Automated tire conveying system and method based on machine vision

Abstract

The present application relates to the technical field of automatic sorting, in particular to a tire automatic conveying system and method based on machine vision, the system comprising an image locking module, a texture contrast module, a region guiding module, a path screening module and a label binding module.In the present application, tire images are acquired by an image acquisition device, the tread texture features are extracted, the texture distribution length and the number of jumps are calculated, the guiding cross-section diagram is generated, the stable image label is generated by comparing the profile lateral offset value and the number of jumps, the long strip region is divided according to the classification result, the line segment slope and the same direction segment length continuity are extracted, the priority sequence is generated by sorting, the path selection comprehensive task queue number and the waiting time are screened, the path number is screened and the classification information is bound, the path delivery combination number is formed, the path scheduling and label management are realized through the mechanical arm movement and label printing, and the path planning accuracy and automation level are improved.

Description

Technical Field

[0001] This invention relates to the field of automated sorting technology, and more particularly to an automated tire conveying system and method based on machine vision. Background Technology

[0002] The field of automated sorting technology encompasses logistics systems based on item feature recognition and automated conveying devices, and is a crucial component of modern logistics automation and intelligent manufacturing. The core of this technology lies in the automatic identification of target objects through sensing technology, feeding the identification results back to the control system, whereby the execution mechanism completes the sorting operation. The system typically includes image acquisition devices, a recognition and processing unit, a conveying structure, and sorting execution devices. By detecting and judging targets through image sensing, edge recognition, and object matching, combined with positioning control and timing response, it achieves path scheduling and precise delivery of objects. This technology is widely used in express delivery, warehousing, and manufacturing, forming a fully automated technology system from identification to action execution.

[0003] Among them, the machine vision-based automated tire conveying system refers to an automated system that uses image acquisition devices to acquire image data of tires during the conveying process, determines the tire's position, posture, and type information through image recognition, and drives the conveying device to transport tires along a specified path based on the information instructions. This system mainly addresses the conveying and sorting problems of tires on production lines and logistics lines. It uses image capture, shape recognition, and label recognition to complete object detection, and then uses control signals to drive a linear conveyor belt and orientation adjustment device to complete the conveying path change and direction guidance. Combined with sensor position judgment and synchronous control, it achieves fully automated execution of the entire process from tire identification to handling.

[0004] While existing technologies in logistics automation and intelligent manufacturing have achieved item feature recognition and automated conveying, they suffer from significant shortcomings in feature extraction and path optimization. During item recognition, image acquisition equipment is susceptible to interference from ambient lighting and background changes in complex scenes, leading to unstable texture extraction and difficulty in forming symmetrical and stable texture blocks, thus affecting subsequent recognition accuracy. The ability to extract texture variations and transitions is weak, particularly for the complex textures of tire surfaces, where stable feature extraction and symmetry judgment are impossible, resulting in large texture matching errors. In path planning, there is a lack of comprehensive consideration of task queues and waiting times, resulting in simplistic path selection that cannot dynamically respond to task backlogs and path congestion, impacting tire conveying efficiency. In labeling operations, the lack of effective integration of image classification labels and path numbers leads to insufficient labeling accuracy. Furthermore, the absence of barcode logs for operation recording makes it difficult to trace and adjust the operation process. In terms of the dynamism and flexibility of tire transportation, existing systems struggle to respond in real time to contour offsets and texture changes. Path change operations are inflexible and unable to adapt to changes in tire posture during transportation, resulting in low levels of intelligence in transportation path planning and label operation. This fails to meet the requirements of modern logistics systems for high efficiency, precision, and dynamic adjustment. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology and to propose an automated tire delivery system and method based on machine vision.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a machine vision-based automated tire delivery system includes:

[0007] The image locking module acquires the tire image, reads the pixel set in the lateral tangent direction of the tire tread, calculates the continuous length of the texture distribution and the number of density jumps, extracts the central texture region, identifies the jump boundaries, filters out short abrupt line segments, and generates a texture-guided cross-sectional image.

[0008] The texture comparison module reads the continuous texture of the texture-guided cross-section map, obtains the texture width map in the dynamic image sequence, aligns the pixel boundaries, compares the contour lateral offset value with the number of jumps, and generates an image type locking label;

[0009] The region guidance module locks the label according to the image type, divides the strip region along the main axis in the image, extracts the slope of the contour direction line segment, calculates the total length of the segment with the same slope, determines whether the length of the continuous segment exceeds the limit, and sorts and generates a sorting priority sequence.

[0010] The path filtering module reads the first two bands of the sorting priority sequence, obtains the number of paths in queue and the waiting time, removes paths with excessive waiting time, and generates a delivery combination number.

[0011] The label binding module controls the robotic arm to move to the path line according to the delivery combination number, prints image labels, attaches them to the tires, records the numbers to the barcode log, and generates an automated tire delivery solution.

[0012] As a further embodiment of the present invention, the texture-guided cross-sectional image includes a central texture region, texture blocks in left and right symmetrical directions, and transition boundaries; the image type locking label includes cross-sectional texture comparison results, texture width image, lateral offset value of contour region, and number of transitions; the sorting priority sequence includes long strip regions, line segment slope values, total length of segments with the same slope, and continuity of segment length in the same direction; the delivery combination number includes path number, image classification information, number of tasks in queue, and waiting time; and the automated tire conveying scheme includes numbered paths, image classification identifiers, robotic arm trajectory lines, label identifiers, action numbers, and barcode logs.

[0013] As a further aspect of the present invention, the image locking module includes:

[0014] The image acquisition submodule acquires an image of the tire located below the positioning and recognition area, reads the gray value sequence of pixels in the horizontal tangent direction, records the gray value continuity length and counts the number of jumps to obtain the horizontal texture continuity length value.

[0015] The texture extraction submodule extracts the densely abrupt regions based on the horizontal texture continuity length value, detects the abrupt change amplitude value and span value of each abrupt boundary segment and sets a threshold, deletes short line segments that are both less than the abrupt change amplitude threshold and the span threshold, integrates the remaining boundary segment information, and obtains the stable texture interval position value.

[0016] The guidance generation submodule calls the jump sequence in the left and right symmetrical directions in the stable texture interval position value, filters the symmetrical texture blocks whose jump point spacing change amplitude is less than the jump stability threshold, extracts the horizontal sequence and directional trend value, and generates a texture guidance cross-section map.

[0017] As a further aspect of the present invention, the texture comparison module includes:

[0018] The section extraction submodule obtains the continuous texture pattern on the section in the texture-guided section image, identifies the texture boundary position in the horizontal direction, extracts the width sequence value of the continuous texture region, establishes a corresponding index structure between the section position and the texture width sequence, and obtains the section texture width value.

