005] The existing method for stacking I-beam wheels by AGV forklifts still has the problem of insufficient optimization ability. In most application scenarios, the positioning and alignment of AGV forklifts rely on fixed rules, usually using simple PID control or adjustment strategies based on heuristic methods. During the stacking process, if the pose adjustment strategy of the AGV forklift is not precise enough, there may be a slight misalignment between the I-beam wheel and the base, and the existing system lacks an intelligent optimization mechanism to dynamically adjust the pose parameters of the forklift. In addition, traditional matching methods (such as the ICP algorithm) are easily affected by the initial values during the point cloud alignment process. If the initial matching deviation is large, it may fall into a local optimal solution, resulting in the inability to accurately align the I-beam wheels. Moreover, traditional optimization algorithms usually utilize insufficient global information during the calculation process and are difficult to efficiently and accurately find the optimal matching solution.

[0006] Therefore, how to provide a method for stacking I-beam wheels by applying stereo vision to AGV forklifts is an urgent problem that needs to be solved by those skilled in the art. Summary of the Invention

[0007] An object of the present invention is to propose a method for stacking I-beam wheels by applying stereo vision to AGV forklifts. The present invention significantly improves the stacking accuracy, pose adjustment accuracy and operation efficiency of the I-beam wheels, reduces the alignment error and the risk of stacking failure, and realizes the efficient automatic operation of the AGV forklift.

[0008] A method for stacking I-beam wheels by applying stereo vision to AGV forklifts according to an embodiment of the present invention includes the following steps:

[0009] S1. Use a TOF camera to obtain the three-dimensional point cloud data of the I-beam wheel and the base, and extract the first position information of the I-beam wheel and the second position information of the base;

[0010] S2. Calculate the matching parameters between the first position information of the I-beam wheel and the second position information of the base based on the whale optimization algorithm to optimize the matching error;

[0011] S3. Calculate the pose adjustment amount of the AGV forklift based on the matching result, and initialize the firefly population to optimize the forward movement adjustment amount, side movement adjustment amount and rotation adjustment amount;

[0012] S4. Use the firefly algorithm to optimize the pose adjustment scheme of the AGV forklift to align the I-beam wheel with the base;

[0013] S5. The AGV forklift executes the pose adjustment scheme to complete the precise stacking of the I-beam wheels, and optimizes the adjustment through a closed-loop feedback system.

[0014] Optionally, the S3 specifically includes:

[0015] S21. Define the I-shaped wheel point cloud data set and the base point cloud data set, and set P = {p i ∣p i ∈R 3 ,i=1,2,…,n}, where each p i is a three-dimensional point extracted from the first position information of the I-shaped wheel; let Q = {q i ∣q i ∈R 3 ,i=1,2,…,n}, where each q i is a three-dimensional point extracted from the second position information of the base;

[0016] S22. Define the candidate matching parameter X to belong to the SE(3) group. The SE(3) group represents a three-dimensional special Euclidean group that describes all rigid transformations in three-dimensional space, where X = (ω, T), ω is a rotation vector, T is a translation vector, and the rotation matrix C is given by the following formula:

[0017] C = exp([ω] × );

[0018] Among them, the antisymmetric matrix [ω] × Defined as:

[0019]

[0020] Among them, exp is the exponential mapping function, ω1, ω2, ω3 are the key parameters describing the rotation angle of the spool;

[0021] S23, construct a robust matching error function f(X), the expression is:

[0022]

[0023] Where i represents the index number in the point cloud, n represents the number of points in the point cloud dataset, the Euclidean norm is denoted as ‖·‖, the regularization parameter is λ, and the Huber loss function ρ(r) is defined as:

[0024]

[0025] δ is the preset threshold;

[0026] S24. Initialize the candidate matching parameter population of the whale optimization algorithm in the SE(3) space, denoted as

[0027]

[0028] in, represents the rotation vector of the jth candidate solution initialized by the whale optimization algorithm, It represents the translation vector of the j-th candidate solution for the initialization of the whale optimization algorithm;

