Visual sorting method based on Delta robot

Abstract

According to the visual sorting method based on the Delta robot, the position of a moving target is recognized in real time through a visual system, the coordinates of a grabbing point are dynamically corrected in combination with the movement speed of a conveying belt, and the tail end of the robot and the target synchronously arrive; planning an n-shaped track for connecting the tail end of the robot with a grabbing point, and optimizing at a right-angle turning section by adopting a transition curve with continuous curvature so as to avoid sudden change of acceleration; a high-order continuous acceleration control algorithm is adopted to drive the robot to move so as to ensure that the speed and the acceleration are continuous and smooth and eliminate mechanical impact. Robot control based on a visual sorting system can be effectively achieved, and accurate recognition, dynamic positioning, reasonable path planning and stable and dynamic sorting of sorted objects are achieved. And the sorting efficiency of the sorting system and the stability of the sorting process are improved. The control method is simple and convenient and has a good application prospect.

Description

Technical Field

[0001] The present invention belongs to the technical fields such as robot control, and particularly relates to a vision sorting method based on a Delta robot. Background Art

[0002] Delta parallel robots have important application values in light industry sorting scenarios (such as food and electronic component sorting) due to their high speed and high precision characteristics. However, in dynamic sorting scenarios, especially in the target grasping task where the conveyor belt moves continuously, the existing technical solutions still have significant deficiencies. Traditional Delta robot systems usually rely on preset paths or manual teaching to complete grasping tasks, and their trajectory planning and motion control logics lack adaptability in dynamic scenarios. For example, conventional vision sorting systems are mostly designed for static targets and do not fully consider the influence of conveyor belt movement on the position of the grasping point, resulting in insufficient dynamic compensation ability. When the end effector of the robot moves at high speed, if the traditional "door" - shaped trajectory planning is adopted, mechanical vibrations are likely to be caused by sudden changes in acceleration at the right - angled turning points, affecting the sorting stability. In addition, existing motion control algorithms (such as trapezoidal acceleration - deceleration and PID control) are difficult to achieve acceleration continuity, and the end - jitter problem is particularly prominent in high - speed scenarios.

[0003] In the field of trajectory optimization, although existing methods attempt to improve smoothness by replacing right - angled segments with arcs or simple curves, their curvatures are fixed and lack dynamic adjustment capabilities, making it difficult to adapt to the speed and precision requirements of different sorting scenarios. For example, although arc transitions can reduce impacts, they cannot flexibly adjust the parameters of the transition segment according to the actual trajectory length and lifting height, resulting in limited adaptability. At the same time, there are still technical bottlenecks in the collaborative design of the vision system and the robot control module: the problems of hand - eye calibration error and mis - grasping caused by repeated target recognition have not been effectively solved. Although some studies attempt to compensate the target position through Kalman filtering or linear prediction, these methods do not deeply integrate the real - time coupling relationship between the robot kinematic model and the dynamic scenario, and there are still problems of cumulative synchronization errors in practical applications. Therefore, there is an urgent need for a sorting solution that integrates dynamic vision compensation, trajectory smoothing optimization, and high - order motion control algorithms to improve the stability and efficiency of Delta robots in high - speed dynamic scenarios. Summary of the Invention

[0004] In view of the deficiencies in the existing technology, such as insufficient dynamic target tracking accuracy, mechanical vibration caused by sudden acceleration changes at the turning points of the trajectory, and poor stability of the motion control algorithm, the present invention provides a vision sorting method based on a Delta parallel robot. Through a vision dynamic sorting compensation mechanism, the coordinates of the grasping point are corrected in real time to achieve synchronous positioning of the robot end and the moving target. The "door"-shaped trajectory is modified for curvature continuity by using an elliptical transition trajectory optimization technique to eliminate sudden acceleration changes. In combination with a 3-4-5th polynomial acceleration and deceleration control algorithm, the speed and acceleration of the end motion are made continuous and smooth. At the same time, a vision duplicate removal mechanism with a preset distance threshold is used to avoid repeated grasping, and a deep learning model trained with a small learning rate and a large amount of data is used to improve the recognition accuracy of dynamic scenes. Finally, a collaborative technical solution integrating dynamic vision positioning, trajectory smoothing optimization, and high-order motion control is formed, significantly improving the grasping efficiency and system stability in high-speed sorting scenarios.

