A visual texture navigation AGV running track real-time monitoring method

Abstract

The application provides a visual texture navigation AGV running track real-time monitoring method, comprising the following steps: obtaining real-time ground image frames and current driving mileage from an image acquisition module and a position solution module on an AGV chassis, combining pre-stored historical track coordinates to obtain initial positioning data; evaluating the similarity pattern repetition degree between adjacent image frames according to the initial positioning data and the height of the preset threshold, marking the positions higher than the threshold as texture confusion high-risk sections to determine potential misjudgment positions; for the candidate anchor points, combining the driving mileage cumulative error to dynamically adjust the filtering threshold value, retaining the anchor points with the threshold value lower than the average length of the texture period to obtain the optimized anchor points; from the optimized anchor points, selecting the texture section with the highest similarity of the texture features of each anchor point and the gray distribution of the current image frame as the matching reference, evaluating the size of the coordinate correction deviation and the allowed range, and determining the corrected track coordinates.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a method for real-time monitoring of the running trajectory of a visual texture navigation AGV. Background Technology

[0002] In modern industry and logistics, navigation technology for Automated Guided Vehicles (AGVs) is a crucial pillar for achieving intelligent transportation and efficient operations. The accurate monitoring of their operating trajectories directly impacts production efficiency and safety. Especially in complex factory or warehouse environments, ensuring AGVs can stably travel along predetermined paths, avoiding deviations or collisions, has become a critical aspect of technological development. Research in this area not only affects the reliability of equipment operation but also plays a decisive role in the synergy of the entire automation system. However, current trajectory monitoring methods often struggle to balance the contradiction between local recognition and global positioning. Many solutions rely too heavily on fixed recognition rules or preset reference points when processing ground features, ignoring the dynamic changes in environmental characteristics. This approach often fails to accurately distinguish between current and historical positions when dealing with complex or highly repetitive ground patterns, leading to a gradual accumulation of positioning errors and affecting the long-term operational stability of the AGV. Focusing on specific technical challenges, the period length of the ground texture becomes a core factor affecting positioning reliability. The texture period length refers to the interval between repeated occurrences of ground patterns, directly determining the distribution density and uniqueness of positioning reference points that the AGV can acquire during operation. If the cycle is short and the pattern repeats frequently, while providing more correction opportunities over short distances, it can easily cause the system to confuse the current position with similar areas in the distance. Conversely, if the cycle is long, although the pattern's recognizability may improve, correction opportunities decrease, increasing the risk of accumulated deviations. This contradiction makes it difficult for the system to maintain stable trajectory monitoring on long-distance straight paths. For example, in a large warehouse, AGVs need to transport goods along a straight aisle hundreds of meters long, with a regular grid-like texture on the ground as a navigation reference. Due to the short texture cycle, similar patterns appear every few meters. The AGV's vision system, while moving at high speed, may mistake its current position for a similar point tens of meters away, leading to incorrect trajectory judgments, deviations from the planned route, and even the risk of collisions with other equipment. Therefore, accurately identifying recognizable reference areas with varying ground texture cycle lengths, avoiding misjudgments caused by similar patterns, and ensuring trajectory stability over long distances has become a critical problem that urgently needs to be solved. Summary of the Invention

[0003] This invention provides a method for real-time monitoring of the running trajectory of a visual texture-guided AGV, mainly including:

[0004] The image acquisition module and position calculation module on the AGV chassis acquire real-time ground image frames and current driving distance, and combine them with pre-stored historical trajectory coordinates to obtain initial positioning data;

[0005] Based on the initial positioning data, the similarity of patterns between adjacent image frames is evaluated and compared with a preset threshold. Locations that exceed the threshold are marked as high-risk areas for texture confusion, thus identifying potential misjudgment locations.

[0006] Based on high-risk areas of texture confusion and potential misjudgment locations, real-time image frames and historical coordinates within the corresponding range are obtained, local texture periodic deviations are analyzed, local gray-level gradients are extracted and their uniqueness in the global path is analyzed, and candidate anchor points are obtained.

[0007] For candidate anchor points, the screening threshold is dynamically adjusted in combination with the cumulative error of driving mileage. Anchor points with threshold values ​​lower than the average length of texture period are retained to obtain the preferred anchor points.

