A robot motion intelligent control method and system based on machine vision

Abstract

The application discloses a kind of robot motion intelligent control method and system based on machine vision, it is related to robot motion intelligent control technical field, by visual image acquisition processing module and ground texture detail extraction module, can make full use of ground crack, joint seam and special mark and so on texture details, form more stable visual feature base.Compared with the scheme of traditional dependence GPS or IMU, this design can still maintain higher positioning accuracy and stability in the environment of GPS signal loss, IMU drift accumulation is serious, reduce the deviation and misjudgment direction that robot appears in narrow space or repetitive trajectory task.Ground mark matching and feature point screening module not only extracts feature point, but also combines texture direction consistency index to carry out stability screening to feature point, eliminate weak feature point in unstable area, output high confidence ground mark matrix;Make robot can rely on reliable ground mark to carry out continuous self-positioning and trajectory keeping.

Description

Technical Field

[0001] This invention relates to the field of intelligent robot motion control technology, specifically to a robot motion intelligent control method and system based on machine vision. Background Technology

[0002] Robotics is one of the core research areas of intelligent equipment and automation systems, and it has been widely applied in various scenarios such as industry, warehousing, and service industries in recent years. Among the branches of robotics development, vision-based perception and positioning control has become a key direction for improving robot autonomy and operational accuracy. Especially in indoor or semi-structured environments, relying on vision sensors to acquire image frames and utilizing features such as ground cracks, seams, and special markings for environmental understanding and path correction is gradually becoming a core technology replacing traditional positioning methods.

[0003] Currently, vision-based robot localization and control mostly rely on auxiliary localization information provided by GPS or IMU, with the vision component often serving as a "supplement." This structure leads to robot localization heavily depending on the cumulative integration results of the IMU in confined spaces or areas lacking GPS signals. However, IMUs are prone to drift errors during long-term operation, and the role of the vision component is limited to local feature recognition, unable to establish a stable and continuous environmental feature map, nor can it maintain high-precision localization in repetitive trajectories or complex environments.

[0004] The reason for this deficiency is that existing visual positioning methods lack a systematic utilization of ground texture details. Feature point matching is limited to sporadic detection and fails to be deeply integrated with SLAM algorithms, resulting in the inability to form a highly recognizable "feature map" in the environment. Once GPS signals in the environment cannot provide a reference, or IMU drift accumulates to a certain extent, the robot may experience path deviation and loss of trajectory repeatability, or even get stuck or misjudge direction in confined spaces. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a robot motion intelligent control method and system based on machine vision, which solves the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solution: a robot motion intelligent control system based on machine vision, including a visual image acquisition and processing module, a ground texture detail extraction module, a landmark matching and feature point filtering module, a texture change rate and drift correction module, a positioning optimization and fusion module, and a control feedback adjustment module;

[0007] The visual image acquisition and processing module acquires continuous image frames through the camera mounted on the robot, processes them, and obtains the image frame set It(x, y).

[0008] The ground texture detail extraction module extracts features from the image frame set It(x,y), including texture orientation gradient Ga, texture gray-level variance Btex, edge density Gde, and texture orientation consistency index Ctex;

[0009] The landmark matching and feature point filtering module extracts feature points from the image frame set It(x,y) using a feature description algorithm, and performs stability filtering by combining them with the texture direction consistency index Ctex to obtain the landmark matrix Lt;

[0010] The texture change rate and drift correction module obtains the initial positioning coordinates Pslam by comparing the landmark matrix Lt of feature points in adjacent frames, detects the visual positioning drift trend, and calculates the drift correction coefficient Kcorr.

[0011] The positioning optimization fusion module fuses the acquired initial positioning coordinates Pslam with the drift correction coefficient Kcorr to calculate the final positioning coordinates Popt.

[0012] The control feedback adjustment module dynamically adjusts the robot's motion commands based on the acquired final positioning coordinates (Popt).

[0013] Preferably, the visual image acquisition and processing module includes a visual image acquisition unit and a brightness correction unit;

[0014] The visual image acquisition unit acquires images through the camera mounted on the robot to form the original image, and processes and corrects the brightness value of the original image through the inverted image optical response function to obtain the brightness value of the corrected original image.

[0015] The brightness value of the corrected original image is obtained by subtracting the average brightness value of the dark area of ​​the image from the brightness value of the original image, and then taking the ratio of the difference between the average brightness value of the bright area and the brightness value of the dark area; the pixel brightness is scaled to the [0,1] interval to standardize the brightness difference.

[0016] The brightness correction unit takes the brightness value of the original image after preliminary correction as a reference for each pixel (x, y), calculates the difference between the average brightness of the pixel (x, y) over a time window and the current pixel brightness, and multiplies this difference by the inter-frame smoothing factor to obtain a smoothing correction amount.

[0017] The obtained smoothing correction amount is added back to the original pixel value to obtain the final enhanced pixel brightness. The final enhanced pixel brightness of multiple frames is then fused to obtain the image frame set It(x, y).

[0018] Preferably, the ground texture detail extraction module includes a local texture feature analysis unit and a global texture consistency calculation unit;

[0019] The local texture feature analysis unit performs basic texture feature calculations on the image frame set It(x,y) and extracts the texture orientation gradient Ga, texture gray-level variance Btex, and edge density Gde;

[0020] The texture orientation gradient Ga is obtained by calculating the rate of change of brightness in the horizontal and vertical directions for each pixel (x, y) in the image frame set It(x, y), comparing the obtained rate of change of brightness, and then obtaining the local orientation angle of the texture through the arctangent function.

[0021] Texture grayscale variance Btex is obtained by calculating the squared difference between the brightness of each pixel and the average brightness of the local neighborhood, and then averaging the result; it is used to measure the texture complexity of the region.

[0022] Edge density (Gde) is calculated by counting the number of edge pixels detected within a given local window area, and then calculating the ratio of edge pixels to window area. It represents the degree of edge crowding in a local region.

[0023] Preferably, the global texture consistency calculation unit analyzes the stability of the overall texture direction based on the texture direction gradient Ga, and calculates and obtains the texture direction consistency index Ctex;

[0024] The texture orientation consistency index Ctex is obtained by taking the texture orientation gradient Ga of each pixel (x, y) in the image, subtracting it from the average orientation angle pGa of the entire image, calculating the cosine value and taking the absolute value, and finally averaging it over all pixels.