[0019] The pixel alignment submodule calls the texture region position index in the cross-sectional texture width value to obtain the texture graphic at the corresponding position in the dynamic image sequence, extracts the start and end coordinates of the texture boundary in consecutive frame images at the same position, compares the boundary offset position, calculates the dynamic offset difference, and adjusts all texture region boundaries uniformly according to the offset difference to obtain the texture pixel offset value.

[0020] The inter-frame judgment submodule calculates the magnitude of the shift difference and the number of jumps between consecutive frames based on the lateral shift and the number of jumps in the texture pixel offset values. It then compares whether the magnitude of the shift and the number of jumps are both lower than the offset stability threshold and the jump consistency threshold, establishes a stable region marker and encodes the image type status, and generates an image type locking label.

[0021] As a further aspect of the present invention, the formula for calculating the dynamic offset difference is specifically as follows:

[0022]

[0023] Where, Δd i,t Represents the dynamic offset difference, where μ represents the cross-sectional width weighting coefficient of the i-th texture region. and These represent the start and end coordinates of the texture boundary of the region in frame t, respectively. i,k,t γ represents the difference in ordinate between the k-th adjacent pixels of the i-th texture region in frame t, γ is the variance smoothing constant, η is the temporal dynamic decay factor, Δt is the time interval parameter between adjacent frames, and ν is the gradient normalization coefficient. θ represents the L2 norm of the gradient magnitude of the i-th texture region in frame t, and θ is the numerical stability adjustment constant.

[0024] As a further aspect of the present invention, the region guidance module includes:

[0025] The classification reading submodule obtains the classification result shown by the image type locking label, establishes an image type index label table in the map sheet, extracts the map sheet area coordinate range corresponding to the image type, and obtains the map sheet area classification index value;

[0026] The strip division submodule divides the map sheet into multiple equal-width strip regions along the main axis direction based on the region marked by the map sheet region classification index value, extracts the start and end coordinates of all contour direction line segments within the strip, calculates the adjustment slope value of the contour direction line segments, and aggregates line segments with the same slope value to obtain the cumulative length of the same direction segment;

[0027] The priority generation submodule calls the length data of the strips in the cumulative length of the same-direction segment, compares it with the stable threshold of the continuous segment length, filters the strip regions whose cumulative length exceeds the threshold, sorts the strip numbers that meet the requirements in descending order of cumulative length, and generates a sorting priority sequence.

[0028] As a further aspect of the present invention, the formula for calculating the adjustment slope value of the contour direction line segment is as follows:

[0029]

[0030] Where, k adj The adjustment slope value of the outline direction line segment, y e The y-coordinate represents the endpoint of the line segment. s The ordinate of the starting point of the line segment, x e The x-coordinate represents the endpoint of the line segment. s λ represents the x-coordinate of the starting point of the line segment. c α represents the weighting coefficient that is dynamically adjusted based on the line segment length. d The balance factor representing the squared difference between the coordinates in the principal axis direction and the horizontal axis direction, ∈ d This represents the horizontal coordinate offset correction amount.

[0031] As a further aspect of the present invention, the path filtering module includes:

[0032] The path extraction submodule obtains the strip region number in the sorting priority sequence, reads the corresponding transport path code, extracts the current task queue number and corresponding waiting time of the path, establishes a path number and waiting parameter table, and obtains the path waiting parameter value.

[0033] The status judgment submodule compares the waiting time with the set reference upper limit threshold based on the path number and waiting time data in the path waiting parameter value, filters out path numbers whose waiting time exceeds the reference upper limit, and outputs the remaining path numbers as available delivery items to obtain the available path number value.

[0034] The combined binding submodule calls the available path number value and the corresponding image classification label, binds them according to the number and classification, constructs a binding sequence table, outputs the binding result number group, and generates the delivery combination number.

[0035] As a further aspect of the present invention, the tag binding module includes:

[0036] The trajectory positioning submodule obtains the number path and image classification identifier in the number of the delivery combination, parses the coordinate information of the trajectory line pointed to by the number path, controls the movement axis parameters of the robotic arm to move to the corresponding coordinates, establishes a lookup table between image identifiers and trajectory points, and obtains the positioning trajectory coordinate values.

[0037] The identification writing submodule calls the trajectory points in the positioning trajectory coordinate value, links the label printing to write the corresponding image classification identification content, collects the current position coordinates and orientation angle of the tire, controls the printing to synchronously attach the writing label to the tire surface, and generates the attached image label number value.

[0038] The action recording submodule synchronously obtains the deployment combination number and label affixing time based on the number information in the attached image label number value, generates action code and establishes a matching relationship with the barcode log, records action entries to the log document sequence, and generates an automated tire delivery solution.

[0039] A machine vision-based automated tire delivery method includes the following steps:

[0040] S1: Acquire the tire image, read the set of horizontal grayscale jump points, calculate the jump interval density, extract the dense continuous region as the central texture area, remove non-continuous boundary segments, select symmetrical and stable texture blocks on both sides, combine them to construct the guide direction graphic, and generate a texture guide cross-section diagram.

[0041] S2: Read the texture-guided cross-sectional image, acquire the texture width image at the same position in the dynamic image, align the frame boundary lines, analyze the synchronous change range of the number of jumps and the offset value, extract the image frame group with consistent changes, and generate image type locking labels;

[0042] S3: Read the image type lock tag, divide the main axis direction strip area, extract the contour line segment direction, determine whether the length of continuous segments in the same direction exceeds the set range, sort and number them according to the strip continuity, and generate a sorting priority sequence.

[0043] S4: Read the first two strips of the sorting priority sequence, obtain the queue number and waiting time of the delivery number, determine and remove the number whose waiting time exceeds the limit, bind the remaining number with the image classification information, and generate the delivery combination number;

[0044] S5: Read the delivery combination number, control the robotic arm to move to the numbered path trajectory line, execute label printing and write image identification, complete the attachment and record it to the log, and generate an automated tire delivery solution.

[0045] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0046] In this invention, tire images are acquired through an image acquisition device, tread texture features are extracted, texture distribution length and number of jumps are calculated, and a guide cross-section diagram is generated. By comparing the lateral offset value of the contour and the number of jumps, a stable image label is generated. Based on the classification results, long strip regions are divided, the slope of line segments and the continuity of segment length in the same direction are extracted, and a priority sequence is generated. The path selection takes into account the number of tasks in the queue and the waiting time, filters the path number and binds the classification information to form a path deployment combination number. Through the movement of the robotic arm and the printing of labels, path scheduling and label management are realized, improving the accuracy and automation level of path planning. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 This is a system flowchart of the present invention;

[0049] Figure 2 This is a system block diagram of the present invention;

[0050] Figure 3 This is a flowchart of the method steps of the present invention. Detailed Implementation

[0051] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0052] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0053] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, their intended meanings are consistent. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, their intended meanings are consistent.

