Fusion image perception and variable-order fractional order motion control method

Abstract

The invention discloses a fusion image perception and variable-order fractional order motion control method, and belongs to the technical field of automatic control. The method comprises the following steps: establishing a sliding platform coordinate system; a target image is collected through binocular vision, and three-dimensional coordinates of a target object are extracted through median filtering, weighted equalization, edge masking and binarization processing; constructing an error function according to the target three-dimensional coordinates and the current position of the motor, and constructing a nonlinear order function in combination with target speed derivative information; the function is substituted into a speed ring of a three-ring control structure, a variable-order fractional order controller is adopted, a fractional order operator is quickly approached through improved polynomial fitting and dichotomy, an optimal control quantity is generated in real time to drive a three-axis motor, and high-precision tracking control over a target object is achieved. According to the method, image perception and variable-order control are deeply fused, and the tracking precision, the response speed and the adaptive capacity of the system in a complex dynamic scene are effectively improved.

Description

TECHNICAL FIELD

[0001] The application relates to a fusion image perception and variable order fractional order motion control method and belongs to the technical field of automatic control. BACKGROUND

[0002] With the continuous development of artificial intelligence and automatic control technology, high-precision and real-time control of target objects in complex systems has become a current research hotspot. In application scenarios such as robot operation, automatic trajectory tracking, and sliding platform control, the target state has significant nonlinearity, time-varying nature, and uncertainty, and traditional control methods cannot balance the response accuracy, tracking accuracy, and anti-interference ability of the system. Fractional order controllers are widely studied due to their stronger dynamic description capability, but the fixed order structure is still insufficient to meet the real-time adjustment requirements when the target state changes dramatically. At the same time, although image perception technology can provide high-precision three-dimensional position information, in complex real-time systems, efficient fusion into the control system still faces challenges such as large delay and complex structure.

[0003] In the prior art, patent CN117111913A proposes a real-time motion control method based on code generation, which realizes current and speed double-loop control of the motor through Simulink modeling and code generation combined with fractional order PIλ control algorithm, can reduce the amount of manual programming, and improve development efficiency. However, this scheme only focuses on code generation and motor control, does not involve image perception and visual servoing, and the fractional order control method is also relatively single. Patent CN214799034U discloses a fast prototype controller based on backplane plug-in real-time scheduling, which expands ADC, DAC, PWM, QEP, etc. interfaces through a modular architecture of DSP+FPGA+co-processor, and is suitable for real-time management of distributed power grid energy routing. This method emphasizes hardware modularization and energy scheduling, but the application scenario is limited to power systems, does not consider motion control and image perception, and does not involve variable order fractional order control. Patent CN115657544A discloses a medium-frequency auxiliary converter DCDC control system, which adopts FPGA, DSP, and ARM chips to realize voltage and current acquisition, PWM control, logic protection, and network communication, and improves the stability of the rail transit power supply system. The research object of this scheme is a power electronic converter, and it does not involve visual servoing and fractional order control methods. Patent CN107300854A proposes a variable order fractional order calculus time domain analysis method based on polynomial fitting, which focuses on fitting the transfer function coefficient matrix through the least squares method to realize fast solution of any order. However, its application object is limited to simulation analysis systems, and it does not involve real-time control or multi-source information fusion. SUMMARY

[0004] The application aims to provide a fusion image perception and variable order fractional order motion control method, which obtains three-dimensional coordinates of a target object based on an image acquisition module, combines a variable order fractional order control algorithm, and constructs a controller transfer function under an arbitrary order through a two-point method and a polynomial fitting, so as to improve the calculation efficiency of the transfer function, realize rapid adaptive adjustment of the controller to state changes, and improve the dynamic response and trajectory tracking accuracy of the system.

[0005] To achieve the above-mentioned purpose / solve the above-mentioned technical problems, the application is implemented by using the following technical scheme.

[0006] In one aspect, the application provides a fusion image perception and variable order fractional order motion control method, comprising:

[0007] establishing a sliding platform Cartesian coordinate system;

[0008] acquiring continuous image frames of a target object from left and right perspectives of the sliding platform, and extracting three-dimensional coordinate information of the target object in the Cartesian coordinate system from the continuous image frames from the left and right perspectives;

[0009] constructing an error function according to the three-dimensional coordinate information of the continuous target object and current position coordinates of three-axis motors driving the target object to move;

[0010] constructing a three-dimensional motion trajectory of the target object according to the three-dimensional coordinate information of the continuous target object, and calculating velocity derivative information of the target object;

[0011] constructing a nonlinear order function according to the velocity derivative information of the target object and the error function, and substituting the nonlinear order function into a velocity loop of a three-loop motion control structure;

[0012] adopting a variable order fractional calculus controller in the velocity loop, combining an improved polynomial fitting method to fit and approximate the fractional order operator, and using a bisection method to quickly determine an optimal fitting coefficient corresponding to a current order interval to generate a control quantity to drive the three-axis motors and realize tracking control of the target object.

[0013] Further, before extracting the three-dimensional coordinate information of the target object in the Cartesian coordinate system from the continuous image frames from the left and right perspectives, performing median filtering processing on the image frames through a grid data window, specifically:

[0014] ;

[0015] wherein, is an original pixel gray value of a pixel position of an image frame, is a median filtered pixel gray value of the pixel position of the image frame, ​​For pixel position on image frame, e, f respectively grid data window value range.