[0025] Determine directional differences in an image based on the value of the texture orientation consistency index Ctex;

[0026] When 0.5 ≤ texture orientation consistency index Ctex < 1, it indicates that the orientation difference in the image is small and the texture orientation is uniform.

[0027] When 0 < texture orientation consistency index Ctex ≤ 0.5, it indicates that there are large differences in orientation in the image and the texture is messy.

[0028] Preferably, the landmark matching and feature point filtering module includes a feature point extraction unit and a feature point stability filtering unit;

[0029] The feature point extraction unit extracts structurally significant feature points from the image frame set It(x,y) by using a feature description algorithm and performing joint analysis based on texture orientation gradient Ga, texture gray variance Btex, and edge density Gde, and obtains a saliency score Sfeat(x,y).

[0030] Among them, the feature points include the intersection of cracks and the corners of splice seams;

[0031] The saliency score Sfeat(x,y) is obtained by weighting and superimposing the absolute value of the texture orientation gradient Ga, the texture gray variance Btex, and the edge density Gde for each pixel (x,y).

[0032] Pixels with high scores are more likely to be prominent feature points such as crack intersections and seam corners.

[0033] The saliency scores Sfeat(x,y) of all feature points are summarized to form a feature point set. Then, the pixel feature points with saliency scores Sfeat(x,y) higher than the feature point threshold are selected to form a feature point candidate set Pcand.

[0034] Preferably, the feature point stability screening unit combines the acquired feature point candidate set Pcand with the texture orientation consistency index Ctex to screen out feature points in regions with chaotic texture orientation, assigns a confidence score R to the screened feature points, and outputs the landmark matrix Lt.

[0035] Each feature point is assigned a stability weight Wstab(x, y) by using the texture orientation consistency index Ctex and the texture orientation gradient Ga.

[0036] The stability weights Wstab(x, y) are obtained using the following formula:

[0037] ;

[0038] In the formula, Ctex(t) represents the texture orientation consistency index at time t, Ga(x,y) represents the texture orientation gradient of pixel (x,y), pGa represents the global average orientation gradient, and π represents pi.

[0039] The confidence score R is obtained by weighted fusion of the significance score Sfeat(x,y) and the stability weight Wstab(x,y) of each candidate point, and candidate points with confidence scores R higher than the preset confidence threshold TRx are officially marked as "landmark points", and the landmark matrix Lt is output.

[0040] Preferably, the texture change rate and drift correction module includes an inter-frame landmark motion estimation unit and a drift trend analysis and correction unit;

[0041] The inter-frame landmark motion estimation unit matches the landmark matrix Lt of feature points in adjacent frames and calculates the inter-frame displacement Δpj of the feature points;

[0042] The inter-frame displacement Δpj is obtained by calculating the difference in displacement in the horizontal and vertical directions for each feature point that is successfully matched in two adjacent frames; the initial positioning coordinates Pslam are obtained by estimating the robot's inter-frame displacement Δpj between two frames using the least squares method.

[0043] The initial positioning coordinates Pslam are obtained by averaging the inter-frame displacements of all matched feature points and superimposing them onto the coordinates Pslam(t-1) of the previous frame.

[0044] The drift trend analysis and correction unit calculates and obtains the feature point drift amount ΔPdr based on the inter-frame displacement Δpj;

[0045] The feature point drift ΔPdr is obtained by summing the squares of the inter-frame displacements Δpj of all feature points, averaging them, and then taking the square root.

[0046] The drift amount ΔPdr of the acquired feature points is combined with the texture direction consistency index Ctex to calculate the drift correction coefficient Kcorr.

[0047] The drift correction factor Kcorr is obtained using the following formula:

[0048] ;

[0049] In the formula, ΔPdr(t) represents the feature point drift at time t, and KC represents the rate of change of the main texture direction between frames. It is obtained by the ratio of the difference between the mean gradient of the texture direction Ga(t) at time t and the mean gradient of the texture direction Ga(t-1) at time t-1 to the time interval Δt between the two frames.

[0050] If the drift is large and the texture direction changes significantly, the denominator changes and Kcorr decreases, indicating that a large correction is needed.

[0051] If the drift is small and the texture direction consistency is high, Kcorr will increase, indicating high positioning reliability and small correction range.

[0052] Preferably, the positioning optimization fusion module includes a fusion offset calculation unit and a final positioning coordinate generation unit;

[0053] The fusion offset calculation unit combines the acquired initial positioning coordinates Pslam with the reference positioning coordinates Pref to obtain the positioning difference ΔP, and then adjusts the ratio using the drift correction coefficient Kcorr to obtain the fusion offset ΔPopt.

[0054] The positioning difference ΔP is obtained by the difference between the reference positioning coordinates Pref and the initial positioning coordinates Pslam;

[0055] The fusion offset ΔPopt is obtained by multiplying the positioning difference ΔP by the drift correction coefficient Kcorr;

[0056] If the drift correction coefficient Kcorr value is large, it indicates that the environmental texture consistency is high, the drift is small, and the difference values ​​are basically completely superimposed.

[0057] If the drift correction coefficient Kcorr value is small, it indicates that the drift is serious or the texture direction changes greatly. The amount of fusion will be compressed to avoid large-scale error correction.

[0058] The final positioning coordinate generation unit combines the initial positioning coordinates Pslam with the fused offset ΔPopt to obtain the final positioning coordinates Popt;

[0059] The final positioning coordinates Popp are obtained by summing the initial positioning coordinates Pslam and the fused offset ΔPopt.

[0060] Preferably, the control feedback adjustment module includes a positioning correction unit and a motion command generation unit;

[0061] The positioning correction unit compares the final positioning coordinates Popt with the target point coordinates Ptar to obtain the position deviation Epos, and corrects the robot's forward direction angle θcu and velocity Vco to obtain the direction correction value θcr and velocity correction value Vcr.

[0062] Position deviation Epos is obtained by the difference between the target point coordinates Ptar and the final positioning coordinates Popp.

[0063] The direction correction value θcr is obtained using the following formula:

[0064] ;

[0065] In the formula, θcr(t) represents the direction correction value at time t, arctan represents the arctangent function, Epos,y(t) represents the longitudinal component of the position deviation at time (t), and Epos,x(t) represents the lateral component of the position deviation at time (t).

[0066] The speed correction value Vcr is obtained using the following formula:

[0067] ;

[0068] In the formula, kv represents the range gain coefficient, and k1 represents the directional suppression coefficient;

[0069] The speed increases when the distance to the target point is large;

[0070] When the steering correction angle is large, reduce speed appropriately to ensure greater stability when turning;

[0071] The motion command generation unit integrates the acquired direction correction value θcr and velocity correction value Vcr into a robot motion control instruction set CMD, which includes the direction adjustment instruction θcmd and the velocity instruction Vcmd, and adjusts the robot's motion state.