[0029] S25. Within the t-th generation, for each candidate solution Update it using the whale optimization algorithm, and the update formula is:

[0030]

[0031] Δt j ∈[0,1];

[0032] Where, is the new matching parameter optimized by the whale optimization algorithm, X * (t) is the current optimal matching parameter found in the t-th generation iteration, is the Euclidean distance between the current candidate solution and the optimal solution, Δt j is a random variable, cos is the trigonometric cosine function, β is a preset constant, is the random perturbation vector, and the symbol "⊕" represents the combination operation within the SE(3) group;

[0033] S26. For each candidate solution Calculate the matching error

[0034] S27. Repeat steps S25 and S26 until the matching error meets the preset convergence condition;

[0035] S28. Output the candidate solution X * that makes the matching error function f(X) obtain the minimum value

[0036] Optionally, the specific content of S3 includes:

[0037] S31. Denote the optimal matching parameter as X * =(ω * ,T * ), where, T * =(T x ,T y ,T z ) T , ω * =(ω1,ω2,ω3) T , T x ,T y ,T z is the translation adjustment amount after the spool and the base are matched;

[0038] S32. Calculate the pose adjustment amount of the AGV forklift, and determine the forward adjustment amount Δx, the side shift adjustment amount Δy, and the rotation adjustment amount Δθ. The calculation formulas are respectively:

[0039] Δx = T x ;

[0040] Δy = T y ;

[0041]

[0042] S33. Initialize the population of the firefly optimization algorithm for optimizing the pose adjustment scheme of the AGV forklift. Denote the initial population as:

[0043]

[0044] Among them, is the initial adjustment amount for the pose adjustment optimization of the AGV forklift,

[0045]

[0046] ∈ x , ∈ y , ∈ θ is the preset perturbation range;

[0047] S34. Output the initial pose adjustment parameter Z0 = (Δx, Δy, Δθ) and the initialized firefly population

[0048] Optionally, the specific steps of S4 include:

[0049] S41. Use the initial population as the initial candidate solution for iteration;

[0050] S42. For each candidate solution calculate the fitness function:

[0051]

[0052] Among them, is the fitness function value of the i-th candidate solution in the t-th generation, are the adjustment amounts of the i-th firefly individual in the x, y, and z axis directions in the t-th generation respectively;

[0053] S43. For any candidate solution and calculate the Euclidean distance:

[0054]

[0055] is the Euclidean distance between the firefly individual i and the firefly individual j in the t-th generation iteration;

[0056] S44. Update each candidate solution using the firefly algorithm update formula. The update formula is:

[0057]

[0058] Among them, is the firefly optimization algorithm in the (t + 1)-th generation iteration, is the pose adjustment parameter after the update of the i-th individual in the (t - 1)-th generation iteration of the firefly optimization algorithm, exp is the exponential mapping function, μ is the momentum factor, β0 is the initial attractiveness constant, γ is the light absorption coefficient, λ is the time decay constant, t is the current iteration number, is the candidate solution and is the Euclidean distance between them, α0 is the initial random perturbation coefficient, δ ∈ (0, 1) is the random perturbation decay coefficient, is a random vector uniformly distributed in [0, 1];

[0059] S45. Calculate the fitness function for the updated candidate solution :

[0060]

[0061] S46. Repeat steps S42 to S45 until the preset iteration number or the fitness function satisfies the preset convergence condition, and output the candidate solution Z that makes the fitness function obtain the minimum value * =(Δx * , Δy * , Δθ * );

[0062] S47. Use the optimal candidate solution Z * =(Δx * , Δy * , Δθ * ) to determine the AGV forklift pose adjustment scheme, and the specific expression is:

[0063] Δx adj =Δx * , Δy adj =Δy * , Δθ adj =Δθ * ;

[0064] And update the pose parameter X new of the AGV forklift as:

[0065]

[0066] where X current is the current pose, and the symbol Represents the combination operation within the SE(3) group, and realizes the precise alignment of the spool and the base based on the updated pose parameters.