[0008] From the preferred anchor points, the texture segment with the highest similarity between the texture features of each anchor point and the grayscale distribution of the current image frame is selected as the matching benchmark. The magnitude of the deviation and allowable range after coordinate correction is evaluated, and the corrected trajectory coordinates are determined.

[0009] Obtain the offset of the corresponding position of the corrected trajectory coordinate record and the real-time image frame, evaluate the offset difference between the previous and next frames and the magnitude of the preset change threshold, and trigger backtracking drift suppression based on historical stable anchor points for offsets exceeding the threshold, that is, backtrack to the previous stable anchor point position and re-perform texture matching localization to obtain the suppressed stable trajectory.

[0010] The historical coordinate database is updated based on the stable trajectory coordinates. The updated historical coordinate database is then sent back to the image acquisition module to evaluate the improvement in overall positioning reliability and obtain the final visual texture navigation AGV trajectory monitoring results.

[0011] Furthermore, the process of acquiring real-time ground image frames and current mileage from the image acquisition module and position calculation module on the AGV chassis, and fusing them with pre-stored historical trajectory coordinates to obtain initial positioning data, includes:

[0012] The system acquires real-time ground image frames from the image acquisition module, obtains the current driving mileage from the location calculation module, matches the real-time ground image frames with the pre-stored historical trajectory coordinates, extracts the texture grayscale distribution features and the period length of the texture pattern from the real-time ground image frames, and generates initial positioning data containing the texture grayscale distribution features and the period length.

[0013] Furthermore, the step of evaluating the repetition degree of similar patterns between adjacent image frames based on the initial positioning data and comparing it with a preset threshold, marking locations exceeding the threshold as high-risk areas for texture confusion, and determining potential misjudgment locations includes:

[0014] A sliding detection window is set up, and grayscale distribution feature vectors of adjacent image frames within the sliding detection window are extracted. The inter-frame correlation coefficient of the grayscale distribution feature vectors is calculated, and the ratio of the number of image frames exceeding a preset similarity threshold to the total number of frames is used as the similarity pattern repetition. The similarity pattern repetition is compared with the preset repetition threshold, and the path intervals exceeding the preset repetition threshold are marked as high-risk areas for texture confusion. The acquisition positions within the high-risk areas for texture confusion are determined as potential misjudgment positions.

[0015] Furthermore, based on high-risk areas of texture confusion and potential misjudgment locations, the process of obtaining real-time image frames and historical coordinates within the corresponding range, analyzing the periodic deviation of local textures, extracting local grayscale gradients and analyzing their uniqueness in the global path to obtain candidate anchor points includes:

[0016] Based on the coordinate range of the high-risk texture confusion area, a real-time image frame sequence is acquired from the image acquisition module. Historical coordinates and texture features of the corresponding range are extracted from the historical coordinate library. The interval distance between adjacent texture patterns is statistically analyzed. The difference between the interval distance and the standard texture period length is calculated to obtain the periodic deviation ratio. Image frame positions exceeding a preset deviation threshold are selected. For the selected positions, the gray-level difference sequence of adjacent pixels is extracted as a local gray-level gradient feature vector. The vector distance between the local gray-level gradient feature vector and the historical texture features is calculated. Positions with a minimum vector distance greater than a preset distinction threshold are determined to be unique positions and marked as candidate anchor points. A set of candidate anchor points is generated by traversing all positions.

[0017] Furthermore, for candidate anchor points, the selection threshold is dynamically adjusted based on the accumulated mileage error, and anchor points with threshold values ​​lower than the average length of the texture period are retained to obtain preferred anchor points, including:

[0018] For the candidate anchor points, the corresponding mileage value is obtained, the cumulative mileage error value is calculated as the screening threshold, the average length of the texture period is obtained from the historical coordinate library, the screening threshold is compared with the average length of the texture period, and candidate anchor points with a length lower than the average length of the texture period are retained to generate a preferred anchor point set.