[0005] The technical solution specifically adopted by the present invention to solve its technical problems is as follows:

[0006] A vision sorting method based on a Delta robot:

[0007] The position of the moving target is recognized in real time through a vision system, and the coordinates of the grasping point are dynamically corrected in combination with the conveyor belt movement speed, so that the robot end reaches synchronously with the target;

[0008] Plan the "door"-shaped trajectory connecting the robot end and the grasping point, and optimize it with a curvature-continuous transition curve at the right-angle turning section to avoid sudden acceleration changes;

[0009] Adopt a high-order continuous acceleration control algorithm to drive the robot to move, so as to ensure that the speed and acceleration are continuous and smooth and eliminate mechanical shock.

[0010] Further, the dynamic correction of the grasping point coordinates includes:

[0011] The time required for the robot to move from the current position to the initial position of the target is calculated in real time and recorded as the grasping time;

[0012] According to the product of the grasping time and the conveyor belt running speed, calculate the moving distance of the conveyor belt during the grasping period, and record it as the compensation distance;

[0013] The compensation distance is deducted from the horizontal coordinate of the initial position of the target in real time to generate the dynamically updated position of the grasping point, ensuring that the robot end reaches synchronously with the moving target.

[0014] Further, the curvature-continuous transition curve is an elliptical curve, and its horizontal length and vertical height are dynamically adjusted in the following manner:

[0015] Dynamically generate the length of the horizontal semi-axis based on the horizontal projection length of the trajectory according to a preset horizontal transition ratio;

[0016] Dynamically generate the height of the vertical semi-axis based on a preset lifting height according to a vertical transition ratio;

[0017] Control the curvature continuity of the elliptical curve through the coordinated adjustment of the horizontal semi-axis and the vertical semi-axis;

[0018] Among them, the interpolation point coordinates of the elliptical transition section are realized through the coordinate transformation from the two-dimensional plane coordinate system to the robot's three-dimensional coordinate system.

[0019] Further, the coordinate transformation from the two-dimensional plane coordinate system to the robot's three-dimensional coordinate system is realized through the following steps:

[0020] Construct a rotation matrix according to the transition section direction angle θ;

[0021] Based on the two-dimensional interpolation point coordinates and the rotation matrix, calculate the three-dimensional absolute coordinates in the robot coordinate system.

[0022] Further, the high-order continuous acceleration control algorithm adopts a fifth-order polynomial motion law and satisfies the following conditions:

[0023] At the starting moment of motion when δ = 0 and the ending moment when δ = 1, the position, velocity, and acceleration are all zero;

[0024] During the motion process, the acceleration curve is continuous and has no mutation;

[0025] Dynamically adjust the motion trajectory through the normalized time parameter δ to ensure the smooth transition of the robot end velocity and acceleration.

[0026] Further, the method for removing duplicates in real-time recognition by the vision system includes:

[0027] Traverse and compare the coordinates of the recognized objects in consecutive key frames;

[0028] If the coordinate deviation of the same object in adjacent frames is less than a preset threshold, it is determined as a duplicate target and ignored;

[0029] Otherwise, add the object coordinates to the list to be grabbed to avoid repeated execution of the grabbing action.

[0030] Further, the vision system realizes target recognition in the following way:

[0031] Adopt a target detection model based on deep learning to perform real-time recognition and positioning of the objects on the conveyor belt;

[0032] Train the model with a small learning rate ≤ 0.005 and a large batch size ≥ 128 to adapt to the real-time requirements of the high-speed sorting scenario;

[0033] Output object category, position and confidence information for dynamic grasping point calculation.