[0067] The beneficial effects of the present invention are as follows:

[0068] (1) A spool stacking method for AGV forklifts based on stereovision proposed by the present invention combines the 3D point cloud acquisition ability of the TOF camera, the global search ability of the whale optimization algorithm, and the dynamic adjustment ability of the firefly optimization algorithm, significantly improving the accuracy, stability, and adaptability of AGV forklifts during the automated stacking of spools. By obtaining high-precision 3D point cloud data of the spool and the base through the TOF camera, the pose information of the spool can be accurately measured in a complex industrial environment. Compared with traditional 2D vision and lidar solutions, this method reduces the influence of factors such as environmental illumination and object reflectivity on pose recognition and improves the matching accuracy.

[0069] (2) The present invention uses the whale optimization algorithm to optimize the matching parameters of the spool and the base. Compared with the traditional ICP (Iterative Closest Point) algorithm, this optimization method can globally search for the optimal matching solution in a larger search space, avoiding the local optimum problem and having stronger robustness to the initial matching error, significantly improving the alignment accuracy between the spool and the base. Based on this optimization strategy, even if the spool undergoes a slight deviation during the operation of the AGV forklift, the system can still accurately calculate the matching error and perform intelligent correction to ensure that the spool can be placed accurately on the base.

[0070] (3) The present invention further optimizes the pose adjustment amount of the AGV forklift through the firefly optimization algorithm, enabling the AGV forklift to adaptively adjust the forward movement amount, side movement amount, and rotation angle to achieve more precise spool stacking. The traditional pose adjustment method of AGV forklifts often relies on fixed path planning or simple proportional-integral-derivative (PID) control, lacking dynamic adaptive optimization ability and being difficult to cope with complex working conditions changes. The present invention utilizes the adaptive search ability of the firefly optimization algorithm, enabling the AGV forklift to adjust its own pose in real time during the stacking process, ensuring that the spool is always in the best stacking position, improving the stacking accuracy while reducing the need for multiple adjustments and enhancing the operation efficiency. Description of the Drawings

[0071] The drawings are used to provide further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention. In the drawings:

[0072] Figure 1 is a flowchart of a spool stacking method for AGV forklifts based on stereovision proposed by the present invention;

[0073] Figure 2 Schematic diagram of optimizing the matching of spools and bases using the whale optimization algorithm proposed by the present invention;

[0074] Figure 3 Schematic diagram of optimizing the pose adjustment of the AGV forklift using the firefly optimization algorithm proposed by the present invention. Specific implementation manner

[0075] The present invention will now be further described in detail with reference to the accompanying drawings. These drawings are all simplified schematic diagrams, only illustrating the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0076] Refer to Figures 1 to 3 , a method for stacking spools applied to an AGV forklift based on stereo vision, comprising the following steps:

[0077] S1. Use a TOF camera to obtain the three-dimensional point cloud data of the spool and the base, and extract the first position information of the spool and the second position information of the base;

[0078] S2. Calculate the matching parameters of the first position information of the spool and the second position information of the base based on the whale optimization algorithm to optimize the matching error;

[0079] S3. Calculate the pose adjustment amount of the AGV forklift based on the matching result, and initialize the firefly population to optimize the forward movement adjustment amount, side movement adjustment amount, and rotation adjustment amount;

[0080] S4. Use the firefly algorithm to optimize the pose adjustment scheme of the AGV forklift to align the spool with the base;

[0081] S5. The AGV forklift executes the pose adjustment scheme to complete the precise stacking of the spools, and optimizes the adjustment through a closed-loop feedback system.

[0082] In this embodiment, S3 specifically includes:

[0083] S21. Define the spool point cloud data set and the base point cloud data set. Let P = {p i ∣p i ∈R 3 , i = 1, 2,..., n}, where each p i is a three-dimensional point extracted from the first position information of the spool; let Q = {q i ∣q i ∈R 3 , i = 1, 2,..., n}, where each q i is a three-dimensional point extracted from the second position information of the base;

[0084] S22. Define that the candidate matching parameter X belongs to the SE(3) group, where SE(3) represents the special Euclidean group in three dimensions, describing all rigid transformations in three-dimensional space. Here, X = (ω, T), ω is the rotation vector, and T is the translation vector. The rotation matrix C is given by:

[0085] C = exp([ω] × );

[0086] where the skew-symmetric matrix [ω] × is defined as:

[0087]

[0088] where exp is the exponential mapping function, and ω1, ω2, ω3 are the key parameters describing the rotation angle of the spool;

[0089] S23. Construct a robust matching error function f(X), and the expression is:

[0090]

[0091] where i represents the index number in the point cloud, n represents the number of points in the point cloud dataset, the Euclidean norm is denoted as ‖·‖, the regularization parameter is λ, and the Huber loss function ρ(r) is defined as:

[0092]

[0093] δ is a preset threshold;

[0094] S24. Initialize the candidate matching parameter population of the whale optimization algorithm in the SE(3) space, denoted as

[0095]

[0096] where represents the rotation vector of the j-th candidate solution initialized by the whale optimization algorithm, represents the translation vector of the j-th candidate solution initialized by the whale optimization algorithm;

[0097] S25. In the t-th generation, update each candidate solution using the whale optimization algorithm, and the update formula is:

[0098]

[0099] Δt j ∈[0, 1];

[0100] where is the new matching parameter optimized by the whale optimization algorithm, X *(t) is the current optimal matching parameter found in the t-th generation iteration, is the Euclidean distance between the current candidate solution and the optimal solution, Δt j is a random variable, cos is the trigonometric cosine function, and β is a preset constant, is the random perturbation vector, and the symbol "⊕" represents the combination operation within the SE(3) group;

[0101] S26. For each candidate solution Calculate the matching error

[0102] S27. Repeat steps S25 and S26 until the matching error meets the preset convergence condition;

[0103] S28. Output the candidate solution X that minimizes the matching error function f(X) * as the optimal matching parameter.

[0104] In this embodiment, the specific steps of S3 include:

[0105] S31. Denote the optimal matching parameter as X * =(ω * , T * ), where T * =(T x , T y , T z ) T , ω * =(ω1, ω2, ω3) T , T x , T y , T z is the translation adjustment amount after the spool and the base are matched;

[0106] S32. Calculate the pose adjustment amount of the AGV forklift, and determine the forward movement adjustment amount Δx, the side movement adjustment amount Δy, and the rotation adjustment amount Δθ. The calculation formulas are as follows:

[0107] Δx = T x ;

[0108] Δy = T y ;

[0109]

[0110] S33. Initialize the population of the firefly optimization algorithm for optimizing the pose adjustment scheme of the AGV forklift. Denote the initial population as:

[0111]

[0112] Among them, is the initial adjustment amount for the pose adjustment and optimization of the AGV forklift position,

[0113]

[0114] ∈ x , ∈ y , ∈ θ is the preset perturbation range;

[0115] S34. Output the initial pose adjustment parameter Z0 = (Δx, Δy, Δθ) and the initialized firefly population

[0116] In this embodiment, the S4 specifically includes:

[0117] S41. Use the initial population as the initial candidate solution for iteration;

[0118] S42. For each candidate solution calculate the fitness function:

[0119]

[0120] where, is the fitness function value of the i-th candidate solution in the t-th generation, are the adjustment amounts of the i-th firefly individual in the x, y, and z axis directions in the t-th generation respectively;

[0121] S43. For any candidate solution and calculate the Euclidean distance:

[0122]

[0123] is the Euclidean distance between the firefly individual i and the firefly individual j in the t-th generation iteration;

[0124] S44. Update each candidate solution using the firefly algorithm update formula, and the update formula is:

[0125]

[0126] where, is in the (t + 1)-th generation iteration of the firefly optimization algorithm, is the pose adjustment parameter after the update of the i-th individual in the (t - 1)-th generation iteration of the firefly optimization algorithm, exp is the exponential mapping function, μ is the momentum factor, β0 is the initial attraction constant, γ is the light absorption coefficient, λ is the time decay constant, t is the current iteration number, is the candidate solution and The Euclidean distance between them, α0 is the initial random perturbation coefficient, and δ ∈ (0, 1) is the random perturbation decay coefficient. is a random vector uniformly distributed within [0, 1];