[0019] Furthermore, the step of selecting the texture segment with the highest similarity between the texture features of each anchor point and the grayscale distribution of the current image frame from the preferred anchor points as the matching benchmark, evaluating the magnitude of the deviation after coordinate correction and the allowable range, and determining the corrected trajectory coordinates includes:

[0020] For the preferred anchor points, the correlation coefficient between the texture feature vector of each anchor point and the grayscale distribution feature vector of the current image frame is calculated. The texture segment corresponding to the anchor point with the highest correlation coefficient is selected as the matching benchmark. The known coordinates of the matching benchmark are obtained. The pixel distance of the matching benchmark texture segment is calculated and converted into a physical distance offset. The offset is superimposed on the known coordinates to generate corrected coordinates. The deviation between the corrected coordinates and the current mileage coordinates is compared with the allowable deviation range to determine the corrected trajectory coordinates.

[0021] Furthermore, the process of obtaining the offset of the corresponding position between the corrected trajectory coordinate record and the real-time image frame, evaluating the offset difference between consecutive frames and the magnitude of a preset change threshold, and triggering backtracking drift suppression based on historical stable anchor points for offsets exceeding the threshold, i.e., reverting to the previous stable anchor point position and re-performing texture matching localization to obtain a suppressed stable trajectory, includes:

[0022] Extract the current position offset corresponding to the corrected trajectory coordinates, calculate the difference between the position offset and the previous image frame position offset as the inter-frame offset difference value, compare the inter-frame offset difference value with a preset change threshold, and mark anchor points whose offset difference value is less than or equal to the preset change threshold as stable anchor points; for offsets exceeding the preset change threshold, select the nearest stable anchor point from the stable anchor point sequence, backtrack to the stable anchor point coordinates, and re-execute texture feature matching and coordinate correction to generate a stable trajectory.

[0023] Furthermore, the process of updating the historical coordinate database based on stable trajectory coordinates, sending the updated historical coordinate database back to the image acquisition module, evaluating the improvement in overall positioning reliability, and obtaining the final visual texture navigation AGV trajectory monitoring results includes:

[0024] Based on the stable trajectory, the coordinates and texture features of the stable anchor points are written into the historical coordinate database and sent back to the image acquisition module. The average deviation between the corrected position and the mileage position in the updated historical coordinate database is calculated and compared with the average deviation before the update. The monitoring results containing the stable trajectory coordinate sequence are then output.

[0025] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0026] This invention discloses a visual texture navigation AGV trajectory monitoring method. Addressing the core business scenario problem of AGV trajectory drift caused by texture confusion, positioning misjudgment, and accumulated mileage errors due to strong periodicity of ground textures, a multi-stage fusion strategy is employed. First, based on real-time image frames and historical trajectories, high-risk areas of texture confusion are initially identified and unique candidate anchor points are extracted. Then, the selection threshold is dynamically adjusted to retain preferred anchor points. Coordinate correction is performed based on the anchor point with the highest similarity. For abnormal offsets after correction, a backtracking drift suppression based on historically stable anchor points is triggered. Finally, the historical coordinate database is updated for continuous optimization. This method effectively suppresses positioning misjudgment and drift accumulation caused by periodic textures, improving the positioning accuracy and trajectory stability of AGVs in complex and repetitive texture environments, and ensuring the reliability and long-term consistency of trajectory monitoring. Attached Figure Description

[0027] Figure 1 This is a flowchart of a method for real-time monitoring of the running trajectory of a visual texture navigation AGV according to the present invention.

[0028] Figure 2 This is a schematic diagram of a method for real-time monitoring of the running trajectory of a visual texture navigation AGV according to the present invention.

[0029] Figure 3 This is another schematic diagram of a method for real-time monitoring of the running trajectory of a visual texture navigation AGV according to the present invention. Detailed Implementation

[0030] To further understand the content of this invention, a detailed description of the invention is provided in conjunction with the accompanying drawings and embodiments. The specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0031] like Figures 1-3 This embodiment of a method for real-time monitoring of the running trajectory of a visual texture navigation AGV may specifically include:

[0032] Step S101: Obtain real-time ground image frames and current mileage from the image acquisition module and position calculation module on the AGV chassis, and fuse them with pre-stored historical trajectory coordinates to obtain initial positioning data.