[0072] The robot's motion state is obtained through matching in the following ways:

[0073] When the position deviation Epos is less than the preset error threshold Terr, the original state is maintained.

[0074] When the direction correction value θcr > the preset direction threshold Tθ, the robot's motion state is in the direction correction state.

[0075] When the direction correction value θcr > the preset direction threshold Tθ, and the position deviation Epos > the preset error threshold Terr, the robot's motion state enters the deceleration state.

[0076] A machine vision-based intelligent motion control method for robots includes the following steps:

[0077] Step 1: The visual image acquisition and processing module acquires continuous image frames through the camera mounted on the robot, processes them, and obtains the image frame set It(x, y).

[0078] Step 2: The ground texture detail extraction module extracts features from the image frame set It(x,y), including texture orientation gradient Ga, texture gray-level variance Btex, edge density Gde, and texture orientation consistency index Ctex;

[0079] Step 3: The landmark matching and feature point filtering module extracts feature points from the image frame set It(x,y) using a feature description algorithm, and performs stability filtering by combining them with the texture direction consistency index Ctex to obtain the landmark matrix Lt;

[0080] Step 4: The texture change rate and drift correction module obtains the initial positioning coordinates Pslam by comparing the landmark matrix Lt of feature points in adjacent frames, detects the visual positioning drift trend, and calculates the drift correction coefficient Kcorr.

[0081] Step 5: The localization optimization and fusion module fuses the acquired initial localization coordinates Pslam with the drift correction coefficient Kcorr to calculate and obtain the final localization coordinates Popt;

[0082] Step Six: The control feedback adjustment module dynamically adjusts the robot's motion commands based on the acquired final positioning coordinates (Popt).

[0083] This invention provides a robot motion intelligent control method and system based on machine vision, which has the following beneficial effects:

[0084] (1) During system operation, the visual image acquisition and processing module and the ground texture detail extraction module can make full use of texture details such as ground cracks, splicing seams and special markings to form a more stable visual feature base. Compared with traditional solutions that rely on GPS or IMU, this design can still maintain high positioning accuracy and stability in environments where GPS signals are missing or IMU drift is severe, reducing the robot's deviation and misjudgment of direction in narrow spaces or repetitive trajectory tasks.

[0085] The landmark matching and feature point filtering module not only extracts feature points but also uses the texture direction consistency index to filter feature points for stability, eliminating weak feature points in unstable areas and outputting a high-confidence landmark matrix. This design improves upon the limitations of existing visual positioning technologies that rely on scattered feature points and cannot construct effective feature maps, enabling robots to perform continuous self-localization and trajectory maintenance based on reliable landmarks.

[0086] (2) The brightness of the original image acquired by the camera is processed and corrected by inverting the image optical response function, instead of relying on traditional static calibration or fixed compensation methods. This dynamic correction method can adjust the brightness value in real time according to different lighting conditions and lens characteristics, eliminating the image distortion problem caused by changes in ambient light, making the input image data more accurate, and ensuring the reliability of subsequent texture extraction and feature point recognition.

[0087] The brightness correction unit calculates the average brightness difference within a time window at the pixel level, generates a smoothing correction amount using an inter-frame smoothing factor, and then adds the correction amount back to the original pixel value, achieving multi-frame brightness fusion. This process ensures a natural brightness transition between consecutive frames, effectively reducing "brightness abrupt changes" caused by flicker, uneven exposure, or lighting jumps, and ensuring the continuity of ground texture details at the visual input end.

[0088] (3) Feature points are extracted through joint analysis of texture direction gradient Ga, texture gray-level variance Btex, and edge density Gde. This not only identifies high-value features such as crack intersections and splice corners, but also generates a saliency score for each pixel, ensuring that only feature points with clear and stable structures are selected. Compared with traditional single-based corner or edge detection schemes, this multi-parameter joint screening method significantly reduces the interference of "noisy feature points," making the landmark points constructed by the system more accurate and providing a reliable data foundation for subsequent positioning calculations.

[0089] The feature point stability screening unit introduces a texture orientation consistency index Ctex, which is combined with the orientation gradient information of each feature point to assign a stability weight to each point, and then a confidence score is generated. Only candidate points with a confidence score higher than the threshold are officially marked as landmark points and output to the landmark matrix. This mechanism effectively avoids the erroneous selection of landmarks in areas with messy textures, making the final landmark matrix Lt more reliable.

[0090] (4) The initial positioning coordinates Pslam and the reference positioning coordinates Pref are compared by the fusion offset calculation unit, and the drift correction coefficient Kcorr is used for proportional adjustment to generate the fusion offset, which is then superimposed on the initial positioning to obtain the final positioning coordinates Popt. The control feedback adjustment module compares the final positioning coordinates with the target point in real time, calculates the position deviation Epos, and converts it into the direction correction value θcr and the velocity correction value Vcr, which are then integrated to generate motion control commands. The robot can not only automatically adjust its direction according to the relative relationship between its current position and the target point, but also dynamically accelerate or decelerate according to the magnitude of the deviation, realizing a control mode of "responding and adjusting at any time". This dynamic feedback mechanism makes the robot's path correction smoother and more natural, rather than abruptly performing large-scale operations. Attached Figure Description

[0091] Figure 1 This is a schematic diagram of the block diagram of a robot motion intelligent control system based on machine vision according to the present invention.

[0092] Figure 2 This is a schematic diagram illustrating the steps of a machine vision-based intelligent motion control method for robots according to the present invention.

[0093] Figure 3 This is a schematic diagram of the final positioning coordinate acquisition process of the present invention;

[0094] Figure 4 This is a line graph of the drift correction coefficient of the present invention. Detailed Implementation

[0095] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0096] Example 1

[0097] This invention provides a machine vision-based intelligent control system for robot motion. Please refer to [link / reference]. Figures 1 to 4It includes a visual image acquisition and processing module, a ground texture detail extraction module, a landmark matching and feature point filtering module, a texture change rate and drift correction module, a positioning optimization and fusion module, and a control feedback adjustment module;

[0098] The visual image acquisition and processing module acquires continuous image frames through the camera mounted on the robot, processes them, and obtains the image frame set It(x, y).