[0054] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0055] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0056] Please see Figure 1 and Figure 2 Machine vision-based automated tire delivery systems include:

[0057] The image locking module acquires an image of the tire below the positioning recognition area, reads the continuous pixel set in the lateral tangent direction of the tire tread, calculates the continuous length and density jump number of the texture distribution in the image, extracts the central texture area, identifies the jump boundary and filters out short abrupt line segments, selects the texture block that meets the jump stability requirement in the left and right symmetrical directions as the main direction recognition reference area, and generates a texture guiding cross-sectional image.

[0058] The texture comparison module reads the continuous texture graphics on the cross section of the texture-guided cross section, obtains the texture width map at the same position in the dynamic image sequence and performs pixel boundary alignment operation, compares the horizontal offset value and the number of jumps in the contour region, determines whether the change in offset value and the number of jump repetitions in consecutive frames simultaneously meet the stability condition, and generates an image type locking label.

[0059] The region guidance module reads the classification results shown by the image type locking label, divides the image into long strip regions along the main axis, extracts the slope values ​​of the contour line segments in the strips and calculates the total length of segments with the same slope, determines whether the length of segments in the same direction in the strips exceeds the set continuous range, sorts them according to the continuity of the strips that meet the conditions, and generates a sorting priority sequence.

[0060] The path filtering module reads the top two strip regions in the sorting priority sequence, obtains the number of tasks queuing and the waiting time on the path corresponding to the delivery number, determines whether there are path numbers whose waiting time exceeds the reference limit and removes them, and binds the excluded numbers with image classification information for delivery scheduling to generate delivery combination numbers.

[0061] The label binding module reads the number path and image classification identifier from the delivery combination number, controls the robotic arm to move to the trajectory line corresponding to the specified number path, retrieves the label printing device to write the image identifier and attaches it to the tire surface, and records the action number in the barcode log to generate an automated tire delivery solution.

[0062] The texture-guided cross-sectional image includes the central texture region, texture blocks in the left and right symmetrical directions, and transition boundaries. The image type locking label includes the cross-sectional texture comparison result, texture width map, lateral offset value of the contour region, and number of transitions. The sorting priority sequence includes long strip regions, line segment slope values, total length of segments with the same slope, and continuity of segment length in the same direction. The delivery combination number includes path number, image classification information, number of tasks in the queue, and waiting time. The automated tire conveying solution includes numbered paths, image classification labels, robotic arm trajectory lines, label labels, action numbers, and barcode logs.

[0063] Please see Figure 1 and Figure 2 The image locking module includes:

[0064] The image acquisition submodule acquires an image of the tire located below the positioning and recognition area, reads the gray value sequence of pixels in the horizontal tangent direction, records the gray value continuity length and counts the number of jumps to obtain the horizontal texture continuity length value.

[0065] The image acquisition submodule acquires an image of the tire located below the positioning and recognition area. First, the industrial camera parameters are initialized, setting the resolution to 1024×768 pixels, the frame rate to 60 frames per second, the exposure time to 10 milliseconds, and using a side-lit white LED with a brightness of 400 lumens to ensure clear details of the tire surface texture. During acquisition, the positioning and recognition area uses a structural sensor to report the tire's current position. The control module triggers the camera to capture the image based on this feedback. After acquisition, the image is transmitted to the image processing unit, which reads the grayscale value sequence of pixels in a specified row along the horizontal tangent direction. For example, taking the 718th row (50 pixels from the bottom of the image), the grayscale values ​​of 1024 pixels from left to right in that row are read, with a range from 0 to 255. The sequence is compared point by point; if the current grayscale value is... If the grayscale difference between the previous pixel and the previous pixel does not exceed 8, it is considered a continuation. For example, if the sequence contains [120, 122, 121, 119, 117], and the grayscale difference of 5 consecutive points does not exceed 8, it is recorded as a continuation segment with a continuation length of 5. If the next point changes to 97, the grayscale difference is 20, the jump count is incremented by 1, and the process continues to slide forward. A window of length 10 is used to traverse and count the entire sequence. Whenever a jump point is detected, the jump position is recorded, and a new continuation length count begins after that segment. If consecutive continuation segments are encountered with values ​​of 3, 4, 6, 2, etc., they are recorded as the corresponding continuation lengths. At the same time, the total number of jump points between these segments is recorded as 4. Finally, the distribution of continuation lengths and the total number of jumps in the row of images are obtained. For example, if the average length of a continuation segment is 4.2 pixels, the total number of jump points is 6.

[0066] The texture extraction submodule extracts densely transitioned regions based on the horizontal texture continuity length value, detects the abrupt change amplitude and span value of each abrupt boundary segment and sets a threshold, deletes short line segments that are both less than the abrupt change amplitude threshold and the span threshold, integrates the remaining boundary segment information, and obtains the stable texture interval position value.

[0067] First, all horizontal pixel rows are traversed, and the concentration of transition points is detected row by row. For each row, a sliding window of 10 pixels is used to calculate the number of transition points within the window. For example, in row 300, there are 3 transition points between pixels 1 and 10, so the transition density is 0.3 / pixel. If the density exceeds a set threshold of 0.2 / pixel, the area is considered a dense transition region. This transition density threshold is derived from extensive experimental statistics. In testing 1000 different tire images, it was found that areas exceeding 0.2 / pixel mostly represent tire tread areas; therefore, this value is taken as a reasonable benchmark. After determining the dense transition region, the abrupt change amplitude and span of each transition boundary segment are detected. For example, in row 45, pixel position 260 has a grayscale value of 130, and pixel position 261 has a grayscale value of 155, so the abrupt change amplitude is 25. A transition boundary segment starts and ends at positions 260 to 264, with a span of 4 pixels. A transition amplitude threshold of 20 and a span threshold of 3 are set to determine if a segment is valid. These two thresholds are derived by comparing typical patterns with worn areas. Wear areas typically have transition amplitudes below 15 and spans below 2, therefore 20 and 3 are set as boundaries. If a boundary segment has a transition amplitude of 18 and a span of 2, it falls below both thresholds and is considered an invalid short segment, which is then deleted. The remaining boundary segments are retained. Valid boundary segments, such as those starting and ending at positions 200 to 208 or 215 to 225, are compared. Adjacent segments are judged; if the distance between two segments does not exceed 2 pixels, they are merged into a complete texture boundary region. The final output is the stable texture interval position value, such as the stable texture interval of 195 to 225 pixels in row 300.

[0068] The guidance generation submodule calls the jump sequence in the left and right symmetrical directions in the stable texture interval position value, filters the symmetrical texture blocks whose jump point spacing change amplitude is less than the jump stability threshold, extracts the horizontal sequence and directional trend value, and generates a texture guidance cross-section map.