[0016] The median filter processing of image frame by grid data window can effectively suppress the salt and pepper noise and random noise introduced in image acquisition process, improve the image signal-to-noise ratio, lay a solid foundation for subsequent image segmentation and feature extraction, thereby indirectly improve the extraction accuracy of three-dimensional coordinate information and system anti-interference ability.

[0017] Further, the three-dimensional coordinate information of the target object in the Cartesian coordinate system is extracted from the continuous image frames of left and right view angles, specifically including:

[0018] The image frame after median filter processing is evenly divided into sub-block regions, and a local gray level histogram is constructed based on each sub-block region, and the expression is:

[0019] ;

[0020] ;

[0021] wherein, is the pth sub-block region, m, n are the number of horizontal and vertical sub-block regions respectively, is the frequency of pixels with gray level q in the pth sub-block region, q is the gray level of the pth sub-block region, , is a conditional indicator function;

[0022] According to the gray level histogram constructed by the sub-block region, the cumulative distribution function is calculated, and the expression is:

[0023] ;

[0024] wherein: is the frequency of pixels with gray level s in the pth sub-block region;

[0025] The gray level histograms constructed for all sub-block regions are merged to construct a global gray level histogram:

[0026] ;

[0027] wherein, is the total frequency of pixels of the entire image frame at the gray level , is the pixel set of the entire image frame;

[0028] According to the global gray level histogram, a global cumulative distribution function is constructed, and the expression is:

[0029] ;

[0030] Where C and D are the number of pixels in the vertical and horizontal directions of the image frame, respectively; This represents the total frequency of pixels in the entire image frame at gray level s;

[0031] Based on the statistical results of the global and local grayscale histograms, for each pixel in the image frame... Perform weighted equalization mapping processing, combined with local Functions and Global The function calculates each pixel in an image frame. The mapped grayscale values ​​yield a weighted equalized image, expressed as:

[0032] ;

[0033] ;

[0034] in, For weighted equalization of images The grayscale value at the pixel location. represents the weighting coefficients for local and global mapping; L represents the number of gray levels.

[0035] For weighted equalized images, edge masks are constructed using gradient magnitudes. To distinguish between edge regions and non-edge regions of a target object in a weighted equalization image, the expression is:

[0036] ;

[0037] in, for Gradient magnitude at pixel location; The threshold for determining the edge of the target object. for Edge mask for pixel location;

[0038] when When, it means The pixel location belongs to the edge region of the target object.

[0039] when When, it means The pixel location belongs to a non-edge area of ​​the target object;

[0040] Blur compensation is performed on the non-edge regions of the target object in the weighted equalization image to obtain the compensated image, expressed as:

[0041] ;

[0042] in, To compensate for image processing, The mean gray level of the neighborhood centered at pixel (x,y);

[0043] The compensated image is binarized, and the contrast between the target object and the background is enhanced by threshold segmentation to extract the foreground target. The expression is as follows:

[0044] ;

[0045] in, The image after binarization. The set segmentation threshold,

[0046] when At that time, the judgment The pixel location belongs to the foreground object.

[0047] when At that time, the judgment Pixel position background area;

[0048] Based on the coordinates of the foreground target in the target area, calculate the two-dimensional centroid coordinates of the target object, expressed as:

[0049] ;

[0050] in, Let x be the x-coordinate of the target object at time t in image frame. Let be the ordinate of the target object at time t in image frame. The total number of target areas. Here, x and y are the x and y coordinates of the k-th foreground target at time t in image frame t, respectively.

[0051] Depth estimation is performed based on the dual-view image input from the sliding platform, and the depth coordinates of the target object are calculated based on the triangulation principle. The expression is as follows:

[0052] ;

[0053] ;

[0054] in, Let t be the depth position of the target object in space at time t. S is the focal length of the camera, and S is the baseline distance between the two cameras. Let be the pixel disparity of the target object in the left and right images acquired by the stereo camera at time t. and These are the x-coordinates of the left and right camera images of the target object at time t, respectively.

[0055] The three-dimensional coordinates of the target object are constructed based on its two-dimensional centroid coordinates and depth coordinates as follows: .

[0056] By segmenting the image and constructing local and global grayscale histograms, combined with weighted equalization mapping, the overall contrast of the image can be enhanced while effectively avoiding information loss in excessively bright or dark areas, significantly improving the image's adaptability under non-uniform lighting conditions. Furthermore, by constructing edge masks and performing blur compensation on non-edge areas, background noise can be smoothed while sharpening the target edges, making subsequent binarization segmentation more accurate. Finally, through centroid calculation and binocular triangulation, high-precision and robust extraction of the target's three-dimensional coordinates is achieved in complex backgrounds.

[0057] Furthermore, the gradient magnitude The calculation method is as follows:

[0058] right The gradient is convolved with a Sobel kernel to obtain the horizontal and vertical gradient components. The gradient magnitude is then calculated based on these components, expressed as follows:

[0059] ;

[0060] ;

[0061] in, The horizontal convolution kernel matrix, This is the vertical convolution kernel matrix.

[0062] Furthermore, the error function constructed based on the three-dimensional coordinate information of the continuous target and the current position coordinates of the three-axis motor driving the target's motion is expressed as follows:

[0063] ;

[0064] ;

[0065] ;

[0066] in: Let be the error function. A continuous three-dimensional coordinate sequence of the target object. This is the sequence of coordinates of the current position of the three-axis motor at time t. These represent adjacent sampling times.