[0033] Real-time ground image frames are acquired from the image acquisition module under the AGV chassis. Based on the installation height of the image acquisition module and the lens field of view, the ground area covered by each image frame is determined. Simultaneously, the current mileage data output by the position calculation module is read, and the mileage data is aligned with the acquisition timestamp of the image frame to obtain a synchronized image sequence with mileage markers. For each image frame in the synchronized image sequence, the grayscale distribution features of the ground texture are extracted using a grayscale histogram statistical method. The pixel count ratio of each grayscale interval is recorded, and the difference in grayscale distribution features between adjacent frames is calculated. If the difference exceeds a preset threshold, the image frame is contrast-stretched, and the grayscale distribution features are re-extracted to obtain a corrected grayscale distribution feature sequence. Pre-stored historical trajectory coordinate data is retrieved, and the historical trajectory coordinate data is spatially mapped to the current mileage marker. The coordinate fusion weight is determined based on the cumulative deviation of the mileage data; the larger the cumulative deviation, the higher the weight of the historical trajectory coordinates. The specific weight calculation is as follows: , Where d is the cumulative deviation, wh is the weight of the historical trajectory coordinates, and wc is the weight of the corrected gray-scale distribution feature sequence. The corrected gray-scale distribution feature sequence and the historical trajectory coordinate data are superimposed and fused according to their weights. The fusion formula is as follows: Where fused is the fusion result, current is the gray-level distribution feature sequence, and historical is the historical trajectory coordinate data, the initial positioning data containing the texture gray-level distribution is obtained.

[0034] For example, in the image acquisition module configuration of the AGV chassis, the image acquisition module is usually installed below the chassis at a preset height above the ground, and the field of view of the lens determines the size of the ground area that a single frame image can cover.

[0035] Once the installation height and field of view parameters are determined, the actual ground coverage width and length corresponding to each frame of the image can be calculated through geometric projection relationships, thereby establishing the mapping basis between image pixel coordinates and ground physical coordinates.

[0036] In one possible implementation, the position calculation module continuously outputs the AGV's cumulative mileage data via an encoder or odometer, which is updated at a fixed frequency. To establish the correspondence between image frames and mileage data, a timestamp is recorded at each image acquisition, and simultaneously, the timestamp at the mileage data output is also recorded. Alignment processing is performed based on the matching degree of the two sets of timestamps, so that each image frame carries the mileage marker value corresponding to its acquisition time, thereby forming a synchronized image sequence with mileage markers.

[0037] Specifically, the grayscale histogram statistical method divides and counts the grayscale values ​​of all pixels in an image into intervals. The grayscale value range is divided into several intervals, and the number of pixels falling into each interval is counted. This number is then divided by the total number of pixels in the image to obtain the pixel proportion of each interval. The grayscale distribution feature is a numerical sequence composed of the pixel proportions of each grayscale interval, reflecting the light and dark distribution pattern of the ground texture. For two adjacent frames, the difference in the proportions of corresponding intervals in their grayscale distribution feature sequences is calculated. The absolute values ​​of all interval differences are summed to obtain the difference value, which is used to measure the severity of texture changes between adjacent frames. Contrast stretching is an image enhancement processing method. Its principle is to linearly expand the original grayscale value range of the image to a wider grayscale interval, making the grayscale distribution more uniform. When the difference value between adjacent frames exceeds a preset threshold, it indicates that the image may have lighting changes or texture blurring. In this case, contrast stretching is performed on the frame, and the grayscale distribution features are re-extracted to obtain a corrected grayscale distribution feature sequence.

[0038] In one embodiment, the historical trajectory coordinate data is a set of location points recorded by the AGV during previous operations, with each location point containing a corresponding mileage value and spatial coordinates. The current mileage marker is compared with the mileage value in the historical trajectory coordinate data to find historical location points with similar mileage values, establishing a spatial mapping relationship between the current frame and historical location points. The cumulative deviation refers to the gradually accumulating error between the mileage data output by the position calculation module and the actual travel distance. This deviation can be obtained through multiple calibration runs or comparison with external reference points.