[0069] The guided generation submodule performs lateral texture guided extraction based on the stable texture interval position values. First, starting from the center point of each stable texture interval, it extracts a 50-pixel abrupt grayscale sequence to the left and right. For example, if a stable interval starts and ends at pixels 195 to 225, with a center point at 210, then the extracted range is a 160-260 pixel abrupt sequence. From this range, it extracts the abrupt point positions. For instance, abrupt points are found at pixels 163, 170, 176, and 181 on the left, and at pixels 215, 221, 228, and 234 on the right. The spacing between adjacent abrupt points is calculated sequentially. For example, if the spacing between adjacent points on the left is 7, 6, and 5 pixels, and the spacing between adjacent points on the right is 6, 7, and 6 pixels, the difference in spacing between points with the same index on the left and right is compared one by one. The maximum difference is 1 pixel. If the difference is less than the set abrupt... A stability threshold of 2 pixels indicates that the transition sequence has good symmetry. This threshold is set by sampling experiments. In standard tire images, the difference in the left and right transition spacing is usually no more than 2 pixels. Therefore, 2 is set as a reasonable judgment boundary. Symmetrical texture blocks that meet the conditions will be recorded. Then, the overall direction of their transition points will be analyzed to extract the lateral sequence trend. For example, there are 8 transition points from pixels 160 to 234 in this sequence. It is proposed that every two transition points form a line segment. The correspondence between the lateral displacement and the vertical row number is calculated segment by segment, and the upward, downward or parallel trend of the sequence is determined accordingly. Finally, these trends are merged to construct a guide cross-section diagram. The diagram uses the horizontal axis to represent the image width and the vertical axis to represent the trend direction. Each texture block is represented by a set of directional arrows, which intuitively presents the lateral arrangement structure of the tire tread.

[0070] Please see Figure 1 and Figure 2 The texture contrast module includes:

[0071] The section extraction submodule obtains continuous texture graphics on the section in the texture-guided section image, identifies the texture boundary position in the horizontal direction, extracts the width sequence value of the continuous texture region, establishes a corresponding index structure between the section position and the texture width sequence, and obtains the section texture width value.

[0072] The section extraction submodule acquires continuous texture graphics on the cross-section of the texture guide cross-section image. First, it reads the generated guide cross-section image data, traverses each cross-sectional region of the image, and scans each row of pixels from left to right, recording adjacent gray-level abrupt change regions as texture boundary points. The validity of the abrupt change is determined by comparing the gray-level difference between adjacent pixels to see if it exceeds a preset abrupt change threshold. This threshold is set to 12, obtained by statistically analyzing the average gray-level abrupt change distribution in 1000 images and setting its average plus standard deviation range to prevent weak texture boundaries from being mistaken for strong boundaries. After identifying the boundary points, it extracts the continuous texture region between adjacent boundaries and calculates its horizontal length as the texture region width value. If the boundary points are at the 200th and 2nd pixels respectively... If the width is between 25 pixels, then the width of the texture region is 25 pixels. The above extraction process is repeated throughout the entire image height range. One or more texture region width values ​​can be extracted for each row. For example, the width sequence [25, 28, 23] is extracted in row 300, and [24, 29, 22] is extracted in row 301. Then, an index structure is built to pair the pixel number of each row in the image with the corresponding texture width sequence. The row number and its corresponding width sequence are recorded using a key-value pair structure, for example, {300: [25, 28, 23], 301: [24, 29, 22]}. A complete cross-sectional texture width data index is constructed to form a structurally stable horizontal texture width index table, and finally the cross-sectional texture width value is obtained.

[0073] The pixel alignment submodule calls the texture region position index in the cross-sectional texture width value to obtain the texture graphic at the corresponding position in the dynamic image sequence, extracts the start and end coordinates of the texture boundary in consecutive frame images at the same position, compares the boundary offset position, calculates the dynamic offset difference, and adjusts all texture region boundaries uniformly according to the offset difference to obtain the texture pixel offset value.

[0074] The specific formula for calculating the dynamic offset difference is as follows:

[0075]

[0076] Where, Δd i,t Represents the dynamic offset difference, where μ represents the cross-sectional width weighting coefficient of the i-th texture region. and These represent the start and end coordinates of the texture boundary of the region in frame t, respectively. i,k,t γ represents the difference in ordinate between the k-th adjacent pixels of the i-th texture region in frame t, γ is the variance smoothing constant, η is the temporal dynamic decay factor, Δt is the time interval parameter between adjacent frames, and ν is the gradient normalization coefficient. θ represents the L2 norm of the gradient magnitude of the i-th texture region in frame t, and θ is the numerical stability adjustment constant.

[0077] Parameter definition and value source:

[0078] μ: Cross-sectional width weighting coefficient, calculated as the ratio of the standard deviation of texture width to the mean, with a value of 0.8. The calculation method is μ = 0.5 + 0.3·std(w) / mean(w), where w is the texture width monitoring data. The measured standard deviation of texture width std(w) = 0.12, and the mean (w) = 0.4. Substituting these values, we get μ = 0.8. This value fluctuates with the texture width distribution.

[0079] and The boundary detection algorithm of a certain texture region was obtained using image processing software (such as OpenCV). The measured starting coordinates of the texture region in frame t are: Pixels, ending coordinates are Pixel.

[0080] y i,k,t With y i,k,t-1 The difference in the ordinate of adjacent pixels is calculated from the pixel grayscale gradient. The measured difference in the ordinate y of the k=1th adjacent pixel in the tth frame is... i,1,t = 2 pixels, the difference in ordinate y in the (t-1)th frame i,1,t-1 = 1 pixel, the square of the difference is (2-1) 2 =1; the sum of the squares of the differences between other adjacent pixels is 4 (e.g., when k=2, the difference is 3-1=2, and the square is 4).

[0081] γ: Variance smoothing constant, set to 0.3, based on the range of grayscale variance monitoring values ​​in the texture region from 0.1 to 0.5, taking the median value.

[0082] η: Time decay factor, set to 0.1, based on the continuous frame interval Δt = 0.033 seconds (30 frames / second video), combined with the empirical value range of 0.05 to 0.2.

[0083] ν: Gradient normalization coefficient, set to 0.3, based on the gradient magnitude. (Calculated using the Sobel operator), with a reasonable range of normalization coefficients of 0.1 to 0.5.

[0084] The L2 norm of the gradient magnitude is used to calculate the gradient component G of a texture region using the Sobel operator. x =20, G y =15, then

[0085] θ: Numerical stability adjustment constant, set to 1×10 -5 This is to prevent the denominator from being too small.

[0086] Formula calculation derivation process:

[0087] Molecular calculations:

[0088] Cross-section width item:

[0089] Dynamic deformation term:

[0090] Total numerator: 40 + 16.491 = 56.491;

[0091] Denominator calculation:

[0092] Gradient normalization term:

[0093] Final result:

[0094]

[0095] Result Analysis:

[0096] The result Δd i,t =7.532 represents the dynamic offset difference of the i-th texture region in frame t, and the value is positively correlated with the texture boundary offset. By comparing the offset differences between adjacent frames, the offset of the boundaries of all texture regions is adjusted to obtain the finally uniformly corrected texture pixel offset value.