[0067] Furthermore, the nonlinear order function constructed based on the velocity derivative information of the target object and the error function is expressed as follows:

[0068] ;

[0069] in: Let be the error function. For the velocity derivative information of the target object, Let the order of the fractional integral operator be denoted as . This is for nonlinear order functions.

[0070] By combining the velocity derivative information of the target object with the error function to construct a nonlinear order function, the order of the fractional-order controller can be dynamically and nonlinearly adaptively adjusted according to the motion state of the target and the tracking error of the system. This gives the controller a more flexible dynamic characteristic adjustment capability, and when faced with non-stationary motions such as sudden changes in the speed of the target object, it can adjust the control strategy more quickly to reduce tracking lag.

[0071] Furthermore, the three-ring motion control structure includes:

[0072] The inner current loop uses an integer-order PI controller for rapid adjustment of current error;

[0073] The intermediate velocity loop employs a variable-order fractional-order proportional integral. The controller has the following structure:

[0074] ;

[0075] in, This is the speed loop proportional coefficient. For integral gain, For the velocity loop transfer function, For fractional integral operators, Let be the order of the fractional integral operator;

[0076] The outermost position loop uses a proportional controller, and its structure is as follows:

[0077] ;

[0078] in, For the position loop transfer function, This is the position loop proportional gain.

[0079] A three-loop control structure consisting of a current loop, a speed loop, and a position loop is adopted. Variable-order fractional-order controllers and proportional controllers are used in the speed loop and position loop respectively, achieving a reasonable allocation of control tasks. The integer-order PI controller in the inner loop ensures rapid current response, the proportional controller in the outermost position loop ensures system stability, and the variable-order fractional-order control in the middle speed loop addresses the nonlinear and time-varying challenges brought about by system inertia and external disturbances. This architecture maximizes the advantages of variable-order fractional-order control in dynamic performance adjustment while ensuring system stability.

[0080] Furthermore, the variable-order fractional-order calculus controller is used in the speed loop, combined with an improved polynomial fitting method to fit and approximate the fractional-order operator, and the optimal fitting coefficients are obtained by quickly determining the corresponding interval of the current order using the bisection method, generating control quantities to drive the three-axis motor and realize the tracking control of the target object, specifically including:

[0081] Step (1): Select M uniformly distributed order points within the order range (0, 1) of the fractional integral operator. ,in ,

[0082] Step (2): In each order Next, using fractional integral operators As the core of the filter, its corresponding discrete-time transfer function is constructed. A unit impulse input is applied to obtain the filter's time-domain output response at that order. Furthermore, based on the filter's input and output data, the Prony algorithm is used to fit its difference equation and extract the corresponding discrete-time transfer function.

[0083] Step (3): Construct the denominator coefficient matrix A based on the discrete transfer function in the time domain;

[0084] Step (4): Construct the molecular coefficient matrix B based on the time-domain discrete transfer function;

[0085] Step (5): Using the discrete point order X = [ , ,… [] is the input, representing the current order. Falling on adjacent discrete points and When the values ​​are in between, take the corresponding denominator coefficient matrix respectively. and The column vectors are output column by column, and the two-point linear interpolation method is used for fitting calculation to obtain the N+1 polynomial of the denominator coefficients;

[0086] Step (6): Using the discrete point order X = [ , ,… [] is the input, representing the current order. Falling on adjacent discrete points and When the values ​​are in between, take the corresponding molecule coefficient matrix respectively. and The column vectors are output column by column, and the two-point linear interpolation method is used for fitting calculation to obtain the N+1 polynomial of the molecular coefficients;

[0087] Step (7): Combine the N+1 polynomials of the denominator coefficients and the N+1 polynomials of the numerator coefficients to obtain the time-domain transfer function of the corresponding fractional calculus in any region. Based on the real-time extraction of the target object's three-dimensional coordinate changes and the error feedback of the servo motor from the image acquisition module, calculate the motion control signal.

[0088] Furthermore, the denominator coefficient matrix A specifically includes:

[0089] ;

[0090] in: Let A represent the Nth row and Mth column element of the denominator coefficient matrix A, which corresponds to the Nth order coefficient of the denominator polynomial at the order point M.

[0091] The molecular coefficient matrix B specifically includes:

[0092] ;

[0093] in: Let represent the Nth row and Mth column element of the numerator coefficient matrix B, corresponding to the Nth order coefficient of the denominator polynomial at order point M.

[0094] Furthermore, the two-point linear interpolation method is used for fitting calculations, specifically including:

[0095] ;

[0096] ;

[0097] in: In order The denominator coefficients are below. For the first A known order point, For the current variable order being calculated, For order point The next The coefficient of the denominator is 1. For coefficient serial numbers, For order point index, In order The molecular coefficients below, For order point The next Molecular coefficient.

[0098] By employing an improved polynomial fitting method combined with the bisection method, fast interpolation is performed between pre-calculated discrete order points, enabling real-time and efficient generation of fractional operator transfer functions of arbitrary order. Compared to traditional online numerical calculations, this significantly reduces computational complexity and real-time processing resource consumption, ensuring that the controller can quickly update parameters within each control cycle. This enables the engineering application of variable-order fractional control in real-time motion control systems, meeting the stringent real-time requirements of control in highly dynamic scenarios.