[0099] The ground texture detail extraction module extracts features from the image frame set It(x,y), including texture orientation gradient Ga, texture gray-level variance Btex, edge density Gde, and texture orientation consistency index Ctex;

[0100] The landmark matching and feature point filtering module extracts feature points from the image frame set It(x,y) using a feature description algorithm, and performs stability filtering by combining them with the texture direction consistency index Ctex to obtain the landmark matrix Lt;

[0101] The texture change rate and drift correction module obtains the initial positioning coordinates Pslam by comparing the landmark matrix Lt of feature points in adjacent frames, detects the visual positioning drift trend, and calculates the drift correction coefficient Kcorr.

[0102] The positioning optimization fusion module fuses the acquired initial positioning coordinates Pslam with the drift correction coefficient Kcorr to calculate the final positioning coordinates Popt.

[0103] The control feedback adjustment module dynamically adjusts the robot's motion commands based on the acquired final positioning coordinates (Popt).

[0104] In this embodiment, the visual image acquisition and processing module and the ground texture detail extraction module can fully utilize texture details such as ground cracks, seams, and special markings to form a more stable visual feature base. Compared with traditional solutions that rely on GPS or IMU, this design can maintain high positioning accuracy and stability in environments where GPS signals are missing or IMU drift is severe, reducing robot deviation and misjudgment of direction in confined spaces or repetitive trajectory tasks.

[0105] The landmark matching and feature point filtering module not only extracts feature points but also uses the texture direction consistency index to filter feature points for stability, eliminating weak feature points in unstable areas and outputting a high-confidence landmark matrix. This design improves upon the limitations of existing visual positioning technologies that rely on scattered feature points and cannot construct effective feature maps, enabling robots to perform continuous self-localization and trajectory maintenance based on reliable landmarks.

[0106] Through the texture change rate and drift correction module, the system can dynamically detect the drift trend of landmarks in consecutive frames and calculate drift correction coefficients to correct the initial localization, reducing the cumulative error of visual SLAM. Then, through the localization optimization and fusion module, the initial localization and correction data are fused, making the final localization coordinates more accurate and smoother, avoiding frequent path correction and task interruption caused by robot drift.

[0107] The control feedback adjustment module dynamically adjusts the robot's direction, speed, and acceleration / deceleration based on the final positioning coordinates, and transmits the execution status back in real time, forming a closed loop of positioning-control-feedback. This closed-loop control enables the robot to not only correct its motion path in real time, but also to adaptively adjust its motion strategy when the environment changes, reducing abnormal actions and safety risks caused by sudden deviations or unstable steering.

[0108] Example 2

[0109] This embodiment is an explanation based on Embodiment 1. Please refer to it. Figure 1 and Figure 3 Specifically: the visual image acquisition and processing module includes a visual image acquisition unit and a brightness correction unit;

[0110] The visual image acquisition unit acquires images through the camera mounted on the robot to form the original image, and processes and corrects the brightness value of the original image through the inverted image optical response function to obtain the brightness value of the corrected original image.

[0111] The brightness value of the corrected original image is obtained by subtracting the average brightness value of the dark area of ​​the image from the brightness value of the original image, and then taking the ratio of the difference between the average brightness value of the bright area and the brightness value of the dark area.

[0112] The brightness correction unit takes the brightness value of the original image after preliminary correction as a reference for each pixel (x, y), calculates the difference between the average brightness of the pixel (x, y) over a time window and the current pixel brightness, and multiplies this difference by the inter-frame smoothing factor to obtain a smoothing correction amount.

[0113] The obtained smoothing correction amount is added back to the original pixel value to obtain the final enhanced pixel brightness. The final enhanced pixel brightness of multiple frames is then fused to obtain the image frame set It(x, y).

[0114] In this embodiment, the visual image acquisition unit processes and corrects the brightness of the original image acquired by the camera by retrieving the image optical response function, rather than relying on traditional static calibration or fixed compensation methods. This dynamic correction method can adjust the brightness value in real time according to different lighting conditions and lens characteristics, eliminating image distortion caused by changes in ambient light, making the input image data more accurate, and ensuring the reliability of subsequent texture extraction and feature point recognition.

[0115] The brightness correction unit calculates the average brightness difference within a time window at the pixel level, generates a smoothing correction amount using an inter-frame smoothing factor, and then adds the correction amount back to the original pixel value, achieving multi-frame brightness fusion. This process ensures a natural brightness transition between consecutive frames, effectively reducing "brightness abrupt changes" caused by flicker, uneven exposure, or lighting jumps, and ensuring the continuity of ground texture details at the visual input end.

[0116] Because the system completes dynamic brightness correction and multi-frame brightness fusion during the image acquisition stage, the output image frame set It(x,y) is clearer, more stable, and retains more detail. This front-end processing method directly improves the extraction quality of texture features such as ground cracks, seams, and special markings by subsequent modules, reduces feature point drift and positioning errors caused by unstable image quality, and lays a solid foundation for robots to achieve high-precision self-localization in narrow spaces or semi-structured environments.

[0117] Example 3

[0118] This embodiment is an explanation based on Embodiment 2. Please refer to it. Figure 1 Specifically: the ground texture detail extraction module includes a local texture feature analysis unit and a global texture consistency calculation unit;

[0119] The local texture feature analysis unit performs basic texture feature calculations on the image frame set It(x,y) and extracts the texture orientation gradient Ga, texture gray-level variance Btex, and edge density Gde;

[0120] The texture orientation gradient Ga is obtained by calculating the rate of change of brightness in the horizontal and vertical directions for each pixel (x, y) in the image frame set It(x, y), comparing the obtained rate of change of brightness, and then obtaining the local orientation angle of the texture through the arctangent function.

[0121] The texture grayscale variance Btex is obtained by calculating the squared difference between the brightness of each pixel and the average brightness of the local neighborhood, and then averaging the result.

[0122] Edge density Gde is obtained by counting the number of edge pixels detected within a given local window area, and then calculating the ratio of edge pixels to window area.

[0123] The global texture consistency calculation unit analyzes the stability of the overall texture direction based on the texture direction gradient Ga and calculates the texture direction consistency index Ctex.

[0124] The texture orientation consistency index Ctex is obtained by taking the texture orientation gradient Ga of each pixel (x, y) in the image, subtracting it from the average orientation angle pGa of the entire image, calculating the cosine value and taking the absolute value, and finally averaging it over all pixels.