[0097] The inter-frame judgment submodule calculates the magnitude of the shift difference and the number of jumps between consecutive frames based on the horizontal offset and the number of jumps in the texture pixel offset values. It compares whether the magnitude of the shift and the number of jumps are both lower than the offset stability threshold and the jump consistency threshold, establishes stable region markers and encodes the image type status, and generates an image type locking label.

[0098] The inter-frame judgment submodule reads the texture pixel offset records at the same position in all consecutive frames based on the lateral offset and jump count in the texture pixel offset values ​​of consecutive frames. For each pair of adjacent frames, it calculates the offset value change amplitude ΔP to determine whether there is a sudden change in the lateral position of the texture boundary between the two frames. For example, if the offset value from frame 1 to frame 2 is 2 pixels and from frame 2 to frame 3 is 3 pixels, then ΔP is 1 pixel. It counts the number of times ΔP exceeds a set stability threshold in the entire sequence. The stability threshold is set to 4 pixels. This value comes from tire running video sampling experiments. If the offset exceeds 4 pixels, it often corresponds to a dynamically unstable image frame. Therefore, it uses this as a boundary to count the number of segments with ΔP less than or equal to 4. If the proportion exceeds 80% of the total number of frames, the texture region offset is judged as a stable region. At the same time, it counts the number of jumps in the same row and the same region in each frame. The number of jumps refers to the gray-level change points in the image row. The number of jumps is compared with the number of adjacent frames for each frame, and the number of jump repetitions is recorded. That is, whether the same jump point appears repeatedly in consecutive frames. If the number of jump points in the same texture area in consecutive frames is higher than the jump consistency threshold, that is, the consistency jump ratio is greater than 80%, this threshold is an empirical statistical value, which is set by comparing the repetition of jump points in the same pattern area in more than 300 frames of images. Then it is considered to be jump consistent. Finally, the texture area that satisfies ΔP≤4 pixels and jump consistency ≥80% is marked as a stable area and a status code is marked for the area. For example, stable state is 1 and unstable state is 0. Combined with the frame number and region number, an image state lock label is formed. The sample structure is as follows: {(frame, line, region): state}. For example, {(12, 300, 1): 1} means that the first region of the 300th line of the 12th frame is locked as a stable state.

[0099] Please see Figure 1 and Figure 2 The regional boot module includes:

[0100] The classification reading submodule obtains the classification results shown by the image type locking label, establishes an image type index label table in the map sheet, extracts the map sheet area coordinate range corresponding to the image type, and obtains the map sheet area classification index value;

[0101] First, the image type locking tags output by the inter-frame judgment submodule are retrieved from the image processing system. The tag data structure is a key-value pair of frame number, region coordinates, and corresponding status code, for example, {(frame, line, region): type}. After reading the tags, the system initializes the current map area and defines an image index mapping table in the two-dimensional coordinate space of the map area. Each tag type is mapped to an independent index number. For example, status code 1 for stable image regions is mapped to index category A, and status code 0 for unstable regions is mapped to index category B. Based on the start and end coordinates of the region and the frame number in each tag, the corresponding coordinate range is found in the map area. For example, if the start and end of region 2 in line 350 is x = 200 to x = 230, and y = 350, then this region is defined as a rectangle in the map area. The coordinate blocks (200, 350)-(230, 350) are assigned to index category A. The processing iterates through all the frame number and region number combinations in the locking tags, reading the start and end coordinates each time and filling them into the index tag table to form index region block information. Then, coordinate aggregation is performed on each image type to extract all region coordinate ranges corresponding to each image type, forming a set of classified regions within the map area. For example, 10 different region blocks are extracted from the stable region. The horizontal and vertical coordinates of each block are recorded as a set of data. Finally, the index table structure is established as {A: [(x1, y1, x2, y2), ...], B: [...]}, which is used to identify the corresponding position coordinates of all classified regions in the map, thereby obtaining the map area classification index value.

[0102] The strip division submodule divides the map sheet into multiple equal-width strip regions along the main axis based on the region marked by the map sheet area classification index value. It extracts the start and end coordinates of all contour direction line segments within the strip and calculates the adjustment slope value of the contour direction line segments. Line segments with the same slope value are aggregated to obtain the cumulative length of the same direction segment.

[0103] The formula for calculating the adjustment slope value of the contour direction line segment is as follows:

[0104]

[0105] Where, k adj The adjustment slope value of the outline direction line segment, y e The y-coordinate represents the endpoint of the line segment. s The ordinate of the starting point of the line segment, x e The x-coordinate represents the endpoint of the line segment. s λ represents the x-coordinate of the starting point of the line segment. c α represents the weighting coefficient that is dynamically adjusted based on the line segment length. d The balance factor representing the squared difference between the coordinates in the principal axis direction and the horizontal axis direction, ∈ d This represents the horizontal coordinate offset correction amount;

[0106] Parameter assignment and data source

[0107] Coordinate values: Coordinates of the starting point of the line segment (x) s ,y s ) and endpoint coordinates (x e ,y e It can be directly extracted using surveying equipment or map data acquisition systems. For example, if the starting point of a line segment is (2.5, 3.8) and the ending point is (5.2, 7.1), then x... s =2.5,y s =3.8,x e =5.2,y e =7.1.

[0108] Weighting coefficient λ c Based on the length of the line segment Dynamic adjustment. When the line segment length L = 5.0, λ c =0.6; when L=10.0, λ c =0.3. This coefficient is set based on the distribution pattern of line segment lengths in the statistical map. The longer the line segment, the smaller its weight, which is used to suppress the excessive influence of long line segments on the slope calculation.

[0109] Balance factor α d Calculated based on the squared mean of the coordinate differences along the principal axis (e.g., the horizontal axis). When the mean difference in coordinates along the principal axis of the map sheet is 2.0, α... d =2.0, used to balance the differences in dimensions between the horizontal and vertical axes.

[0110] Correction amount ∈ d The value is fixed at 0.001, defined by engineering specifications to avoid a denominator of zero and to prevent significant interference with the calculation results.