[0099] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0100] Compared to existing technologies that only focus on code generation, motor dual closed-loop control, or power system energy dispatching, this invention innovatively integrates image perception and variable-order fractional-order control algorithms to achieve high-precision real-time control of moving targets in complex dynamic environments. This significantly improves the system's control accuracy, response speed, and anti-interference capability. At the same time, it has good stability and scalability, making it suitable for fields such as intelligent manufacturing, robot control, and automated assembly. Attached Figure Description

[0101] Figure 1 This is a schematic diagram of the process of the present invention;

[0102] Figure 2 The diagram shown is a flowchart of one embodiment of the scalable real-time measurement and control system that integrates image perception and variable-order fractional-order measurement and control according to the present invention.

[0103] Figure 3 The diagram shown is a structural schematic of one embodiment of the scalable real-time measurement and control system that integrates image perception and variable-order fractional-order measurement and control according to the present invention. Detailed Implementation

[0104] It should be noted that:

[0105] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0106] The term "and / or" simply describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. Additionally, the character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0107] Example 1

[0108] like Figure 1 The embodiment shown provides a method for fusing image perception and variable-order fractional-order motion control, including:

[0109] S1: Establish the Cartesian coordinate system for the sliding platform;

[0110] S2: Acquire continuous image frames of the target object from the left and right perspectives on the sliding platform, and extract the three-dimensional coordinate information of the target object in the Cartesian coordinate system from the continuous image frames from the left and right perspectives.

[0111] S3: Construct an error function based on the three-dimensional coordinate information of the continuous target and the current position coordinates of the three-axis motor that drives the target's motion;

[0112] S4: Construct the three-dimensional motion trajectory of the target object based on the three-dimensional coordinate information of the continuous target object, and calculate the velocity derivative information of the target object;

[0113] S5: Construct a nonlinear order function based on the velocity derivative information of the target object and the error function, and substitute it into the velocity loop of the three-loop motion control structure;

[0114] S6: A variable-order fractional-order calculus controller is used in the speed loop. An improved polynomial fitting method is used to fit and approximate the fractional-order operator. The bisection method is used to quickly determine the corresponding interval of the current order to obtain the optimal fitting coefficients, generate control quantities to drive the three-axis motor, and realize the tracking control of the target object.

[0115] Before extracting the 3D coordinate information of the target object in the Cartesian coordinate system from consecutive image frames from left and right perspectives, the image frames are processed by median filtering through a raster data window, specifically:

[0116] ;

[0117] in, For image frames The original pixel grayscale value at the pixel location. For image frames The median-filtered grayscale value of a pixel at a given pixel location. Let be the pixel position on the image frame, and e and f be the value ranges of the raster data window, respectively.

[0118] The extraction of the three-dimensional coordinate information of the target object in the Cartesian coordinate system from consecutive image frames from left and right perspectives specifically includes:

[0119] The image frames after median filtering are evenly divided into... Each sub-block region is divided into several sub-block regions, and a local grayscale histogram is constructed based on each sub-block region. The expression is as follows:

[0120] ;

[0121] ;

[0122] in, Let m be the p-th sub-block region, and m and n be the number of horizontal and vertical sub-block regions, respectively. Let be the frequency of pixels with gray level q within the p-th sub-block region, where q is the gray level of the p-th sub-block region. , For conditional indicator functions;

[0123] Calculate the cumulative distribution function based on the gray-level histogram constructed from the sub-block regions. The expression is:

[0124] ;

[0125] in: Let be the frequency of pixels with gray level s within the p-th sub-block region;

[0126] The grayscale histograms of all sub-block regions are merged to construct a global grayscale histogram:

[0127] ;

[0128] in, For the entire image frame in grayscale Total pixel frequency It is the set of pixels for the entire image frame;

[0129] Construct a global cumulative distribution function based on the global grayscale histogram. The expression is:

[0130] ;

[0131] Where C and D are the number of pixels in the vertical and horizontal directions of the image frame, respectively; This represents the total frequency of pixels in the entire image frame at gray level s;

[0132] Based on the statistical results of the global and local grayscale histograms, for each pixel in the image frame... Perform weighted equalization mapping processing, combined with local Functions and Global The function calculates each pixel in an image frame. The mapped grayscale values ​​yield a weighted equalized image, expressed as:

[0133] ;

[0134] ;

[0135] in, For weighted equalization of images The grayscale value at the pixel location. represents the weighting coefficients for local and global mapping; L represents the number of gray levels.

[0136] For weighted equalized images, edge masks are constructed using gradient magnitudes. To distinguish between edge regions and non-edge regions of a target object in a weighted equalization image, the expression is:

[0137] ;

[0138] in, for Gradient magnitude at pixel location; The threshold for determining the edge of the target object. for Edge mask for pixel location;

[0139] when When, it means The pixel location belongs to the edge region of the target object.

[0140] when When, it means The pixel location belongs to a non-edge area of ​​the target object;

[0141] Blur compensation is performed on the non-edge regions of the target object in the weighted equalization image to obtain the compensated image, expressed as:

[0142] ;

[0143] in, To compensate for image processing, The mean gray level of the neighborhood centered at pixel (x,y);

[0144] The compensated image is binarized, and the contrast between the target object and the background is enhanced by threshold segmentation to extract the foreground target. The expression is as follows:

[0145] ;

[0146] in, The image after binarization. The set segmentation threshold,

[0147] when At that time, the judgment The pixel location belongs to the foreground object.