[0034] Furthermore, the vision system optimizes and calculates the homogeneous transformation matrix through multi-point calibration, and converts the object coordinates in the conveyor belt coordinate system into the robot absolute coordinate system.

[0035] And, a vision sorting system based on a Delta parallel robot, comprising:

[0036] Vision processing module: used to identify the position of moving targets on the conveyor belt in real time, and dynamically correct the grasping point coordinates based on the conveyor belt movement speed;

[0037] Trajectory planning module: used to generate a "door"-shaped trajectory connecting the robot end and the grasping point, and optimize the curvature continuity of the right-angle turning section through an elliptic curve;

[0038] Motion control module: used to execute a high-order continuous acceleration control algorithm to drive the parallel robot to achieve smooth motion;

[0039] Communication and execution module: used to transmit control instructions to the robot drive unit and synchronize the actions of the end effector.

[0040] Furthermore, the vision processing module includes:

[0041] Deep learning object detection unit, adopting the pre-trained YOLOv5 model;

[0042] Coordinate conversion unit, realizing the mapping from image coordinates to robot absolute coordinates through hand-eye calibration and homogeneous transformation matrix;

[0043] Dynamic compensation unit, calculating the real-time compensation distance based on the grasping time and the conveyor belt speed.

[0044] And, an electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned method when executing the program.

[0045] A non-transitory computer-readable storage medium, on which a computer program is stored, and the computer program implements the steps of the above-mentioned method when executed by a processor.

[0046] Compared with the prior art, the present invention and its preferred solutions at least include the following beneficial effects:

[0047] Improved dynamic sorting synchronization: By means of the vision dynamic compensation mechanism, the grasping point position is corrected in real time, effectively solving the problem of grasping lag caused by target movement in the conveyor belt scenario, and significantly improving the synchronization and grasping success rate of dynamic sorting;

[0048] Trajectory smoothness optimization: An elliptical transition curve is used to replace the right-angle turning section of the traditional "door"-shaped trajectory, and the sudden change in acceleration is eliminated through curvature continuity design to reduce the impact of mechanical vibration on sorting stability;

[0049] Enhanced motion control stability: Based on a high-order continuous acceleration control algorithm (such as 3-4-5th order polynomials), smooth acceleration and deceleration of the end motion are achieved, avoiding end jitter caused by discontinuous acceleration in traditional trapezoidal or PID control;

[0050] Improved visual recognition efficiency: A duplicate removal mechanism combined with a preset distance threshold is used to reduce repeated grasping operations, and the target recognition accuracy and real-time performance are improved through a deep learning model training strategy adapted to dynamic scenarios;

[0051] Strengthened system coordination: Through the deep integration of visual positioning, trajectory planning, and motion control modules, a closed-loop technical solution that dynamically adapts to high-speed sorting scenarios is formed, taking into account the dual requirements of efficiency and stability. Description of the Drawings

[0052] The present invention will be further described in detail below with reference to the drawings and specific embodiments:

[0053] Figure 1 is the overall flowchart of the embodiments of the present invention;

[0054] Figure 2 is a schematic diagram of the relationship between the robot coordinate system and the conveyor belt coordinate system in the hand-eye calibration shown in the embodiments of the present invention;

[0055] Figure 3 is the workflow diagram of duplicate result removal shown in the embodiments of the present invention;

[0056] Figure 4 is the schematic diagram of dynamic sorting shown in the embodiments of the present invention;

[0057] Figure 5 is the schematic diagram of the "door"-shaped trajectory shown in the embodiments of the present invention;

[0058] Figure 6 is the network structure diagram of YOLOV5 in the embodiments of the present invention;

[0059] Figure 7 is the schematic diagram of the test platform in the embodiments of the present invention. Detailed Description of the Invention