[0111] Calculation process for the example;

[0112] Taking the line segment with starting point (1.0, 2.0) and ending point (4.0, 6.0) as an example:

[0113] Step 1: Extract coordinate values;

[0114] x s =1.0,y s =2.0,x e =4.0,y e =6.0;

[0115] Step 2: Calculate the length L of the line segment;

[0116]

[0117] Step 3: Set λ c ;

[0118] Given L = 5.0, we can find λ from the table. c =0.6;

[0119] Step 4: Set α d ;

[0120] The mean difference in principal axis coordinates is 3.0, α d =2.0;

[0121] Step 5: Substitute into the formula to calculate k adj ;

[0122]

[0123]

[0124] Numerical Result Analysis:

[0125] Calculate k adj ≈1.996 represents the slope value after line segment adjustment. This value integrates the line segment length weight and coordinate difference balance factor, compared to the traditional slope k = (y e -y s ) / (x e -x s The adjusted slope (4.0 / 3.0 ≈ 1.333) emphasizes the directional characteristics of line segments and reduces noise interference from short line segments. In subsequent aggregation processing, the cumulative length of line segments with the same slope is determined by statistically analyzing all k... adj Achieve this by summing the lengths of approximately equal line segments.

[0126] The priority generation submodule calls the length data of the strips in the cumulative length of the same-direction segment, compares it with the stable threshold of the continuous segment length, filters the strip regions whose cumulative length exceeds the threshold, sorts the strip numbers that meet the requirements in descending order of cumulative length, and generates a sorting priority sequence.

[0127] The priority generation submodule calls the strip length data from the cumulative length of the same-direction segments. First, it iterates through all strip numbers and their corresponding cumulative length values ​​for each slope group in the record table. From each strip, it selects the slope group with the largest cumulative length as the representative length value. This value is then compared with a set stable threshold for continuous segment length. This stable threshold is empirically set to 40 pixels based on the average length of the main texture segment in the tire image. That is, if the cumulative length value of any same-direction slope group in a strip exceeds 40 pixels, the strip is considered to have priority. For all strips that meet the condition, their numbers and corresponding values ​​are extracted. The cumulative lengths are used to form a set to be sorted. For example, if the cumulative value of strip S3 is 55 pixels, S7 is 63 pixels, and S11 is 44 pixels, then these strips are added to the list of valid strips. Then, the strip set is sorted in descending order by cumulative length, with the first being S7, the second being S3, and the third being S11. The sorting structure is recorded in list form [(S7, 63), (S3, 55), (S11, 44)]. The strip numbers are extracted in sequence to form the priority sequence S7→S3→S11. Finally, the strip priority number structure sorted by length is output, generating the corresponding sorting priority sequence.

[0128] Please see Figure 1 and Figure 2 The path filtering module includes:

[0129] The path extraction submodule obtains the strip area number in the sorting priority sequence, reads the corresponding conveying path code, extracts the current task queue number and corresponding waiting time for the path, establishes a path number and waiting parameter table, and obtains the path waiting parameter value.

[0130] First, the system receives a sequence of stripe numbers arranged in descending order of cumulative length, output by the priority generation submodule. For example, the priority sequence is [S5, S3, S7]. The system reads the stripe numbers one by one from this sequence and determines the corresponding transport path code for each stripe according to the preset stripe-path mapping table. For example, stripe S5 corresponds to path number P12, stripe S3 corresponds to P08, and stripe S7 corresponds to P15. Then, it enters the transport management unit and reads two key parameters for each path: the current task queue number and the waiting time. The queue number refers to the number of sorting units waiting to be executed on this path, in units of... The number of tasks and the waiting time refer to the average time from when a task is added to the queue to when it has been waiting, in seconds. For example, path P12 currently has 6 tasks in the queue with an average waiting time of 42 seconds, path P08 has 3 tasks with a waiting time of 20 seconds, and path P15 has 5 tasks with a waiting time of 37 seconds. The path number and the corresponding two parameters are used to form a key-value pair record table structure of {path number: (queue number, waiting time)}, forming a preliminary path waiting parameter structure. The reading and binding are performed sequentially for all stripes corresponding to the paths, and the complete path number and waiting parameter table are generated. Finally, the path waiting parameter values ​​are obtained.

[0131] The status judgment submodule compares the waiting time with the set reference upper limit threshold based on the path number and waiting time data in the path waiting parameter value, filters out path numbers whose waiting time exceeds the reference upper limit, and outputs the remaining path numbers as available delivery items to obtain the available path number value.

[0132] The status judgment submodule reads the path number and waiting time data from the path waiting parameter value in the parameter table sequentially, and judges whether the waiting time of each path exceeds the set reference upper limit threshold. This upper limit threshold is calculated from the device operation monitoring data and is set to 40 seconds. This value is based on the upper boundary of the average waiting time of 80% of tasks in the daily operation sample. That is, if the waiting time of a path exceeds 40 seconds, the current path is considered unusable. The waiting time in the data of each path is compared with this threshold. If the waiting time is greater than 40 seconds, the path number is filtered out. For example, the waiting time of path P12 is 42 seconds, which exceeds the threshold and is rejected. Paths P08 and P15 are 20 seconds and 37 seconds respectively, which are less than or equal to the threshold, so their path numbers are retained. The path numbers that meet the conditions are written into the output queue in sequence to form a set of path numbers that can be used for the current frame image. Finally, this set is output as the available path number values. For example, in this example, the output is [P08, P15].

[0133] The combined binding submodule calls the available path number value and the corresponding image classification label, binds the number and category, constructs a binding sequence list and outputs the binding result number group, and generates the delivery combination number;

[0134] First, the system reads the set of available path numbers output by the previous module, such as [P08, P15]. Then, it calls the image classification label structure generated in the classification reading submodule, mapping the labels to categories A, B, and C. In the current image frame, P08 is bound to category A, and P15 is bound to category C. The system iterates and binds the paths using the path number as the primary key. For each available path number, the system searches for the corresponding image region's classification identifier in the classification label index table. Once a one-to-one correspondence is found, the path number and classification type are merged to form a binding sequence element. For example, path P08 is bound to A, path P15 to B, and path P15 to C. 5. Bind C to form an initial binding structure of [(P08, A), (P15, C)]. After recording each pair of binding elements, generate number groups in sequence. The system forms a unique delivery combination code based on the image frame number, path number, and category tag number. The structure can be {frame_id: number, path_id: path number, class_tag: category tag}. Finally, output the set of numbers of all binding combinations as the result. For example, if the current frame combination number is {001, P08, A} and {001, P15, C}, output this number group as the final generated delivery combination number.

[0135] Please see Figure 1 and Figure 2 The tag binding module includes:

[0136] The trajectory positioning submodule obtains the number path and image classification identifier in the delivery combination number, parses the trajectory line coordinate information pointed to by the number path, controls the movement axis parameters of the robotic arm to move to the corresponding coordinates, establishes a lookup table between image identifiers and trajectory points, and obtains the positioning trajectory coordinate values.