[0039] Preferably, the coordinate fusion weights are adjusted based on the magnitude of the cumulative deviation. When the cumulative deviation is large, it indicates a decrease in the reliability of the current mileage data, thus increasing the weight of historical trajectory coordinates in the fusion. Conversely, when the cumulative deviation is small, the weight of historical trajectory coordinates is decreased. The weighted superposition fusion involves multiplying the position estimate corresponding to the corrected grayscale distribution feature sequence and the historical trajectory coordinates by their respective weights, then summing the results to obtain the fused position coordinates. These position coordinates are then combined with the corresponding grayscale distribution features to form initial positioning data containing textured grayscale distribution. Optionally, a Fast Fourier Transform (FFT) is performed on the corrected grayscale distribution feature sequence to convert the time-domain signal into a frequency-domain signal. By analyzing the peak distribution in the frequency domain, the most significant frequency component is identified, and the period corresponding to this frequency is the texture period length of the ground. This is because recurring ground textures produce obvious peaks in the frequency domain, and the reciprocal of the frequency corresponding to this peak is the texture period length.

[0040] Step S102: Based on the initial positioning data, evaluate the repetition degree of similar patterns between adjacent image frames and the level of a preset threshold. Mark the positions that exceed the threshold as high-risk areas for texture confusion and determine potential misjudgment positions.

[0041] Based on the texture period length in the initial positioning data, a sliding detection window is set along the AGV's travel direction. The length of the sliding detection window is set to a preset multiple of the texture period length. Gray-level distribution feature vectors of adjacent image frames are extracted within the window. A normalized cross-correlation method is used to calculate the correlation coefficient between the gray-level distribution feature vectors of consecutive image frames, obtaining the inter-frame similarity value. Based on the inter-frame similarity value, the number of image frames within the sliding detection window whose inter-frame similarity value exceeds a preset similarity threshold is counted. The ratio of this number to the total number of frames within the window is calculated as the similarity pattern repetition. The similarity pattern repetition is compared with the preset repetition threshold. If the similarity pattern repetition is higher than the preset repetition threshold, the path interval corresponding to the current sliding detection window is marked as a high-risk area for texture confusion. The start and end coordinates of the high-risk area for texture confusion are obtained. Combined with the travel mileage records corresponding to each image frame, the acquisition position of each image frame within the high-risk area for texture confusion is determined as a potential misjudgment position.

[0042] Specifically, the sliding detection window moves along the AGV's travel direction. The window length is determined by a preset multiple of the texture cycle length, typically set between three and five times, to cover a sufficient number of texture repetition cycles. The window contains multiple continuously acquired ground images, each corresponding to a capture time and travel position.

[0043] In one possible implementation, the grayscale distribution feature vector is extracted using a line-by-line scanning method. For each frame of the image within the window, the image is divided into several equal-width strip regions along the AGV's travel direction. The mean grayscale value of all pixels within each strip region is calculated, and these mean grayscale values ​​are arranged sequentially to form a one-dimensional vector, which is the grayscale distribution feature vector of that frame. The grayscale distribution feature vectors of adjacent frames are used to calculate similarity using a normalized cross-correlation method. This method first performs zero-mean processing on both vectors, then calculates the dot product of the two processed vectors and divides it by the product of their respective vector magnitudes. The resulting correlation coefficient ranges from negative one to positive one; the closer the value is to positive one, the more similar the texture distribution of the two frames. The similarity pattern repetition is statistically analyzed within the sliding detection window. The inter-frame similarity values ​​of all adjacent frame pairs within the window form a numerical sequence. The number of pairs exceeding a preset similarity threshold in this sequence is counted, and this number is divided by the total number of frame pairs within the window to obtain the similarity pattern repetition.

[0044] Preferably, the preset repetition threshold is adjusted based on the texture complexity in the actual application scenario. When the repetition of similar patterns exceeds the preset repetition threshold, it indicates that there are a large number of similar texture patterns within the path range covered by the current window, which can easily lead to positioning confusion. Therefore, this path range is marked as a high-risk area for texture confusion. Based on the above marking results, the start and end coordinates of the high-risk area for texture confusion are obtained. Combined with the mileage record during image frame acquisition, the acquisition position of each image frame within the area is determined as a potential misjudgment position.