[0125] Determine directional differences in an image based on the value of the texture orientation consistency index Ctex;

[0126] When 0.5 ≤ texture orientation consistency index Ctex < 1, it indicates that the orientation difference in the image is small and the texture orientation is uniform.

[0127] When 0 < texture orientation consistency index Ctex ≤ 0.5, it indicates that there are large differences in orientation in the image and the texture is messy.

[0128] In this embodiment, the local texture feature analysis unit not only extracts three basic features—texture direction gradient, texture grayscale variance, and edge density—but also analyzes the horizontal and vertical variation trends of brightness at the pixel level, statistically analyzes local brightness differences and edge distribution density, and characterizes the structural characteristics of the ground from multiple dimensions. This multi-dimensional analysis method overcomes the limitations of single feature point recognition, enabling the robot to capture more complete ground detail information, including cracks, seams, and edge variation areas, providing a richer and more accurate data foundation for subsequent landmark extraction and SLAM localization.

[0129] The global texture consistency calculation unit analyzes the stability of the overall texture direction based on the texture orientation gradient of the entire image, calculates the texture orientation consistency index Ctex, and provides a clear criterion for judging orientation differences. A higher value indicates that the ground orientation texture is uniform and suitable for reliable landmark extraction.

[0130] A low value indicates a cluttered texture and the presence of unstable feature points. This mechanism provides a scientific basis for subsequent feature point selection, allowing the system to automatically determine which areas' information can be used for localization and which areas should be ignored.

[0131] Through a dual analysis of "local features + global consistency," this module can not only generate high-quality texture direction, brightness variation, and edge density data, but also provide an overall index, Ctex, to measure texture reliability. This allows the robot to dynamically adjust the weighting of feature points during visual SLAM or landmark matching, reducing the use of cluttered or unreliable visual information and mitigating the risk of positioning drift and error accumulation from the outset.

[0132] Example 4

[0133] This embodiment is an explanation based on Embodiment 3. Please refer to it. Figure 4 Specifically: the landmark matching and feature point filtering module includes a feature point extraction unit and a feature point stability filtering unit;

[0134] The feature point extraction unit extracts structurally significant feature points from the image frame set It(x,y) by using a feature description algorithm and performing joint analysis based on texture orientation gradient Ga, texture gray variance Btex, and edge density Gde, and obtains a saliency score Sfeat(x,y).

[0135] Among them, the feature points include the intersection of cracks and the corners of splice seams;

[0136] The saliency score Sfeat(x,y) is obtained by weighting and superimposing the absolute value of the texture orientation gradient Ga, the texture gray variance Btex, and the edge density Gde for each pixel (x,y).

[0137] The saliency scores Sfeat(x,y) of all feature points are summarized to form a feature point set. Then, the pixel feature points with saliency scores Sfeat(x,y) higher than the feature point threshold are selected to form a feature point candidate set Pcand.

[0138] The feature point stability screening unit combines the acquired feature point candidate set Pcand with the texture orientation consistency index Ctex to screen out feature points in regions with chaotic texture orientation, assigns a confidence score R to the screened feature points, and outputs the landmark matrix Lt.

[0139] Each feature point is assigned a stability weight Wstab(x, y) by using the texture orientation consistency index Ctex and the texture orientation gradient Ga.

[0140] The stability weights Wstab(x, y) are obtained using the following formula:

[0141] ;

[0142] In the formula, Ctex(t) represents the texture orientation consistency index at time t, Ga(x,y) represents the texture orientation gradient of pixel (x,y), pGa represents the global average orientation gradient, and π represents pi.

[0143] The confidence score R is obtained by weighted fusion of the significance score Sfeat(x,y) and stability weight Wstab(x,y) of each candidate point, and candidate points with confidence scores R higher than the preset confidence threshold TRx are officially marked as landmark points, and the landmark matrix Lt is output.

[0144] The texture change rate and drift correction module includes an inter-frame landmark motion estimation unit and a drift trend analysis and correction unit;

[0145] The inter-frame landmark motion estimation unit matches the landmark matrix Lt of feature points in adjacent frames and calculates the inter-frame displacement Δpj of the feature points;

[0146] The inter-frame displacement Δpj is obtained by calculating the difference in displacement in the horizontal and vertical directions for each feature point that is successfully matched in two adjacent frames; the initial positioning coordinates Pslam are obtained by estimating the robot's inter-frame displacement Δpj between two frames using the least squares method.

[0147] The initial positioning coordinates Pslam are obtained by averaging the inter-frame displacements of all matched feature points and superimposing them onto the coordinates Pslam(t-1) of the previous frame.

[0148] The drift trend analysis and correction unit calculates and obtains the feature point drift amount ΔPdr based on the inter-frame displacement Δpj;

[0149] The feature point drift ΔPdr is obtained by summing the squares of the inter-frame displacements Δpj of all feature points, averaging them, and then taking the square root.

[0150] The drift amount ΔPdr of the acquired feature points is combined with the texture direction consistency index Ctex to calculate the drift correction coefficient Kcorr.

[0151] The drift correction factor Kcorr is obtained using the following formula:

[0152] ;

[0153] In the formula, ΔPdr(t) represents the feature point drift at time t, and KC represents the rate of change of the main texture direction between frames. It is obtained by the ratio of the difference between the mean gradient of the texture direction Ga(t) at time t and the mean gradient of the texture direction Ga(t-1) at time t-1 to the time interval Δt between the two frames.

[0154] Specific examples:

[0155] Table 1: Calculation table of drift correction coefficients;

[0156] Group number Ctex(t) ΔPdr(t) KC Kcorr Group 1 0.92 0.05 0.9 0.880 Group 2 0.85 0.08 1.1 0.781 Group 3 0.78 0.12 1.3 0.674 Group 4 0.95 0.03 0.8 0.927 Group 5 0.88 0.07 1.0 0.822

[0157] In this embodiment, feature points are extracted through joint analysis of texture orientation gradient Ga, texture gray-level variance Btex, and edge density Gde. This not only identifies high-value features such as crack intersections and splice seam corners, but also generates a saliency score for each pixel, ensuring that only structurally clear and stable feature points are selected. Compared to traditional schemes based solely on corner or edge detection, this multi-parameter joint screening method significantly reduces the interference of "noisy feature points," making the landmark points constructed by the system more accurate and providing a reliable data foundation for subsequent localization calculations.