[0137] First, the path number and image classification identifier are extracted from the deployment combination number data. For example, in combination number {001, P08, A}, path number P08 and classification identifier A are extracted. Then, the system searches for the corresponding transport trajectory line data in the trajectory path database using path number P08. The trajectory line is represented by a series of continuous coordinate points. For example, the trajectory line corresponding to path P08 contains 5 control points, namely (100, 50), (120, 60), (140, 65), (160, 68), and (180, 70). The system obtains the motion target points of the robotic arm according to the control point sequence, and sets the target values ​​of the servo control axes in the X and Y directions as the current target coordinates. For example, the first segment of the trajectory target... Given (100, 50), the system sends the servo position code value corresponding to 100 pixels to the X-axis controller and the position code value corresponding to 50 pixels to the Y-axis controller, controlling the robotic arm to move forward point by point until it reaches the final trajectory point. During the movement of the robotic arm, the system continuously analyzes the coordinates of each point in the trajectory line and performs path easing control to ensure that the robotic arm stops stably at the target delivery point. Then, a mapping relationship is established between the image classification identifier A and the trajectory endpoint coordinate point (180, 70), forming a comparison record item between the image identifier and the trajectory point, for example, recorded as {A: (180, 70)}. The same operation is performed on multiple image classification identifiers, and finally a comparison table of all classified images and trajectory points is obtained, outputting the positioning trajectory coordinate values.

[0138] The label writing submodule calls the trajectory points in the positioning trajectory coordinate value, links the label printing to write the corresponding image classification label content, collects the current position coordinates and orientation angle of the tire, controls the printing to synchronously attach the label to the tire surface, and generates the attached image label number value.

[0139] The label writing submodule calls the trajectory points in the positioning trajectory coordinate values, reads the corresponding category label and coordinate values ​​of each trajectory point one by one, starts the linkage printing control unit, inputs the image category label data into the label printing controller, and formats and encodes the label content, for example, the category label A is converted into the barcode string "TY-A-001". Then the robotic arm grabs the label and prepares for the affixing action. The control module reads the current position coordinates and orientation angle information of the tire at the trajectory endpoint (e.g., 180, 70). This information is obtained through the bottom vision system to identify the tire's positioning center and rotation angle on the track. For example, if the current tire center is (180, 70) and the angle is 20°, the system will control the robotic arm end effector. The angle is reversed to ensure the label's attachment direction aligns with the tire's orientation. The control module then slowly presses the label onto the tire surface from its current position along a specific path. Once the label is successfully attached, the control module binds the printed information number, recording the printed content as "TY-A-001", the printing device number as PR-03, and the label attachment position as (180, 70). A unique image label number is assigned, such as TL-A-20250509-01, with the structure recorded as {tag_id: TL-A-20250509-01, path: P08, class: A, pos: (180, 70)}. Finally, the attached image label number value is generated.

[0140] The action recording submodule synchronously obtains the deployment combination number and label affixing time based on the number information in the attached image label number value, generates action code and establishes a matching relationship with the barcode log, records action items to the log document sequence, and generates an automated tire delivery solution.

[0141] The action recording submodule, based on the number information in the attached image label number value, synchronously obtains the corresponding deployment combination number and label attachment time. First, it retrieves the generated image label number TL-A-20250509-01, finding its corresponding path number P08 and category identifier A. Simultaneously, it consults the deployment combination number record table to confirm that the combination structure corresponding to this number is {001, P08, A}. The system obtains the current operation time through the operation timestamp recording module, for example, a time record of 11:43:27 on May 9, 2025. Then, it generates an action code by concatenating the current label number, path number, deployment combination number, and timestamp to form a unique action code, such as ACD-P08-A-20250509-114327, and records this action. The operation type is "label attachment", the target device is "robotic arm device MB-2", the above action code is bound to the current barcode content "TY-A-001" and recorded in the barcode log sequence. The log structure is {timestamp, action_code, tag_content, device_id, label_position}. For example, the record entry is {2025-05-09 11:43:27, ACD-P08-A-20250509-114327, TY-A-001, MB-2, (180, 70)}. After the entry is written, the log is synchronously uploaded to the data center for aggregation and storage. Combined with the trajectory information and action code sequence, the final tire automated delivery solution is generated.

[0142] Please see Figure 3 A machine vision-based automated tire delivery method includes the following steps:

[0143] S1: Acquire the tire image, read the set of horizontal grayscale jump points, calculate the jump interval density, extract the dense continuous region as the central texture area, remove non-continuous boundary segments, select symmetrical and stable texture blocks on both sides, combine them to construct the guide direction graphic, and generate a texture guide cross-section diagram.

[0144] S2: Read the texture-guided cross-sectional image, acquire the texture width image at the same position in the dynamic image, align the frame boundary line, analyze the synchronous change range of the number of jumps and the offset value, extract the image frame group with consistent changes, and generate image type locking labels;

[0145] S3: Read the image type lock label, divide the main axis direction strip area, extract the contour line segment direction, determine whether the length of continuous segments in the same direction exceeds the set range, sort and number according to the strip continuity, and generate a sorting priority sequence.

[0146] S4: Read the first two strips of the sorting priority sequence, obtain the queue number and waiting time of the delivery number, determine and remove the number whose waiting time exceeds the limit, bind the remaining number with the image classification information, and generate the delivery combination number;

[0147] S5: Read the delivery combination number, control the robotic arm to move to the numbered path trajectory line, execute label printing and write image identification, complete the attachment and record it to the log, and generate an automated tire delivery solution.