[0127] S45. Calculate the fitness function for the updated candidate solution :

[0128]

[0129] S46. Repeat steps S42 to S45 until the preset number of iterations is reached or the fitness function satisfies the preset convergence condition, and output the candidate solution Z that minimizes the fitness function * =(Δx * , Δy * , Δθ * );

[0130] S47. Use the optimal candidate solution Z * =(Δx * , Δy * , Δθ * ) to determine the AGV forklift position and pose adjustment plan. The specific expression is:

[0131] Δx adj =Δx * , Δy adj =Δy * , Δθ adj =Δθ * ;

[0132] And update the pose parameters X of the AGV forklift new as:

[0133]

[0134] where X current is the current pose, and the symbol represents the combined operation within the SE(3) group. Based on the updated pose parameters, the accurate alignment of the spool and the base is achieved.

[0135] Example:

[0136] To verify the feasibility of the present invention in implementation, the present invention is applied to a large steel manufacturing plant. Among them, the spool is one of the key raw materials and is used for the storage and transportation of products such as steel wires and cables. The plant covers an area of more than 50,000 square meters. During the production process, thousands of spools need to be stacked every day. Each spool weighs between 1.2 tons and 1.5 tons. The traditional manual or semi-automatic stacking methods not only take a long time and have low efficiency, but also often cause stacking deviations due to human judgment errors, resulting in relatively large safety hazards. To improve the efficiency and accuracy of spool stacking and reduce safety risks, the present invention adopts a method for spool stacking by an AGV forklift based on stereo vision. The three-dimensional point cloud data of the spool and the base is obtained in real time through a TOF camera, and then the collected data is globally matched and optimized using the whale optimization algorithm. Subsequently, the firefly optimization algorithm is used to optimize the pose adjustment of the AGV forklift to ensure precise alignment between the spool and the base, thereby realizing automatic stacking.

[0137] In practical applications, a dedicated automated stacking workshop is set up in the plant area, with a flat ground and a preset AGV driving path laid. The ambient light is reasonably regulated to ensure the stability of TOF camera data acquisition. During the implementation process, after the AGV forklift transports the spool to the stacking area, the TOF camera simultaneously collects the three-dimensional point cloud data of the spool and the stacking base. The data acquisition frequency can reach 30 frames per second, and the acquisition range accuracy reaches the millimeter level. Through data preprocessing, the collected point cloud information is transmitted to the embedded computing unit, and then global matching is realized relying on the whale optimization algorithm. The matching process is completed within 10 seconds, and the error is controlled within 0.8 millimeters. Subsequently, according to the matching result, the system calculates the forward movement, side movement, and rotation adjustment amounts required by the AGV forklift, and initializes the population of the firefly optimization algorithm. The initial population size is set to 50. After about 15 iterations, the pose adjustment scheme optimization time is within 5 seconds, and the convergence error of the final adjustment parameters is less than 0.5 millimeters, and the rotation angle error is less than 0.2 degrees. The average total time for the entire stacking process from the spool transportation to the final stacking completion is shortened to 45 seconds, which is more than 60% shorter than the average time of 120 seconds for traditional manual stacking.

[0138] To further verify the beneficial effects of the present invention, key indicators of the traditional method and the method of the present invention were compared and tested in actual applications. The test site was set in the automated stacking workshop within the factory area, and the test time was one month continuously, with an average of 50 stacking operations per day. The data showed that after adopting the method of the present invention, the matching success rate of spools increased from 85% of the traditional method to 98%, the pose adjustment accuracy decreased from the traditional error of 3 mm to within 0.5 mm, the rotation alignment angle error decreased from 2 degrees to within 0.2 degrees, the average time consumed for each stacking operation was shortened from the original 120 seconds to 45 seconds, and the number of times the AGV forklift needed to be readjusted due to adjustment failure decreased from an average of 0.8 times per operation to 0.1 times. In addition, the closed-loop feedback mechanism effectively reduced the cumulative error in the subsequent stacking process, and the overall vehicle operation stability increased by nearly 90%.