[0148] when At that time, the judgment Pixel position background area;

[0149] Based on the coordinates of the foreground target in the target area, calculate the two-dimensional centroid coordinates of the target object, expressed as:

[0150] ;

[0151] in, Let x be the x-coordinate of the target object at time t in image frame. Let be the ordinate of the target object at time t in image frame. The total number of target areas. Here, x and y are the x and y coordinates of the k-th foreground target at time t in image frame t, respectively.

[0152] Depth estimation is performed based on the dual-view image input from the sliding platform, and the depth coordinates of the target object are calculated based on the triangulation principle. The expression is as follows:

[0153] ;

[0154] ;

[0155] in, Let t be the depth position of the target object in space at time t. S is the focal length of the camera, and S is the baseline distance between the two cameras. Let be the pixel disparity of the target object in the left and right images acquired by the stereo camera at time t. and These are the x-coordinates of the left and right camera images of the target object at time t, respectively.

[0156] The three-dimensional coordinates of the target object are constructed based on its two-dimensional centroid coordinates and depth coordinates as follows: .

[0157] The gradient magnitude The calculation method is as follows:

[0158] right The gradient is convolved with a Sobel kernel to obtain the horizontal and vertical gradient components. The gradient magnitude is then calculated based on these components, expressed as follows:

[0159] ;

[0160] ;

[0161] in, The horizontal convolution kernel matrix, This is the vertical convolution kernel matrix.

[0162] The error function is constructed based on the three-dimensional coordinate information of the continuous target and the current position coordinates of the three-axis motor driving the target's motion. The expression is as follows:

[0163] ;

[0164] ;

[0165] ;

[0166] in: Let be the error function. A continuous three-dimensional coordinate sequence of the target object. This is the sequence of coordinates of the current position of the three-axis motor at time t. These represent adjacent sampling times.

[0167] The nonlinear order function constructed based on the velocity derivative information of the target object and the error function is expressed as follows:

[0168] ;

[0169] in: Let be the error function. For the velocity derivative information of the target object, It is a nonlinear order function. This is for nonlinear order functions.

[0170] The three-ring motion control structure includes:

[0171] The inner current loop uses an integer-order PI controller for rapid adjustment of current error;

[0172] The intermediate velocity loop employs a variable-order fractional-order proportional integral. The controller has the following structure:

[0173] ;

[0174] in, This is the speed loop proportional coefficient. For integral gain, For the velocity loop transfer function, For fractional integral operators, Let be the order of the fractional integral operator;

[0175] The outermost position loop uses a proportional controller, and its structure is as follows:

[0176] ;

[0177] in, For the position loop transfer function, This is the position loop proportional gain;

[0178] The process involves employing a variable-order fractional-order calculus controller in the velocity loop, combined with an improved polynomial fitting method to approximate the fractional-order operator, and using a bisection method to quickly determine the optimal fitting coefficients for the current order interval. This generates control inputs to drive the three-axis motor, achieving target tracking control. Specifically, this includes:

[0179] Step (1): Select M uniformly distributed order points within the order range (0, 1) of the fractional integral operator. ,in ,

[0180] Step (2): In each order Next, using fractional integral operators As the core of the filter, its corresponding discrete-time transfer function is constructed. A unit impulse input is applied to obtain the filter's time-domain output response at that order. Furthermore, based on the filter's input and output data, the Prony algorithm is used to fit its difference equation and extract the corresponding discrete-time transfer function.

[0181] Step (3): Construct the denominator coefficient matrix A based on the discrete transfer function in the time domain;

[0182] Step (4): Construct the molecular coefficient matrix B based on the time-domain discrete transfer function;

[0183] Step (5): Using the discrete point order X = [ , ,… [] is the input, representing the current order. Falling on adjacent discrete points and When the values ​​are in between, take the corresponding denominator coefficient matrix respectively. and The column vectors are output column by column, and the two-point linear interpolation method is used for fitting calculation to obtain the N+1 polynomial of the denominator coefficients;

[0184] Step (6): Using the discrete point order X = [ , ,… [] is the input, representing the current order. Falling on adjacent discrete points and When the values ​​are in between, take the corresponding molecule coefficient matrix respectively. and The column vectors are output column by column, and the two-point linear interpolation method is used for fitting calculation to obtain the N+1 polynomial of the molecular coefficients;

[0185] Step (7): Combine the N+1 polynomials of the denominator coefficients and the N+1 polynomials of the numerator coefficients to obtain the time-domain transfer function of the corresponding fractional calculus in any region. Based on the real-time extraction of the target object's three-dimensional coordinate changes and the error feedback of the servo motor from the image acquisition module, calculate the motion control signal.

[0186] The denominator coefficient matrix A specifically includes:

[0187] ;

[0188] in: Let A represent the Nth row and Mth column element of the denominator coefficient matrix A, which corresponds to the Nth order coefficient of the denominator polynomial at the order point M.

[0189] The molecular coefficient matrix B specifically includes:

[0190] ;

[0191] in: Let the element in the Nth row and Mth column of the numerator coefficient matrix B be the Nth order coefficient of the denominator polynomial at the corresponding order point M.