[0148] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A machine vision-based automated tire conveying system, characterized in that: The system includes: The image locking module acquires the tire image, reads the pixel set in the lateral tangent direction of the tire tread, calculates the continuous length of the texture distribution and the number of density jumps, extracts the central texture region, identifies the jump boundaries, filters out short abrupt line segments, and generates a texture-guided cross-sectional image. The texture comparison module reads the continuous texture of the texture-guided cross-section map, obtains the texture width map in the dynamic image sequence, aligns the pixel boundaries, compares the contour lateral offset value with the number of jumps, and generates an image type locking label; The region guidance module locks the label according to the image type, divides the strip region along the main axis in the image, extracts the slope of the contour direction line segment, calculates the total length of the segment with the same slope, determines whether the length of the continuous segment exceeds the limit, and sorts and generates a sorting priority sequence. The path filtering module reads the first two bands of the sorting priority sequence, obtains the number of paths in queue and the waiting time, removes paths with excessive waiting time, and generates a delivery combination number. The label binding module controls the robotic arm to move to the path line according to the delivery combination number, prints image labels, attaches them to the tires, records the numbers to the barcode log, and generates an automated tire delivery solution. The texture-guided cross-sectional image includes a central texture region, texture blocks in left and right symmetrical directions, and transition boundaries. The image type locking label includes cross-sectional texture comparison results, texture width image, lateral offset value of contour region, and number of transitions. The sorting priority sequence includes long strip regions, line segment slope values, total length of segments with the same slope, and continuity of segment length in the same direction. The delivery combination number includes path number, image classification information, number of tasks in queue, and waiting time. The automated tire conveying scheme includes numbered paths, image classification identifiers, robotic arm trajectory lines, label identifiers, action numbers, and barcode logs. The image locking module includes: The image acquisition submodule acquires an image of the tire located below the positioning and recognition area, reads the gray value sequence of pixels in the horizontal tangent direction, records the gray value continuity length and counts the number of jumps to obtain the horizontal texture continuity length value. The texture extraction submodule extracts the densely abrupt regions based on the horizontal texture continuity length value, detects the abrupt change amplitude value and span value of each abrupt boundary segment and sets a threshold, deletes short line segments that are both less than the abrupt change amplitude threshold and the span threshold, integrates the remaining boundary segment information, and obtains the stable texture interval position value. The guidance generation submodule calls the jump sequence in the left and right symmetrical directions in the stable texture interval position value, filters out symmetrical texture blocks whose jump point spacing change amplitude is less than the jump stability threshold, extracts the horizontal sequence and directional trend value, and generates a texture guidance cross-section map. The texture contrast module includes: The section extraction submodule obtains the continuous texture pattern on the section in the texture-guided section image, identifies the texture boundary position in the horizontal direction, extracts the width sequence value of the continuous texture region, establishes a corresponding index structure between the section position and the texture width sequence, and obtains the section texture width value. The pixel alignment submodule calls the texture region position index in the cross-sectional texture width value to obtain the texture graphic at the corresponding position in the dynamic image sequence, extracts the start and end coordinates of the texture boundary in consecutive frame images at the same position, compares the boundary offset position, calculates the dynamic offset difference, and adjusts all texture region boundaries uniformly according to the offset difference to obtain the texture pixel offset value. The inter-frame judgment submodule calculates the magnitude of the shift difference and the number of jumps between consecutive frames based on the horizontal shift and the number of jumps in the texture pixel offset values. It compares whether the magnitude of the shift and the number of jumps are both lower than the offset stability threshold and the jump consistency threshold, establishes a stable region marker and encodes the image type status, and generates an image type locking label. The specific formula for calculating the dynamic offset difference is as follows: ； in, Represents the dynamic offset difference. Representing the The cross-sectional width weighting coefficient of each texture region and These represent the regions in the [number]th [year]. The start and end coordinates of the frame's texture boundary. Representing the The texture region in the first The first frame The difference in the ordinate of adjacent pixels. Let Variance be the smoothing constant. The time-dynamic decay factor, This is the time interval parameter between adjacent frames. These are the gradient normalization coefficients. Representing the The texture region in the first The L2 norm of the gradient magnitude of the frame. This is the numerical stability adjustment constant; The region guidance module includes: The classification reading submodule obtains the classification result shown by the image type locking label, establishes an image type index label table in the map sheet, extracts the map sheet area coordinate range corresponding to the image type, and obtains the map sheet area classification index value; The strip division submodule divides the map sheet into multiple equal-width strip regions along the main axis direction based on the region marked by the map sheet region classification index value, extracts the start and end coordinates of all contour direction line segments within the strip, calculates the adjustment slope value of the contour direction line segments, and aggregates line segments with the same slope value to obtain the cumulative length of the same direction segment; The priority generation submodule calls the length data of the strips in the cumulative length of the same-direction segment, compares it with the stable threshold of the continuous segment length, filters the strip regions whose cumulative length exceeds the threshold, sorts the strip numbers that meet the requirements in descending order of cumulative length, and generates a sorting priority sequence. The formula for calculating the adjustment slope value of the contour direction line segment is as follows: ； in, The slope value represents the adjustment slope of the outline line segment. The ordinate of the endpoint of the line segment. The ordinate represents the starting point of the line segment. The x-coordinate of the endpoint of the line segment. The x-coordinate represents the starting point of the line segment. This represents a weighting coefficient that is dynamically adjusted based on the line segment length. The balance factor representing the squared difference between the coordinates in the principal axis direction and the horizontal axis direction. This represents the horizontal coordinate offset correction amount.

2. The automated tire conveying system based on machine vision according to claim 1, characterized in that, The path filtering module includes: The path extraction submodule obtains the strip region number in the sorting priority sequence, reads the corresponding transport path code, extracts the current task queue number and corresponding waiting time of the path, establishes a path number and waiting parameter table, and obtains the path waiting parameter value. The status judgment submodule compares the waiting time with the set reference upper limit threshold based on the path number and waiting time data in the path waiting parameter value, filters out path numbers whose waiting time exceeds the reference upper limit, and outputs the remaining path numbers as available delivery items to obtain the available path number value. The combined binding submodule calls the available path number value and the corresponding image classification label, binds them according to the number and classification, constructs a binding sequence table, outputs the binding result number group, and generates the delivery combination number.

3. The automated tire conveying system based on machine vision according to claim 1, characterized in that, The tag binding module includes: The trajectory positioning submodule obtains the number path and image classification identifier in the number of the delivery combination, parses the coordinate information of the trajectory line pointed to by the number path, controls the movement axis parameters of the robotic arm to move to the corresponding coordinates, establishes a lookup table between image identifiers and trajectory points, and obtains the positioning trajectory coordinate values. The identification writing submodule calls the trajectory points in the positioning trajectory coordinate value, links the label printing to write the corresponding image classification identification content, collects the current position coordinates and orientation angle of the tire, controls the printing to synchronously attach the writing label to the tire surface, and generates the attached image label number value. The action recording submodule synchronously obtains the deployment combination number and label affixing time based on the number information in the attached image label number value, generates action code and establishes a matching relationship with the barcode log, records action entries to the log document sequence, and generates an automated tire delivery solution.

4. A machine vision-based automated tire conveying method, characterized in that, The method is used to implement the automated tire delivery system based on machine vision as described in any one of claims 1-3, and includes the following steps: S1: Acquire the tire image, read the set of horizontal grayscale jump points, calculate the jump interval density, extract the dense continuous region as the central texture area, remove non-continuous boundary segments, select symmetrical and stable texture blocks on both sides, combine them to construct the guide direction graphic, and generate a texture guide cross-section diagram. S2: Read the texture-guided cross-sectional image, acquire the texture width image at the same position in the dynamic image, align the frame boundary lines, analyze the synchronous change range of the number of jumps and the offset value, extract the image frame group with consistent changes, and generate image type locking labels; S3: Read the image type lock tag, divide the main axis direction strip area, extract the contour line segment direction, determine whether the length of continuous segments in the same direction exceeds the set range, sort and number them according to the strip continuity, and generate a sorting priority sequence. S4: Read the first two strips of the sorting priority sequence, obtain the queue number and waiting time of the delivery number, determine and remove the number whose waiting time exceeds the limit, bind the remaining number with the image classification information, and generate the delivery combination number; S5: Read the delivery combination number, control the robotic arm to move to the numbered path trajectory line, execute label printing and write image identification, complete the attachment and record it to the log, and generate an automated tire delivery solution.

Images