[0139] Table 1: Comparison Table of Key Indicators between Traditional Method and Method of the Present Invention

[0140] Index Traditional method Method of the present invention Matching error 3.2 mm 0.8 mm Pose adjustment error 2.9 mm 0.5 mm Rotation angle error 2.1 degrees 0.2 degrees Time consumption for each stacking operation 120 seconds 45 seconds Number of adjustments 0.8 times 0.1 times

[0141] Table 1 "Comparison Table of Key Indicators between Traditional Method and Method of the Present Invention" shows that in terms of matching error, the average error of the traditional method is 3.2 mm, while that of the method of the present invention is only 0.8 mm; in terms of pose adjustment error, the average adjustment error of the traditional method reaches 2.9 mm, and the method of the present invention is controlled within 0.5 mm; in terms of rotation angle error, the average error of the traditional method is about 2.1 degrees, while that of the method of the present invention is less than 0.2 degrees; in terms of the time consumed for each stacking operation, the average time consumed by the traditional method is 120 seconds, while the method of the present invention only needs 45 seconds; and in terms of the number of adjustments, the traditional method needs to be readjusted an average of 0.8 times per operation, while the method of the present invention only needs 0.1 times. It can be seen that the method of the present invention is significantly superior to the traditional method in terms of stacking efficiency and accuracy, and greatly improves the operation safety.

[0142] Through the above embodiments, it can be seen that the present invention not only effectively solves the problem of stacking failure caused by manual errors, inaccurate matching, and untimely pose adjustment during the spool stacking process in actual applications, but also significantly improves the automatic stacking efficiency and stability. The application prospect of this technology in large-scale industrial production and warehousing logistics management is broad, which can achieve unmanned operation, reduce labor intensity, and improve the production efficiency and safety management level of enterprises.

[0143] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. An I - shaped wheel stacking method based on stereo vision and applied to AGV forklifts, characterized in that, The steps are as follows: S1. Use a TOF camera to obtain the three-dimensional point cloud data of the spool and the base, and extract the first position information of the spool and the second position information of the base; S2. Calculate the matching parameters between the first position information of the spool and the second position information of the base based on the whale optimization algorithm to optimize the matching error; S3. Calculate the pose adjustment amount of the AGV forklift based on the matching result, and initialize the firefly population to optimize the forward movement adjustment amount, lateral movement adjustment amount, and rotation adjustment amount; S4. Use the firefly algorithm to optimize the pose adjustment plan of the AGV forklift to align the spool with the base; S5. The AGV forklift executes the pose adjustment plan to complete the precise stacking of the spools, and optimizes the adjustment through a closed-loop feedback system.

2. The method for stacking spools according to claim 1, which is applied to an AGV forklift based on stereo vision, is characterized in that, The specific content of S3 includes: S21. Define the spool point cloud data set and the base point cloud data set. Let \(P = \{p i |p i \in\mathbb{R} 3 , i = 1, 2, \ldots, n\}\), where each \(p i \) is a three-dimensional point extracted from the first position information of the spool; Let \(Q = \{q i |q i \in\mathbb{R} 3 , i = 1, 2, \ldots, n\}\), where each \(q i \) is a three-dimensional point extracted from the second position information of the base; S22. Define that the candidate matching parameter X belongs to the SE(3) group. The SE(3) group represents the three-dimensional special Euclidean group, which describes all rigid transformations in three-dimensional space, where X = (ω, T), ω is the rotation vector, T is the translation vector, and the rotation matrix C is given by the following formula: C = exp([ω] × ); Among them, the skew-symmetric matrix [ω] × is defined as: where exp is the exponential mapping function, and ω1, ω2, ω3 are the key parameters describing the rotation angle of the spool; S23. Construct a robust matching error function f(X), and the expression is: where i represents the index number in the point cloud, n represents the number of points in the point cloud dataset, the Euclidean norm is denoted as ‖·‖, the regularization parameter is λ, and the Huber loss function ρ(r) is defined as: δ is a preset threshold; S24. Initialize the population of candidate matching parameters of the whale optimization algorithm in the SE(3) space, denoted as Among them, represents the rotation vector of the j-th candidate solution for the initialization of the whale optimization algorithm, represents the translation vector of the j-th candidate solution for the initialization of the whale optimization algorithm; S25. In the t-th generation, for each candidate solution update it using the whale optimization algorithm, and the update formula is: Δt j ∈[0,1]; Among them, is the new matching parameter optimized by the whale optimization algorithm, and X * (t) is the current optimal matching parameter found in the t-th generation of iteration, is the Euclidean distance between the current candidate solution and the optimal solution, and Δt j is a random variable, cos is the trigonometric cosine function, and β is a preset constant, is the random perturbation vector, and the symbol represents the combination operation within the SE(3) group; S26. For each candidate solution Calculate the matching error S27. Repeat steps S25 and S26 until the matching error meets the preset convergence condition; S28. Output the candidate solution X that minimizes the matching error function f(X). * As the optimal matching parameter.