[0045] Step S103: Based on the high-risk areas of texture confusion and potential misjudgment locations, obtain the real-time image frames and historical coordinates within the corresponding range, analyze the periodic deviation of local textures, extract the local grayscale gradients and analyze their uniqueness in the global path to obtain candidate anchor points.

[0046] Based on the start and end coordinates of the high-risk texture confusion area, a real-time image frame sequence within the corresponding range is acquired from the image acquisition module. Historical coordinates and corresponding texture feature records stored within this range are extracted from the historical coordinate database to obtain a set of image frames to be processed and a set of historical texture features. For the set of image frames to be processed, the interval distance between repeated occurrences of adjacent texture patterns is statistically analyzed along the AGV's travel direction. The difference between this interval distance and the pre-stored standard texture period length is calculated. The difference is divided by the standard texture period length to obtain the periodic deviation ratio. Image frame positions with periodic deviation ratios exceeding a preset deviation threshold are selected to obtain a set of periodic abnormal positions. For each position in the set of periodic abnormal positions, the gray-level difference sequence of adjacent pixels along the travel direction of the corresponding image frame is extracted as a local gray-level gradient feature vector. Euclidean distance is used to calculate the vector distance between this local gray-level gradient feature vector and the feature vectors at each position in the historical texture feature set, obtaining the distance distribution sequence of the current position. Based on the distance distribution sequence, the minimum vector distance between the current position and all positions in the historical texture feature set is obtained. If the minimum vector distance is greater than a preset discrimination threshold, the current position is determined to be unique in the global path and marked as a candidate anchor point. After traversing the periodic abnormal position set to complete all uniqueness determinations, a candidate anchor point set is obtained.

[0047] In one possible implementation, during the candidate anchor point acquisition process, the start and end coordinates of the high-risk texture confusion segment define the path range to be processed. Based on the coordinate boundaries of this range, the image acquisition module retrieves the corresponding real-time image frame sequence from the storage buffer, and simultaneously extracts the recorded historical coordinates and corresponding texture feature data within this range from the historical coordinate library, forming a pairing relationship between the set of image frames to be processed and the historical texture feature set.

[0048] Specifically, for each frame in the image frame set to be processed, the boundary positions of adjacent texture patterns are detected along the AGV's traveling direction. The pixel distance between the boundaries of two identical patterns is calculated and converted into actual physical distance, which is the measured texture period at the current position. The difference between the measured texture period and the standard texture period length pre-stored in the historical coordinate library is calculated. The difference reflects the degree of deviation between the texture distribution at the current position and the standard distribution. The periodic deviation ratio is calculated by dividing the difference by the standard texture period length, resulting in a dimensionless ratio value. The larger this value, the more significant the deviation between the texture period at the current position and the standard value. The periodic deviation originates from various factors in the actual ground environment.

[0049] For example, in warehouse aisles, slight differences in the width of the grout lines when tiles are laid, or wear and tear in certain areas after long-term use causing blurred texture edges, can all cause deviations between the measured texture period and the standard value. When the periodic deviation ratio exceeds a preset deviation threshold, it indicates that the texture distribution at that location has abnormal characteristics that distinguish it from the surrounding area. These abnormal characteristics constitute a unique identifier for that location.

[0050] Preferably, the preset deviation threshold is set based on the overall uniformity of the ground texture in the actual application scenario. In environments with relatively regular texture coverage, the deviation threshold is set lower to filter out more locations with subtle differences; in environments where the texture itself has large fluctuations, the deviation threshold is set higher to avoid misidentifying normal fluctuations as abnormal features. Based on the filtering results of the periodic abnormal location set, the local gray-level gradient feature vector of each location is further extracted. The gray-level gradient reflects the intensity of the change in gray-level values ​​between adjacent pixels in the image. The image frame is scanned column by column along the AGV's travel direction, and the difference between the average gray-level value of each column of pixels and the average gray-level value of adjacent columns of pixels is calculated. These differences are arranged in column order to form a one-dimensional numerical sequence, which is the local gray-level gradient feature vector of that location.

[0051] For example, if a location happens to be at the boundary between two tiles of different colors, the grayscale gradient feature vector at that location will show a large numerical jump at the boundary, while the grayscale gradient feature vector inside a single-color tile will have a generally flat value. This differentiated feature vector distribution provides a basis for comparison in subsequent uniqueness determination.