[0192] Furthermore, the two-point linear interpolation method is used for fitting calculations, specifically including:

[0193] ;

[0194] ;

[0195] in: In order The denominator coefficients are below. For the first A known order point, For the current variable order being calculated, For order point The next The coefficient of the denominator is 1. For coefficient serial numbers, For order point index, In order The molecular coefficients below, For order point The next Molecular coefficient.

[0196] Example 2

[0197] like Figures 2-3As shown, based on the same inventive concept as Embodiment 1, this embodiment introduces a motion control system that integrates image perception and variable order fractional order, including: a host computer (1), a code execution board (2), a synchronization controller (3), a data recording and storage board (4), an image acquisition module (5), an absolute encoder processing module (6), an incremental encoder processing module (7), a PWM drive output module (8), an AD acquisition board (9), a DA analog output module (10), a digital IO input and output module (11), and a system bus (12).

[0198] The host computer is used to construct a mathematical model of the image perception algorithm and variable-order fractional-order control strategy system according to the control and functional requirements of the system; convert the control algorithm into control instructions suitable for the code execution board and image acquisition board and send them; and update and adjust the control instructions and control parameters online according to the feedback after receiving and processing the data.

[0199] The code execution board is used to receive control commands and external device data, run the core control algorithm, including variable order fractional calculus calculation and polynomial fitting solution, generate real-time control commands based on data from the image acquisition module and various sensors, and send them to the host computer or actuator drive module.

[0200] The synchronous controller is used to manage the clock and trigger of the three-axis motor and the acquisition module in a unified manner, ensuring that the data acquisition and execution commands are strictly aligned in the time domain.

[0201] Data recording and storage board, used to record acquired data, control commands and operating status in real time, and supports high-speed storage and historical data playback;

[0202] The image acquisition module is used to complete image acquisition, median filtering, Sobel edge detection, weighted equalization mapping, binarization, and target centroid calculation to obtain the three-dimensional coordinates of the moving target.

[0203] The absolute encoder processing module is used to acquire and calculate high-precision angle and position information of the absolute encoder motor.

[0204] The incremental encoder processing module is used to collect and calculate the high-frequency dynamic displacement and speed information of the incremental encoder motor, so as to realize fast response status feedback.

[0205] The PWM drive output module is used to convert control commands into multi-channel PWM signals to drive actuators such as servo motors;

[0206] The AD acquisition board is used to acquire external analog signals (voltage and current) and convert them into digital signals for processing by the code execution board.

[0207] The DA analog output module is used to convert digital control quantities into analog signal outputs to adapt to specific actuators or external system interfaces.

[0208] The digital I / O input / output module is used to acquire digital switch input signals and control digital output signals to realize system status monitoring and auxiliary control functions;

[0209] The system bus provides a unified data communication and clock synchronization channel for various functional modules, supporting high-speed parallel data transmission and multi-board collaborative operation.

[0210] The three-axis motors correspond to the three orthogonal motion directions of the sliding platform, with motor one driving linear motion in the X-axis direction, motor two driving linear motion in the Y-axis direction, and motor three driving lifting or pitching motion in the Z-axis direction.

[0211] The image acquisition module includes an image sensor and an image preprocessing unit. Two disparately positioned image sensors are used to simultaneously acquire left and right view image frames of the target sphere. The image preprocessing unit performs median filtering, Sobel edge detection, improved region adaptive background weighted equalization mapping, and binarization on the left and right image frames respectively, extracting the centroid pixel coordinates of the target sphere. A binocular triangulation model was constructed, and disparity was calculated. Obtain the Z-axis depth coordinates and calculate the 3D coordinates using the camera. ;

[0212] The code execution board uses a variable-order fractional-order calculus control method, and the image acquisition module extracts the continuous three-dimensional coordinate sequence of the target ball. Combined with the current displacement value of the three-axis motor collected by the encoder Construct the error function: Based on the rate of change of velocity in the three-dimensional coordinates of the image Sum of error functions Constructing nonlinear order functions A polynomial fitting control function is constructed to achieve dynamic adaptive modeling and control of non-stationary trajectories. The controller is built based on a polynomial fitting method, and in each control cycle, a variable-order discrete-time transfer function is dynamically constructed based on the error between the target 3D coordinate velocity extracted by the image module and the encoder feedback velocity. The transfer function is in the form of:

[0213] ;

[0214] Where Z is the discrete domain independent variable. Let be the denominator and numerator coefficients of the u-th order at the r-th sampling point. Fit the denominator coefficient matrix and numerator coefficient matrix:

[0215] ; ;

[0216] The controller uses a binary search method to handle different... The function fitting results are dynamically selected, and combined with the three-dimensional coordinates of the image and the speed error of the servo motor, the control signal is updated in real time to achieve high responsive and stable tracking of the three-dimensional coordinate trajectory.