[0060] To make the features and advantages of the present invention more obvious and understandable, specific embodiments are given below and described in detail as follows:

[0061] An embodiment of the present invention provides a vision sorting method based on a Delta parallel robot. First, the forward and inverse kinematic solutions are obtained through the kinematic analysis of the robot, and the workspace of the robot is calculated through the forward and inverse kinematic solutions. Then, a suitable workspace is selected to limit the manipulation angle of the robot's motor. Images of the grasped objects are collected to train the vision sorting model. The real-time position of the grasped object relative to the end of the robot is calculated by a vision sorting system based on a deep learning vision algorithm. The coordinates of the object are updated in real time, and the absolute position of the grasping point is calculated. Finally, through the "door"-shaped trajectory planning and the 3-4-5 polynomial for acceleration and deceleration control, the movement of the end is controlled to achieve dynamic sorting. The implementation process of the present invention is simple. Through the "door"-shaped trajectory planning scheme, the efficiency and stability of the sorting process of the Delta parallel robot are significantly improved, which to a certain extent improves the disadvantages of the relatively inefficient sorting path and unstable movement process of traditional vision robots. At the same time, it also provides a reference for the operation of other types of robots.

[0062] It can effectively realize the control of the robot based on the vision sorting system, achieve the accurate recognition, dynamic positioning, reasonable path planning and stable dynamic sorting of the sorted objects. The sorting efficiency of the sorting system and the stability of the sorting process are improved. The control method of the present invention is simple and has good application prospects.

[0063] The overall construction and design process of its solution includes the following steps:

[0064] Step S1: Conduct kinematic analysis on the robot and obtain the forward and inverse kinematic solutions of the robot through geometric relationships;

[0065] Step S2: Obtain the workspace range of the robot through the forward and inverse kinematic solutions, and then obtain the actual workspace of the robot according to actual needs;

[0066] Step S3: Obtain the rotation angle range of the motor according to the actual workspace of the robot;

[0067] Step S4: Perform perspective transformation on the image of the recognition area to remove some redundant images;

[0068] Step S5: Perform hand-eye calibration on the image after perspective transformation;

[0069] Step S6: Write a vision recognition model;

[0070] Step S7: Collect images of the recognized objects to form a training set, and train and test the model;

[0071] Step S8: Preprocess and postprocess the image, and then use the vision recognition model to obtain the absolute position of the key frame of the object;

[0072] Step S9: Calculate the real-time update of the object position and the grasping point position by calculating the absolute position of the key frame of the object;

[0073] Step S10: Implement trajectory planning for the robot through a "gate"-shaped trajectory;

[0074] Step S11: Optimize the "gate"-shaped trajectory through an elliptical transition trajectory;

[0075] Step S12: Use a 3-4-5 polynomial to implement acceleration and deceleration control for the optimized "gate"-shaped trajectory.

[0076] As a preferred solution of this embodiment, in step S3, hand-eye calibration is performed on the perspective-transformed image, and the visual recognition system can obtain the absolute coordinates of the object. Specifically, according to the image coordinates (x 2 , y 2 ) of the object, its image homogeneous coordinates N can be known, and N can be expressed as:

[0077]

[0078] Assume that the thicknesses of different types of objects are b i (where i = 1, 2, 3), then the coordinates of the center point on the surface of the object in the conveyor belt coordinate system can be calculated according to the linear mapping matrix K, and K can be expressed as:

[0079]

[0080] According to the above image homogeneous coordinates N and the linear mapping matrix K, the coordinates M of the object in the conveyor belt coordinate system can be obtained, and M can be expressed as:

[0081]

[0082] Finally, the coordinates are converted into absolute coordinates through a homogeneous transformation matrix. The homogeneous transformation matrix is optimized and calculated from the corresponding coordinates of multiple points. Assume that t points are selected for the calculation of the homogeneous transformation matrix, and the coordinates of each point in the conveyor belt coordinate system are Q t , and the absolute coordinates of each point are P t . By optimizing and calculating the corresponding relationship of multiple points through the following formula, the homogeneous transformation matrix T between the two coordinates can be calculated.