[0052] In one embodiment, Euclidean distance is used as a measure of the similarity between two feature vectors. For a given location in the set of periodic anomaly locations, its local gray-level gradient feature vector is compared with the feature vectors of each recorded location in the historical texture feature set using Euclidean distance calculation. The calculation process involves subtracting the values ​​of the corresponding positions of the two feature vectors, squared the result, summed the squared values ​​of all positions, and taking the square root to obtain a distance value that characterizes the overall difference between the two vectors. A smaller distance value indicates that the texture features of the two locations are more similar, while a larger distance value indicates that the texture features of the two locations are more significantly different.

[0053] Understandably, the distance distribution sequence records the distribution of feature differences between the current location and all recorded locations on the global path. The minimum vector distance is extracted from this sequence; this minimum value represents the degree of difference between the current location and the most similar location in the historical records. If the minimum vector distance is greater than a preset discrimination threshold, it indicates that even the most similar historical location has sufficiently significant feature differences, thus determining that the current location is unique in the global path and marking it as a candidate anchor point. Further, the uniqueness determination process is sequentially executed on all locations in the periodic abnormal location set. After completing the entire traversal, all marked locations are aggregated to form a candidate anchor point set. Each anchor point in this set simultaneously satisfies two conditions: its texture periodic deviation exceeds the normal range, and its grayscale gradient features are unique in the global path, thus enabling it to serve as a reliable positioning reference point.

[0054] Step S104: For candidate anchor points, dynamically adjust the screening threshold value based on the cumulative error of driving mileage, and retain anchor points with threshold values ​​lower than the average length of texture period to obtain the preferred anchor points.

[0055] For each anchor point in the candidate anchor point set, the mileage value corresponding to that anchor point is obtained. A preset cumulative error coefficient per unit distance is read from the location calculation module. The mileage value is multiplied by the cumulative error coefficient to obtain the cumulative mileage error value for that anchor point. Using the cumulative mileage error value as a filtering threshold, the average texture period length of the global path is obtained from the historical coordinate database. The filtering threshold is compared with the average texture period length. If the filtering threshold is less than the average texture period length, the current candidate anchor point is marked as a preferred anchor point and retained; otherwise, the candidate anchor point is excluded. After traversing the candidate anchor point set and completing all judgments, a preferred anchor point set is obtained.

[0056] In one possible implementation, during the preferred anchor point selection process, the mileage value is obtained based on the cumulative distance traveled by the AGV since its departure from the starting point. For each anchor point in the candidate anchor point set, the mileage count value corresponding to the time the anchor point was detected is read from the position calculation module. This value reflects the total distance traveled by the AGV from its current starting position.

[0057] The cumulative error coefficient per unit distance represents the average positioning deviation generated by the AGV per unit distance traveled. This coefficient is pre-calibrated by the position calculation module based on encoder accuracy, wheel wear, and the probability of ground slippage. For example, if the AGV's encoder resolution and wheel diameter parameters determine that approximately 0.5 meters of cumulative deviation will occur for every 100 meters traveled, then the cumulative error coefficient per unit distance is set to 0.5%. Multiplying the traveled mileage value by this coefficient yields the cumulative mileage error value for the current position, which represents the maximum possible positioning deviation range of the AGV at the current position. The cumulative mileage error value is directly used as a screening threshold for subsequent comparison and judgment. The average length of the texture cycle is obtained from the historical coordinate database and represents the average interval distance between repeated occurrences of ground texture patterns on the current path. The comparison between the screening threshold and the average length of the texture cycle has a clear business meaning. When the screening threshold is lower than the average length of the texture cycle, it indicates that the cumulative positioning error at the current position has not yet exceeded one complete texture cycle, and the AGV's visual positioning can still distinguish the current position from other positions within adjacent texture cycles. Therefore, the reliability of this candidate anchor point is relatively high.