[0217] The controller employs a polynomial fitting method, selecting M uniformly distributed order points within the order range (0, 1) applicable to integral control and the order range (-1, 0) applicable to derivative control. For each order point, a time-domain transfer function model of the corresponding continuous-time fractional calculus operator is constructed using Laplace transform, and its discrete-time transfer function set is obtained through numerical discretization. The numerator and denominator coefficients of the discrete transfer functions corresponding to each order are arranged in the same structure to construct a unified position coefficient matrix. Based on this, with the order as the independent variable and the position coefficient as the dependent variable, a polynomial fitting is performed on the column vector of this coefficient matrix to obtain a fitting function set that maps the order input to the transfer function coefficient output. Through the above fitting function set, during the controller operation, the fractional calculus transfer function corresponding to any order can be quickly calculated and generated in real time based on the currently estimated target dynamic order, thereby constructing a high-precision time-domain controller that meets the requirements of variable-order control, significantly improving the response capability and accuracy of the control system to non-stationary state changes. The controller ultimately generates a variable-order fractional-order control signal, which is a function of the order of the control quantity as it changes with the current time. Real-time adjustments are made to achieve adaptive control based on three-dimensional coordinate state driving; the control quantity is modulated by the PWM module and output to the power drive board to drive the three-axis servo motor, thereby realizing high-precision trajectory tracking and positioning control of the target ball in the sliding platform.

[0218] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0219] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0220] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0221] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0222] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A method for integrating image perception and variable-order fractional-order motion control, characterized in that, include: Establish a Cartesian coordinate system for the sliding platform; Acquire continuous image frames of the target object from the left and right perspectives on the sliding platform, and extract the three-dimensional coordinate information of the target object in the Cartesian coordinate system from the continuous image frames from the left and right perspectives; An error function is constructed based on the three-dimensional coordinate information of the continuous target and the current position coordinates of the three-axis motor driving the target's motion; Construct the three-dimensional motion trajectory of the target object based on its three-dimensional coordinate information, and calculate the velocity derivative information of the target object; A nonlinear order function is constructed based on the velocity derivative information of the target object and the error function, and then substituted into the velocity loop of the three-loop motion control structure. A variable-order fractional-order calculus controller is used in the speed loop. An improved polynomial fitting method is used to fit and approximate the fractional-order operator. The bisection method is used to quickly determine the corresponding interval of the current order to obtain the optimal fitting coefficients, and the control quantity is generated to drive the three-axis motor to achieve the tracking control of the target object.

2. The method for fusing image perception and variable-order fractional-order motion control according to claim 1, characterized in that, Before extracting the 3D coordinate information of the target object in the Cartesian coordinate system from consecutive image frames from left and right perspectives, the image frames are processed by median filtering through a raster data window, specifically: ； in, For image frames The original pixel grayscale value at the pixel location. For image frames The median-filtered grayscale value of a pixel at a given pixel location. Let be the pixel position on the image frame, and e and f be the value ranges of the raster data window, respectively.

3. The method for fusing image perception and variable-order fractional-order motion control according to claim 2, characterized in that, The extraction of the three-dimensional coordinate information of the target object in the Cartesian coordinate system from consecutive image frames from left and right perspectives specifically includes: The image frames after median filtering are evenly divided into... Each sub-block region is divided into several sub-block regions, and a local grayscale histogram is constructed based on each sub-block region. The expression is as follows: ； ； in, Let m be the p-th sub-block region, and m and n be the number of horizontal and vertical sub-block regions, respectively. Let be the frequency of pixels with gray level q within the p-th sub-block region, where q is the gray level of the p-th sub-block region. , For conditional indicator functions; Calculate the cumulative distribution function based on the gray-level histogram constructed from the sub-block regions. The expression is: ； in: Let be the frequency of pixels with gray level s within the p-th sub-block region; The grayscale histograms constructed from all sub-block regions are merged to construct a global grayscale histogram: ； in, For the entire image frame in grayscale Total pixel frequency It is the set of pixels for the entire image frame; Construct a global cumulative distribution function based on the global gray-level histogram. The expression is: ； Where C and D are the number of pixels in the vertical and horizontal directions of the image frame, respectively; This represents the total frequency of pixels in the entire image frame at gray level s; Based on the statistical results of the global and local grayscale histograms, for each pixel in the image frame... Perform weighted equalization mapping processing, combined with local Functions and Global The function calculates each pixel in an image frame. The mapped grayscale values ​​yield a weighted equalized image, expressed as: ； ； in, For weighted equalization of images The grayscale value at the pixel location. represents the weighting coefficients for local and global mapping; L represents the number of gray levels. Edge masks are constructed using gradient magnitudes on the weighted equalized image. To distinguish between edge regions and non-edge regions of a target object in a weighted equalization image, the expression is: ； in, for Gradient magnitude at pixel location; The threshold for determining the edge of the target object. for Edge mask for pixel location; when When, it means The pixel location belongs to the edge region of the target object. when When, it means The pixel location belongs to a non-edge area of ​​the target object; Blur compensation is performed on the non-edge regions of the target object in the weighted equalization image to obtain the compensated image, expressed as: ； in, To compensate for image processing, The mean gray level of the neighborhood centered at pixel (x,y); The compensated image is binarized, and the contrast between the target object and the background is enhanced by threshold segmentation to extract the foreground target. The expression is as follows: ； in, The image after binarization. The set segmentation threshold, when At that time, the judgment The pixel location belongs to the foreground object. when At that time, the judgment Pixel position background area; Based on the coordinates of the foreground target in the target area, calculate the two-dimensional centroid coordinates of the target object, expressed as: ； in, Let x be the x-coordinate of the target object at time t in image frame. Let be the ordinate of the target object at time t in image frame. The total number of target areas, Here, x and y are the x and y coordinates of the k-th foreground target at time t in image frame t, respectively. Depth estimation is performed based on the dual-view image input from the sliding platform, and the depth coordinates of the target object are calculated based on the triangulation principle. The expression is as follows: ； ； in, Let be the depth position of the target object in space at time t. S is the focal length of the camera, and S is the baseline distance between the two cameras. Let be the pixel disparity of the target object in the left and right images acquired by the stereo camera at time t. and These are the x-coordinates of the left and right camera images of the target object at time t, respectively. The three-dimensional coordinates of the target object are constructed based on its two-dimensional centroid coordinates and depth coordinates as follows: .