[0083] P t = TQ t

[0084] The homogeneous transformation matrix T obtained by optimizing and calculating with multiple points is:

[0085]

[0086] According to the homogeneous transformation matrix T, the coordinates M of the object in the conveyor belt coordinate system can be converted into absolute coordinates G, and G can be expressed as

[0087] G = TM.

[0088] As a preferred solution of this embodiment, for the method of removing duplicate results in the image preprocessing and postprocessing of step S8, the process is to first identify the current key-frame image. Assuming that there are n recognition results in total, where i ∈ 1, 2, 3... n, and then perform coordinate transformation on it to obtain the absolute coordinates (X i , Y i , Z i ). At the same time, assume that there are m results in the to-be-grabbed list, where j ∈ 1, 2, 3... n, and its absolute coordinates are (A i , B i , C i ). Finally, traverse and compare (X i , Y i , Z i ) with (A i , B i , C i ). If the coordinate distances of two objects are very close, then ignore the duplicate result. Otherwise, add the absolute coordinates of the object to the to-be-grabbed list. Then repeat the above process and continue to identify the next key-frame image.

[0089] Generally, there are certain errors in the visual recognition results of the front and rear key-frame images, and there are also certain errors in the absolute coordinates of the same object recognized twice before and after. Therefore, as long as the coordinate distances of two points are very close, it can be considered that these two results correspond to the same object. The coordinate discrimination formula is:

[0090]

[0091] where ε is the specified distance, preferably preset ε = 2, unit mm.

[0092] As a preferred solution of this embodiment, for step S9 of real-time updating the item position and calculating the grasping point position to achieve precise grasping of the item, specifically, according to the robot coordinate P robot and the object coordinate P object , perform the grasping trajectory planning, and then calculate the distance S init of this path. Then, according to the expected average speed V robot , the time T robot required to move from the robot coordinate P object to the object coordinate P grab can be calculated. Take T grab as the grasping time, which can be expressed as:

[0093]

[0094] According to the grasping time T grab and the conveyor belt movement speed V belt Calculate the distance S that the conveyor belt moves within this time period belt , S belt Can be expressed as:

[0095] S belt = T grab ×V belt

[0096] Then use the conveyor belt movement speed V belt Calculate the grasping point position P where the object arrives after the grasping time grab , P grab Can be expressed as:

[0097] P grab = (X - S belt , Y, Z)

[0098] Finally, according to the grasping point position P grab and the robot position P robot Calculate interpolation, and the object and the robot end can reach the grasping point simultaneously

[0099] The actual average speed of the robot end is V move , and the distance S of the actual grasping path can be calculated according to the grasping point position P grab and the robot position P robot Finally, combined with the grasping time T grab Calculate the actual average speed V grab , V move , V move Can be expressed as:

[0100]

[0101] As a preferred solution of this embodiment, the formulas for the "door"-shaped trajectory planning and elliptical transition trajectory optimization in steps S10 and S11 are specifically: When the robot receives the grasping instruction, it obtains the position of the end effector and the position of the grasping point, connects the position of the end effector and the grasping point through the "door"-shaped trajectory, and then optimizes the "door"-shaped trajectory through the elliptical transition trajectory, replacing the right-angle position of the "door"-shaped with an elliptical trajectory

[0102] The improved trajectory path is composed of five parts as a whole. The trajectory path of the robot end is divided into five parts: the P1P2 vertical section, the P2P3 transition section, the P3P4 horizontal section, the P4P5 transition section, and the P5P6 vertical section. Among them, the coordinates of point P i are the coordinates in the robot coordinate system (for the convenience of narration, hereinafter referred to as absolute coordinates), and are expressed as:

[0103]

[0104] The absolute coordinates of points P1 and P6 are the starting and ending points of the robot's end motion trajectory respectively. The transition section of the trajectory path is controlled by two parameters: the horizontal transition ratio ρ h and the vertical transition ratio ρ v Two parameters are used for control.