[0158] The feature point stability screening unit introduces a texture orientation consistency index Ctex, which is combined with the orientation gradient information of each feature point to assign a stability weight to each point, and then a confidence score is generated. Only candidate points with a confidence score higher than the threshold are officially marked as landmark points and output to the landmark matrix. This mechanism effectively avoids the erroneous selection of landmarks in areas with messy textures, making the final landmark matrix Lt more reliable.

[0159] In this embodiment, the texture change rate and drift correction module calculates the inter-frame displacement of feature points by matching the landmark matrices of adjacent frames, and uses the least squares method to obtain the robot's initial localization coordinates Pslam in consecutive frames. The drift trend analysis unit calculates the drift correction coefficient Kcorr based on the feature point drift amount ΔPdr and the texture orientation consistency index Ctex. This means that the system can identify the cumulative trend of localization error in real time during visual SLAM and perform dynamic correction. This can significantly reduce the path deviation and localization drift of the robot after long-term operation.

[0160] Because this embodiment tightly integrates "saliency screening of feature points" with "stability judgment" and incorporates drift detection and correction, the robot's initial positioning coordinates (Pslam) are already subject to "drift control" during the generation process. Combined with subsequent Kcorr correction, this ensures that the robot's positioning information is not only continuous but also adaptive to environmental changes. This allows the robot to maintain high-precision positioning in areas where GPS is unavailable, such as narrow passages and semi-structured terrain, without experiencing directional jumps or trajectory chaos.

[0161] Example 5

[0162] This embodiment is an explanation based on Embodiment 4. Please refer to it. Figure 3 Specifically: the positioning optimization and fusion module includes a fusion offset calculation unit and a final positioning coordinate generation unit;

[0163] The fusion offset calculation unit combines the acquired initial positioning coordinates Pslam with the reference positioning coordinates Pref to obtain the positioning difference ΔP, and then adjusts the ratio using the drift correction coefficient Kcorr to obtain the fusion offset ΔPopt.

[0164] The positioning difference ΔP is obtained by the difference between the reference positioning coordinates Pref and the initial positioning coordinates Pslam;

[0165] The fusion offset ΔPopt is obtained by multiplying the positioning difference ΔP by the drift correction coefficient Kcorr;

[0166] The final positioning coordinate generation unit combines the initial positioning coordinates Pslam with the fused offset ΔPopt to obtain the final positioning coordinates Popt;

[0167] The final positioning coordinates Popp are obtained by summing the initial positioning coordinates Pslam and the fused offset ΔPopt.

[0168] The control feedback adjustment module includes a positioning correction unit and a motion command generation unit;

[0169] The positioning correction unit compares the final positioning coordinates Popt with the target point coordinates Ptar to obtain the position deviation Epos, and corrects the robot's forward direction angle θcu and velocity Vco to obtain the direction correction value θcr and velocity correction value Vcr.

[0170] Position deviation Epos is obtained by the difference between the target point coordinates Ptar and the final positioning coordinates Popp.

[0171] The direction correction value θcr is obtained using the following formula:

[0172] ;

[0173] In the formula, θcr(t) represents the direction correction value at time t, arctan represents the arctangent function, Epos,y(t) represents the longitudinal component of the position deviation at time (t), and Epos,x(t) represents the lateral component of the position deviation at time (t).

[0174] The speed correction value Vcr is obtained using the following formula:

[0175] ;

[0176] In the formula, kv represents the range gain coefficient, and k1 represents the directional suppression coefficient;

[0177] The motion command generation unit integrates the acquired direction correction value θcr and velocity correction value Vcr into a robot motion control instruction set CMD, which includes the direction adjustment instruction θcmd and the velocity instruction Vcmd, and adjusts the robot's motion state.

[0178] The robot's motion state is obtained through matching in the following ways:

[0179] When the position deviation Epos is less than the preset error threshold Terr, the original state is maintained.

[0180] When the direction correction value θcr > the preset direction threshold Tθ, the robot's motion state is in the direction correction state.

[0181] When the direction correction value θcr > the preset direction threshold Tθ, and the position deviation Epos > the preset error threshold Terr, the robot's motion state enters the deceleration state.

[0182] In this embodiment, the initial positioning coordinates Pslam and the reference positioning coordinates Pref are compared by a fusion offset calculation unit. A drift correction coefficient Kcorr is used for proportional adjustment to generate a fusion offset, which is then superimposed on the initial positioning to obtain the final positioning coordinates Popt. The control feedback adjustment module compares the final positioning coordinates with the target point in real time, calculates the position deviation Epos, and converts it into a direction correction value θcr and a velocity correction value Vcr. These are then integrated to generate motion control commands. The robot can not only automatically adjust its direction based on the relative relationship between its current position and the target point, but also dynamically accelerate or decelerate based on the magnitude of the deviation, achieving a "respond and adjust in real time" control method. This dynamic feedback mechanism allows the robot to perform path corrections more smoothly and naturally, rather than abruptly performing large-scale operations.

[0183] The system incorporates automatic logic for determining the robot's motion state, enabling it to automatically switch between "hold state," "direction correction state," or "deceleration state" based on different positional deviations and direction correction values. For example, when the robot approaches a target point, it maintains its original state to avoid unnecessary movements; when turning is required, it automatically decelerates to ensure stable turning; and when the deviation is large, it simultaneously performs direction adjustment and deceleration. This multi-state closed-loop control not only improves the robot's operational flexibility but also reduces safety hazards during high-speed operation.

[0184] Example 6

[0185] A machine vision-based intelligent control method for robot motion; please refer to [reference needed]. Figure 2 Specifically, it includes the following steps:

[0186] Step 1: The visual image acquisition and processing module acquires continuous image frames through the camera mounted on the robot, processes them, and obtains the image frame set It(x, y).

[0187] Step 2: The ground texture detail extraction module extracts features from the image frame set It(x,y), including texture orientation gradient Ga, texture gray-level variance Btex, edge density Gde, and texture orientation consistency index Ctex;

[0188] Step 3: The landmark matching and feature point filtering module extracts feature points from the image frame set It(x,y) using a feature description algorithm, and performs stability filtering by combining them with the texture direction consistency index Ctex to obtain the landmark matrix Lt;

[0189] Step 4: The texture change rate and drift correction module obtains the initial positioning coordinates Pslam by comparing the landmark matrix Lt of feature points in adjacent frames, detects the visual positioning drift trend, and calculates the drift correction coefficient Kcorr.

[0190] Step 5: The localization optimization and fusion module fuses the acquired initial localization coordinates Pslam with the drift correction coefficient Kcorr to calculate and obtain the final localization coordinates Popt;

[0191] Step Six: The control feedback adjustment module dynamically adjusts the robot's motion commands based on the acquired final positioning coordinates Popp.