4. The method for fusing image perception and variable-order fractional-order motion control according to claim 2, characterized in that, The gradient magnitude The calculation method is as follows: right The gradient is convolved with a Sobel kernel to obtain the horizontal and vertical gradient components. The gradient magnitude is then calculated based on these components, expressed as follows: ； ； in, The horizontal convolution kernel matrix, This is the vertical convolution kernel matrix.

5. The method for fusing image perception and variable-order fractional-order motion control according to claim 2, characterized in that, The error function is constructed based on the three-dimensional coordinate information of the continuous target and the current position coordinates of the three-axis motor driving the target's motion. The expression is as follows: ； ； ； in: Let be the error function. A continuous three-dimensional coordinate sequence of the target object. This is the sequence of coordinates of the current position of the three-axis motor at time t. These represent adjacent sampling times.

6. The method for fusing image perception and variable-order fractional-order motion control according to claim 5, characterized in that, The nonlinear order function constructed based on the velocity derivative information of the target object and the error function is expressed as follows: ； in: Let be the error function. For the velocity derivative information of the target object, Let the order of the fractional integral operator be denoted as . This is for nonlinear order functions.

7. The method for fusing image perception and variable-order fractional-order motion control according to claim 6, characterized in that, The three-ring motion control structure includes: The inner current loop uses an integer-order PI controller for rapid adjustment of current error; The intermediate velocity loop employs a variable-order fractional-order proportional integral. The controller has the following structure: ； in, This is the speed loop proportional coefficient. For integral gain, For the velocity loop transfer function, For fractional integral operators, Let be the order of the fractional integral operator; The outermost position loop uses a proportional controller, and its structure is as follows: ； in, For the position loop transfer function, This is the position loop proportional gain.

8. The method for fusing image perception and variable-order fractional-order motion control according to claim 6, characterized in that, The process involves employing a variable-order fractional-order calculus controller in the velocity loop, combined with an improved polynomial fitting method to approximate the fractional-order operator, and using a bisection method to quickly determine the optimal fitting coefficients for the current order interval. This generates control inputs to drive the three-axis motor, achieving target tracking control. Specifically, this includes: Step (1): Select M uniformly distributed order points within the order range (0, 1) of the fractional integral operator. ,in , Step (2): In each order Next, using fractional integral operators As the core of the filter, its corresponding discrete-time transfer function is constructed. A unit impulse input is applied to obtain the filter's time-domain output response at that order. Furthermore, based on the filter's input and output data, the Prony algorithm is used to fit its difference equation and extract the corresponding discrete-time transfer function. Step (3): Construct the denominator coefficient matrix A based on the discrete transfer function in the time domain; Step (4): Construct the molecular coefficient matrix B based on the time-domain discrete transfer function; Step (5): Using the discrete point order X = [ , ,… [] is the input, representing the current order. Falling on adjacent discrete points and When the interval is between, take the corresponding denominator coefficient matrix respectively. and The column vectors are output column by column, and the two-point linear interpolation method is used for fitting calculation to obtain the N+1 polynomial of the denominator coefficients; Step (6): Using the discrete point order X = [ , ,… [] is the input, representing the current order. Falling on adjacent discrete points and When the values ​​are in between, take the corresponding molecule coefficient matrix respectively. and The column vectors are output column by column, and the two-point linear interpolation method is used for fitting calculation to obtain the N+1 polynomial of the molecular coefficients; Step (7): Combine the N+1 polynomials of the denominator coefficients and the N+1 polynomials of the numerator coefficients to obtain the time-domain transfer function of the corresponding fractional calculus in any region. Based on the real-time extraction of the target object's three-dimensional coordinate changes and the error feedback of the servo motor from the image acquisition module, calculate the motion control signal.

9. The method for fusing image perception and variable-order fractional-order motion control according to claim 8, characterized in that, The denominator coefficient matrix A specifically includes: ； in: Let A represent the Nth row and Mth column element of the denominator coefficient matrix A, which corresponds to the Nth order coefficient of the denominator polynomial at the order point M. The molecular coefficient matrix B specifically includes: ； in: Let represent the Nth row and Mth column element of the numerator coefficient matrix B, corresponding to the Nth order coefficient of the denominator polynomial at order point M.

10. The method for fusing image perception and variable-order fractional-order motion control according to claim 8, characterized in that, The two-point linear interpolation method described above is used for fitting calculations, specifically including: ； ； in: In order The denominator coefficients are below. For the first A known order point, For the current variable order being calculated, For order point The next The coefficient of the denominator is 1. For coefficient serial numbers, For order point index, In order The molecule coefficient below, For order point The next Molecular coefficient.

Images