[0105] The above elliptical transition trajectory contains two parameters: the vertical semi-axis h and the horizontal semi-axis l. h and l can be expressed as:

[0106] l = 0.5L×ρ h

[0107] h = H×ρ v

[0108] Among them, H is the specified lifting height of the trajectory, and L is the length of the line segment P1P6 projected onto the horizontal plane. L can be expressed as:

[0109]

[0110] And the parameters h1 and h2 can be expressed as:

[0111]

[0112] When calculating the coordinates of the interpolation points on the elliptical transition section trajectory, since the transition curve is a plane curve in a two-dimensional coordinate system, it is necessary to transform the coordinates (x 1 , y 1 ) in the two-dimensional coordinate system o1-xy of the curve into the three-dimensional coordinates (X, Y, Z) in the robot coordinate system O-XYZ. The transformation relationship is:

[0113]

[0114] Among them, (X o1 , Y o1 , Z o1 ) are the absolute coordinates of the origin o1 of the two-dimensional coordinate system o1-xy. It can be expressed as:

[0115]

[0116] According to the above given various parameters and calculation equations, the coordinates of each interpolation point can be obtained in turn, thus completing the interpolation of the modified "door"-shaped trajectory.

[0117] As a preferred solution of this embodiment, in step S12, a 3-4-5 polynomial is used to implement the acceleration and deceleration control of the optimized "door"-shaped trajectory. The 3-4-5 polynomial is:

[0118] L(δ) = aδ5 +bδ 4 +cδ 3 +dδ 2 +eδ + f

[0119] Among them, δ is the normalized motion time, and δ can be expressed as:

[0120]

[0121] t is the moment when the trajectory is in motion, T is the total duration of the trajectory motion, L(δ) is the normalized motion distance, and L(δ) can be expressed as:

[0122]

[0123] S(δ) is the transformation law of the trajectory motion distance with respect to the normalized time, and S a is the total distance of the trajectory motion, and a, b, c, d, e, and f are the coefficients of the 3 - 4 - 5th degree polynomial respectively.

[0124] According to the boundary conditions, when δ = 0, there is

[0125]

[0126] When δ = 1, there is

[0127]

[0128] Combined with the boundary conditions of the initial and final positions, the coefficients of the 3 - 4 - 5th degree polynomial can be calculated as:

[0129]

[0130] Finally, according to the L(δ) expression, the motion law of the 3 - 4 - 5th degree polynomial is expressed as:

[0131]

[0132] In the formula, S(δ) is the transformation law of the distance with respect to the normalized time, V(δ) is the transformation law of the velocity with respect to the normalized time, and A(δ) is the transformation law of the acceleration with respect to the normalized time.

[0133] As Figure 1 - Figure 7 shown, taking the vision sorting system written in the YOLOV5 algorithm as an example, the vision sorting method based on the Delta parallel robot is as follows:

[0134] Step S1: Analyze the coordinate system of the robot and obtain the forward and inverse solutions of the robot through geometric relationships;

[0135] Step S2: Obtain the workspace range of the robot through forward and inverse kinematics, and then calculate the actual workspace of the robot according to actual requirements;

[0136] Step S3: Calculate the rotation angle range of the motor according to the actual workspace of the robot;

[0137] Step S4: Perform perspective transformation on the recognized area image to remove some redundant images;

[0138] Step S5: Perform hand-eye calibration on the image after perspective transformation;

[0139] Step S6: Write a visual recognition model using the YOLOV5 algorithm;

[0140] Step S7: Collect images of the objects to be recognized to form a training set, and train and test the model;

[0141] Step S8: Preprocess and postprocess the image, and then use the visual recognition model to obtain the absolute position of the key frame of the object;

[0142] Step S9: Calculate the absolute position of the key frame of the item, and update the item position and calculate the grasping point position in real time.