3. A method for stacking spools based on stereoscopic vision applied to an AGV forklift according to claim 1, characterized in that, The specific content of S3 includes: S31. The optimal matching parameter is denoted as X * =(ω * , T * ), where T * =(T x , T y , T z ), T ω * =(ω1, ω2, ω3) T , T x , T y , T z is the translation adjustment amount after the spool and the base are matched; S32. Calculate the pose adjustment amount of the AGV forklift, and determine the forward movement adjustment amount Δx, lateral movement adjustment amount Δy, and rotation adjustment amount Δθ. The calculation formulas are as follows: Δx = T x ; Δy = T y ; S33. Initialize the population of the firefly optimization algorithm to optimize the pose adjustment plan of the AGV forklift. Denote the initial population as: Among them, is the initial adjustment amount for the pose adjustment and optimization of the AGV forklift position, ∈ x , ∈ y , ∈ θ is the preset disturbance range; S34. Output the initial pose adjustment parameter Z0 = (Δx, Δy, Δθ) and initialize the firefly population 4. A method for stacking spools based on stereoscopic vision applied to an AGV forklift according to claim 1, characterized in that, The specific content of S4 includes: S41. Use the initial population as the initial candidate solution for iteration; S42. For each candidate solution Calculate the fitness function: Among them, is the fitness function value of the $i$-th candidate solution in the $t$-th generation, are the adjustment amounts of the $i$-th firefly individual in the $x$, $y$, and $z$ axis directions in the $t$-th generation, respectively; S43. For any candidate solution and calculate the Euclidean distance: is the Euclidean distance between firefly individual i and firefly individual j in the t-th generation iteration; S44. Use the firefly algorithm update formula to update each candidate solution. The update formula is: Among them, For the firefly optimization algorithm in the (t + 1)-th generation iteration, is the pose adjustment parameter after the update of the i-th individual in the (t - 1)-th generation iteration of the firefly optimization algorithm, exp is the exponential mapping function, μ is the momentum factor, β0 is the initial attractiveness constant, γ is the light absorption coefficient, λ is the time decay constant, and t is the current iteration number. is the candidate solution and is the Euclidean distance between them, α0 is the initial random perturbation coefficient, δ ∈ (0, 1) is the random perturbation decay coefficient, is a random vector uniformly distributed in [0, 1]; S45. For the updated candidate solution Calculate the fitness function: S46. Repeat the steps from S42 to S45 until a preset number of iterations is reached or the fitness function meets the preset convergence condition, and output the candidate solution Z that minimizes the fitness function * =(Δx * , Δy * , Δθ * ); S47. Utilize the optimal candidate solution Z * =(Δx * , Δy * , Δθ * ), to determine the AGV forklift pose adjustment plan, and the specific expression is: Δx adj = Δx * , Δy adj = Δy * , Δθ adj = Δθ * ; And update the pose parameter X of the AGV forklift new as follows: Among them, X current is the current pose, and the symbol represents the combined operation within the SE(3) group, and realizes the precise alignment of the spool and the base based on the updated pose parameters.

Images