[0192] In this embodiment, the method uses continuous image frames captured by the robot's onboard camera as the core data source. Through a visual image acquisition and processing module and a ground texture detail extraction module, key texture features such as ground cracks, seams, and special markings are extracted. This allows the robot to reduce its over-reliance on external devices such as GPS or IMU, and instead form a stable visual feature map based on its own perception in indoor or semi-structured scenes, laying a high-precision foundation for subsequent positioning calculations.

[0193] In the landmark matching and feature point selection module, the method extracts structurally significant feature points through a feature description algorithm and combines this with a texture direction consistency index for stability assessment, retaining only landmark points with high credibility to form a landmark matrix. This step significantly reduces misidentification problems caused by texture chaos, uneven lighting, or noisy points, ensuring that the feature points used for subsequent positioning are more reliable, thus improving positioning accuracy and robustness from the source.

[0194] A drift detection mechanism is introduced into the texture change rate and drift correction module. The initial positioning coordinates are calculated by comparing the landmark matrix of consecutive frames, and drift correction coefficients are generated based on the drift trend to correct the cumulative error of SLAM in a timely manner. This means that the robot will not exhibit significant directional drift or positioning deviation during long-term operation or in complex environments, ensuring the continuity of the operation path.

[0195] The control feedback adjustment module uses the final positioning coordinates to adjust the robot's direction, speed, and acceleration / deceleration in real time, and triggers direction correction or deceleration strategies based on deviations. This dynamic feedback allows the robot to not only correct its path in a timely manner during task execution, but also flexibly adjust its movement mode according to its positioning status, ultimately forming a closed loop of positioning-control-feedback, enabling the robot to have adaptive adjustment and safety protection capabilities.

[0196] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A robot motion intelligent control system based on machine vision, characterized in that: It includes a visual image acquisition and processing module, a ground texture detail extraction module, a landmark matching and feature point filtering module, a texture change rate and drift correction module, a positioning optimization and fusion module, and a control feedback adjustment module; The visual image acquisition and processing module acquires continuous image frames through the camera mounted on the robot, processes them, and obtains the image frame set It(x, y). The ground texture detail extraction module extracts features from the image frame set It(x,y), including texture orientation gradient Ga, texture gray-level variance Btex, edge density Gde, and texture orientation consistency index Ctex; The landmark matching and feature point filtering module extracts feature points from the image frame set It(x,y) using a feature description algorithm, and performs stability filtering by combining them with the texture direction consistency index Ctex to obtain the landmark matrix Lt; The landmark matching and feature point filtering module includes a feature point extraction unit and a feature point stability filtering unit; The feature point stability screening unit combines the acquired feature point candidate set Pcand with the texture orientation consistency index Ctex to screen out feature points in regions with chaotic texture orientation, assigns a confidence score R to the screened feature points, and outputs the landmark matrix Lt. Each feature point is assigned a stability weight Wstab(x, y) by using the texture orientation consistency index Ctex and the texture orientation gradient Ga. The stability weights Wstab(x, y) are obtained using the following formula: ； In the formula, Ctex(t) represents the texture orientation consistency index at time t, Ga(x,y) represents the texture orientation gradient of pixel (x,y), pGa represents the global average orientation gradient, and π represents pi. The confidence score R is obtained by weighted fusion of the significance score Sfeat(x,y) and the stability weight Wstab(x,y) of each candidate point, and candidate points with confidence scores R higher than the preset confidence threshold TRx are officially marked as landmark points, and the landmark matrix Lt is output. The texture change rate and drift correction module obtains the initial positioning coordinates Pslam by comparing the landmark matrix Lt of feature points in adjacent frames, detects the visual positioning drift trend, and calculates the drift correction coefficient Kcorr. The positioning optimization fusion module fuses the acquired initial positioning coordinates Pslam with the drift correction coefficient Kcorr to calculate the final positioning coordinates Popt. The control feedback adjustment module dynamically adjusts the robot's motion commands based on the acquired final positioning coordinates (Popt).

2. The intelligent robot motion control system based on machine vision according to claim 1, characterized in that: The visual image acquisition and processing module includes a visual image acquisition unit and a brightness correction unit; The visual image acquisition unit acquires images through the camera mounted on the robot to form the original image, and processes and corrects the brightness value of the original image through the inverted image optical response function to obtain the brightness value of the corrected original image. The brightness value of the corrected original image is obtained by subtracting the average brightness value of the dark area of ​​the image from the brightness value of the original image, and then taking the ratio of the difference between the average brightness value of the bright area and the brightness value of the dark area. The brightness correction unit takes the brightness value of the original image after preliminary correction as a reference for each pixel (x, y), calculates the difference between the average brightness of the pixel (x, y) over a time window and the current pixel brightness, and multiplies this difference by the inter-frame smoothing factor to obtain a smoothing correction amount. The obtained smoothing correction amount is added back to the original pixel value to obtain the final enhanced pixel brightness. The final enhanced pixel brightness of multiple frames is then fused to obtain the image frame set It(x, y).

3. The intelligent robot motion control system based on machine vision according to claim 2, characterized in that: The ground texture detail extraction module includes a local texture feature analysis unit and a global texture consistency calculation unit; The local texture feature analysis unit performs basic texture feature calculations on the image frame set It(x,y) and extracts the texture orientation gradient Ga, texture gray-level variance Btex, and edge density Gde; The texture orientation gradient Ga is obtained by calculating the rate of change of brightness in the horizontal and vertical directions for each pixel (x, y) in the image frame set It(x, y), comparing the obtained rate of change of brightness, and then obtaining the local orientation angle of the texture through the arctangent function. The texture grayscale variance Btex is obtained by calculating the squared difference between the brightness of each pixel and the average brightness of the local neighborhood, and then averaging the result. Edge density Gde is obtained by counting the number of edge pixels detected within a given local window area, and then calculating the ratio of edge pixels to window area.

4. The robot motion intelligent control system based on machine vision according to claim 3, characterized in that: The global texture consistency calculation unit analyzes the stability of the overall texture direction based on the texture direction gradient Ga and calculates the texture direction consistency index Ctex. The texture orientation consistency index Ctex is obtained by taking the texture orientation gradient Ga of each pixel (x, y) in the image, subtracting it from the average orientation angle pGa of the entire image, calculating the cosine value and taking the absolute value, and finally averaging it over all pixels. Determine directional differences in an image based on the value of the texture orientation consistency index Ctex; When 0.5 ≤ texture orientation consistency index Ctex < 1, it indicates that the orientation difference in the image is small and the texture orientation is uniform. When 0 < texture orientation consistency index Ctex ≤ 0.5, it indicates that there are large differences in orientation in the image and the texture is messy.