[0143] Step S10: Implement the trajectory planning of the robot through a "door"-shaped trajectory;

[0144] Step S11: Optimize the "door"-shaped trajectory through an elliptical transition trajectory;

[0145] Step S12: Use a 3-4-5 polynomial to implement the acceleration and deceleration control of the optimized "door"-shaped trajectory.

[0146] Table 1 shows some training parameters of the YOLOV5 model in the embodiment of the present invention

[0147] Parameter Name Parameter Value Initial Learning Rate 0.005 Learning Rate Decay Factor 0.01 Batch Size 128 Number of Training Iterations 500 Image Size 384×384 Prediction Box Loss Coefficient 0.15 Confidence Loss Coefficient 1.0 Classification Loss Coefficient 0.5

[0148] Based on the same inventive concept, the present invention also provides a computer device, which includes: one or more processors, and a memory for storing one or more computer programs; the program includes program instructions, and the processor is configured to execute the program instructions stored in the memory. The processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, and is used to implement one or more instructions. Specifically, it is used to load and execute one or more instructions in the computer storage medium to implement the above method.

[0149] It should be further noted that, based on the same inventive concept, the present invention also provides a computer storage medium, on which a computer program is stored, and the computer program, when run by a processor, executes the above method. The storage medium may adopt any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may, for example, but not be limited to, be an electrical, magnetic, optical, electrical, magnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (non-exhaustive list) of the computer-readable storage medium include: an electrical connection having one or more wires, a portable computer disk, a hard disk, a Random Access Memory (RAM), a Read-Only Memory (ROM), an Erasable Programmable Read-Only Memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, the computer-readable storage medium may be any tangible medium that contains or stores a program, and the program can be used by or combined with an instruction execution system, apparatus, or device.

[0150] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present invention shall have the ordinary meanings understood by those of ordinary skill in the field to which the present invention pertains. The "first", "second" and similar terms used in the present invention do not denote any order, quantity or importance, but are only used to distinguish different components. Words such as "comprising" or "including" mean that the elements or objects appearing before this word cover the elements or objects listed after this word and their equivalents, without excluding other elements or objects. Words such as "connected" or "coupled" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Upper", "lower", "left", "right", etc. are only used to indicate relative positional relationships, and when the absolute position of the object being described changes, the relative positional relationship may also change accordingly.

[0151] As described above, it is only the preferred embodiment of the present invention, and it is not a limitation to the present invention in other forms. Any person skilled in the art may use the technical content disclosed above to make changes or modifications into equivalent embodiments with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution content of the present invention still fall within the protection scope of the technical solution of the present invention.

[0152] The present invention is not limited to the above best mode. Anyone can obtain other various forms of a vision sorting method based on a Delta robot under the inspiration of the present invention. All equal changes and modifications made according to the scope of the patent application of the present invention shall fall within the scope covered by the present invention.

Claims

1. A visual sorting method based on a Delta robot, characterized in that: The position of the moving target is identified in real time by the visual system, and the coordinates of the grabbing point are dynamically corrected in combination with the conveyor belt movement speed, so that the robot end and the target arrive synchronously; Plan a "gate"-shaped trajectory connecting the robot end and the grasping point, and use a transition curve with continuous curvature to optimize the right-angle turning section to avoid sudden acceleration; A high-order continuous acceleration control algorithm is used to drive the robot movement to ensure continuous and smooth speed and acceleration and eliminate mechanical shock.

2. A visual sorting method based on a Delta robot according to claim 1, characterized in that: The dynamically corrected grab point coordinates include: Calculate the time required for the robot to move from the current position to the target initial position in real time, which is recorded as the grasping time; According to the product of the grabbing time and the running speed of the conveyor belt, the moving distance of the conveyor belt in the grabbing time period is calculated and recorded as the compensation distance; The compensation distance is deducted from the horizontal coordinate of the target initial position in real time to generate a dynamically updated grasping point position to ensure that the robot end arrives synchronously with the moving target.

3. A visual sorting method based on a Delta robot according to claim 1, characterized in that: The transition curve with continuous curvature is an elliptical curve, and its horizontal length and vertical height are dynamically adjusted in the following manner: Based on the horizontal projection length of the trajectory, the horizontal semi-axis length is dynamically generated according to the preset horizontal transition ratio; Based on the preset lifting height, the vertical semi-axis height is dynamically generated according to the vertical transition ratio; The curvature continuity of the elliptical curve is controlled by coordinating the horizontal semi-axis and the vertical semi-axis. The interpolation point coordinates of the elliptical transition segment are realized by coordinate transformation from a two-dimensional plane coordinate system to a three-dimensional coordinate system of the robot.

4. A visual sorting method based on a Delta robot according to claim 3, characterized in that: The coordinate transformation from the two-dimensional plane coordinate system to the robot three-dimensional coordinate system is achieved by the following steps: According to the transition section direction angle θ, a rotation matrix is ​​constructed; Based on the two-dimensional interpolation point coordinates and the rotation matrix, the three-dimensional absolute coordinates in the robot coordinate system are calculated.

5. The visual sorting method based on a Delta robot according to claim 1, characterized in that: The high-order continuous acceleration control algorithm adopts the fifth-order polynomial motion law and meets the following conditions: At the start of the movement when δ=0 and at the end when δ=1, the position, velocity, and acceleration are all zero; During the movement, the acceleration curve is continuous and has no sudden changes; The motion trajectory is dynamically adjusted by normalizing the time parameter δ to ensure a smooth transition between the velocity and acceleration of the robot end.

6. The visual sorting method based on a Delta robot according to claim 1, characterized in that: The deduplication method for real-time recognition by the visual system includes: Traverse and compare the coordinates of objects identified in consecutive key frames; If the coordinate deviation of the same object in adjacent frames is less than a preset threshold, it is considered a duplicate target and ignored; Otherwise, add the object coordinates to the list to be grasped to avoid repeated grasping actions.

7. The visual sorting method based on a Delta robot according to claim 1, characterized in that: The visual system achieves target recognition by: Use a deep learning-based target detection model to identify and locate objects on the conveyor belt in real time; The model is trained with a small learning rate of ≤0.005 and a large batch of data of ≥128 to meet the real-time requirements of high-speed sorting scenarios; Output object category, location and confidence information for dynamic grasping point calculation.

8. The visual sorting method based on a Delta robot according to claim 1, characterized in that: The visual system calculates the homogeneous transformation matrix through multi-point calibration optimization, and converts the object coordinates in the conveyor belt coordinate system into the robot absolute coordinate system.

9. A visual sorting system based on a Delta parallel robot, characterized in that: include: Visual processing module: used to identify the position of the moving target on the conveyor belt in real time and dynamically correct the coordinates of the grabbing point based on the conveyor belt movement speed; Trajectory planning module: used to generate a "door"-shaped trajectory connecting the robot end and the grasping point, and optimize the curvature continuity of the right-angle turning section through an elliptical curve; Motion control module: used to execute high-order continuous acceleration control algorithm to drive the parallel robot to achieve smooth motion; Communication and execution module: used to transmit control instructions to the robot drive unit and synchronize the end effector action.

10. A visual sorting system based on a Delta robot according to claim 9, characterized in that: The visual processing module comprises: Deep learning object detection unit, using the pre-trained YOLOv5 model; The coordinate conversion unit realizes the mapping from image coordinates to robot absolute coordinates through hand-eye calibration and homogeneous transformation matrix; Dynamic compensation unit calculates the real-time compensation distance based on the grab time and conveyor speed.

Images