5. The intelligent robot motion control system based on machine vision according to claim 4, characterized in that: The feature point extraction unit extracts structurally significant feature points from the image frame set It(x,y) by using a feature description algorithm and performing joint analysis based on texture orientation gradient Ga, texture gray variance Btex, and edge density Gde, and obtains a saliency score Sfeat(x,y). Among them, the feature points include the intersection of cracks and the corners of splice seams; The saliency score Sfeat(x,y) is obtained by weighting and superimposing the absolute value of the texture orientation gradient Ga, the texture gray variance Btex, and the edge density Gde for each pixel (x,y). The saliency scores Sfeat(x,y) of all feature points are summarized to form a feature point set. Then, the pixel feature points with saliency scores Sfeat(x,y) higher than the feature point threshold are selected to form a feature point candidate set Pcand.

6. The intelligent robot motion control system based on machine vision according to claim 5, characterized in that: The texture change rate and drift correction module includes an inter-frame landmark motion estimation unit and a drift trend analysis and correction unit; The inter-frame landmark motion estimation unit matches the landmark matrix Lt of feature points in adjacent frames and calculates the inter-frame displacement Δpj of the feature points; The inter-frame displacement Δpj is obtained by calculating the difference in displacement in the horizontal and vertical directions for each feature point that is successfully matched in two adjacent frames; the initial positioning coordinates Pslam are obtained by estimating the robot's inter-frame displacement Δpj between two frames using the least squares method. The initial positioning coordinates Pslam are obtained by averaging the inter-frame displacements of all matched feature points and superimposing them onto the coordinates Pslam(t-1) of the previous frame. The drift trend analysis and correction unit calculates and obtains the feature point drift amount ΔPdr based on the inter-frame displacement Δpj; The feature point drift ΔPdr is obtained by summing the squares of the inter-frame displacements Δpj of all feature points, averaging them, and then taking the square root. The drift amount ΔPdr of the acquired feature points is combined with the texture direction consistency index Ctex to calculate the drift correction coefficient Kcorr. The drift correction factor Kcorr is obtained using the following formula: ； In the formula, ΔPdr(t) represents the feature point drift at time t, and KC represents the rate of change of the main texture direction between frames. It is obtained by the ratio of the difference between the mean gradient of the texture direction Ga(t) at time t and the mean gradient of the texture direction Ga(t-1) at time t-1 to the time interval Δt between the two frames.

7. The robot motion intelligent control system based on machine vision according to claim 6, characterized in that: The positioning optimization and fusion module includes a fusion offset calculation unit and a final positioning coordinate generation unit; The fusion offset calculation unit combines the acquired initial positioning coordinates Pslam with the reference positioning coordinates Pref to obtain the positioning difference ΔP, and then adjusts the ratio using the drift correction coefficient Kcorr to obtain the fusion offset ΔPopt. The positioning difference ΔP is obtained by the difference between the reference positioning coordinates Pref and the initial positioning coordinates Pslam; The fusion offset ΔPopt is obtained by multiplying the positioning difference ΔP by the drift correction coefficient Kcorr; The final positioning coordinate generation unit combines the initial positioning coordinates Pslam with the fused offset ΔPopt to obtain the final positioning coordinates Popt; The final positioning coordinates Popp are obtained by summing the initial positioning coordinates Pslam and the fused offset ΔPopt.

8. The robot motion intelligent control system based on machine vision according to claim 7, characterized in that: The control feedback adjustment module includes a positioning correction unit and a motion command generation unit; The positioning correction unit compares the final positioning coordinates Popt with the target point coordinates Ptar to obtain the position deviation Epos, and corrects the robot's forward direction angle θcu and velocity Vco to obtain the direction correction value θcr and velocity correction value Vcr. Position deviation Epos is obtained by the difference between the target point coordinates Ptar and the final positioning coordinates Popp. The direction correction value θcr is obtained using the following formula: ； In the formula, θcr(t) represents the direction correction value at time t, arctan represents the arctangent function, Epos,y(t) represents the longitudinal component of the position deviation at time (t), and Epos,x(t) represents the lateral component of the position deviation at time (t). The speed correction value Vcr is obtained using the following formula: ； In the formula, kv represents the range gain coefficient, and k1 represents the directional suppression coefficient; The motion command generation unit integrates the acquired direction correction value θcr and velocity correction value Vcr into a robot motion control instruction set CMD, which includes the direction adjustment instruction θcmd and the velocity instruction Vcmd, and adjusts the robot's motion state. The robot's motion state is obtained through matching in the following ways: When the position deviation Epos is less than the preset error threshold Terr, the original state is maintained. When the direction correction value θcr > the preset direction threshold Tθ, the robot's motion state is in the direction correction state. When the direction correction value θcr > the preset direction threshold Tθ, and the position deviation Epos > the preset error threshold Terr, the robot's motion state enters the deceleration state.

9. A machine vision-based intelligent robot motion control method, applied to the machine vision-based intelligent robot motion control system described in any one of claims 1 to 8, characterized in that: Includes the following steps: Step 1: The visual image acquisition and processing module acquires continuous image frames through the camera mounted on the robot, processes them, and obtains the image frame set It(x, y). Step 2: The ground texture detail extraction module extracts features from the image frame set It(x,y), including texture orientation gradient Ga, texture gray-level variance Btex, edge density Gde, and texture orientation consistency index Ctex; Step 3: The landmark matching and feature point filtering module extracts feature points from the image frame set It(x,y) using a feature description algorithm, and performs stability filtering by combining them with the texture direction consistency index Ctex to obtain the landmark matrix Lt; Step 4: The texture change rate and drift correction module obtains the initial positioning coordinates Pslam by comparing the landmark matrix Lt of feature points in adjacent frames, detects the visual positioning drift trend, and calculates the drift correction coefficient Kcorr. Step 5: The localization optimization and fusion module fuses the acquired initial localization coordinates Pslam with the drift correction coefficient Kcorr to calculate and obtain the final localization coordinates Popt; Step Six: The control feedback adjustment module dynamically adjusts the robot's motion commands based on the acquired final positioning coordinates Popp.

Images