[0058] Specifically, when traversing the candidate anchor point set, the above comparison and judgment are performed on each anchor point. If the screening threshold is lower than the average length of the texture period, the anchor point is marked as a preferred anchor point; if the screening threshold is higher than or equal to the average length of the texture period, it indicates that the cumulative error has exceeded the distinguishable range, and the anchor point is at risk of being confused with other period positions, and is therefore excluded. After completing the entire traversal, the remaining anchor points constitute the preferred anchor point set, in which all anchor points meet the condition that the positioning error is controllable.

[0059] Step S105: From the preferred anchor points, select the texture segment with the highest similarity between the texture features of each anchor point and the grayscale distribution of the current image frame as the matching benchmark, evaluate the magnitude of the deviation and allowable range after coordinate correction, and determine the corrected trajectory coordinates.

[0060] For each anchor point in the preferred anchor point set, the texture feature vector corresponding to that anchor point is obtained. The current real-time image frame is acquired from the image acquisition module, and its grayscale distribution feature vector is extracted. The correlation coefficient between the texture feature vector of each anchor point and the grayscale distribution feature vector of the current image frame is calculated using the normalized cross-correlation method to obtain the similarity value of each anchor point. Based on the similarity value of each anchor point, the anchor point with the highest similarity value is selected from the preferred anchor point set. The texture segment corresponding to this anchor point is used as the matching reference. The known coordinates of the matching reference recorded in the historical coordinate database are obtained. The pixel distance of the matching reference texture segment in the current image frame relative to the center of the image frame is calculated and converted into a physical distance offset. The known coordinates of the matching reference are superimposed with the physical distance offset to obtain the corrected coordinates. The deviation value between the corrected coordinates and the current mileage coordinates recorded by the position calculation module is obtained. The deviation value is compared with a preset allowable deviation range. If the deviation value is within the allowable deviation range, the corrected coordinates are determined as the corrected trajectory coordinates; if the deviation value exceeds the allowable deviation range, the current mileage coordinates are kept as the corrected trajectory coordinates.

[0061] In one possible implementation, for each anchor point in the preferred set of anchor points, the texture feature vector stored at that anchor point is read from the historical coordinate library, and the grayscale distribution feature vector is extracted from the current real-time image frame. The dimensions of the two vectors are kept consistent for comparison operations.

[0062] The calculation process of the normalized cross-correlation method includes two stages: vector standardization and correlation coefficient calculation.

[0063] Specifically, the anchor point texture feature vector and the current image frame grayscale distribution feature vector are both zero-mean processed, i.e., each vector element is subtracted from its arithmetic mean to obtain a standardized vector after removing the DC component. The two standardized vectors are then multiplied element-wise and summed, and then divided by the product of their respective magnitudes to obtain a correlation coefficient ranging from negative one to positive one. This coefficient is the similarity value of the current anchor point. The closer the correlation coefficient is to positive one, the more closely the texture distribution of the current image frame matches the texture features stored at the anchor point. After traversing the optimized set of anchor points to complete all similarity calculations, the anchor point with the highest similarity value is selected as the matching benchmark. The texture segment corresponding to this anchor point has calibrated physical coordinates in the historical coordinate database, and these known coordinates become the reference benchmark point for the correction calculation.

[0064] Preferably, the coordinate conversion process converts the pixel distance in the image domain into actual physical coordinates. The center pixel position of the matching reference texture segment is located in the current image frame, and the pixel distance between this position and the center point of the image frame is calculated. Based on the calibration parameters of the image acquisition module, the pixel distance is converted into a physical distance offset. The known coordinates of the matching reference are then superimposed with this physical distance offset to obtain the corrected coordinates. The validity of the correction result is determined by comparing the deviation between the corrected coordinates and the current mileage coordinates. If the deviation is within the allowable deviation range, it indicates that the texture matching correction result and the mileage measurement result corroborate each other, and the corrected coordinates have high reliability. If the deviation exceeds the allowable deviation range, the current mileage coordinates are retained as the output result to avoid abnormal correction values ​​affecting the continuity of trajectory recording.

[0065] Step S106: Obtain the offset of the corresponding position of the corrected trajectory coordinate record and the real-time image frame, evaluate the offset difference between the previous and next frames and the magnitude of the preset change threshold, and trigger backtracking drift suppression based on historical stable anchor points for offsets exceeding the threshold, that is, backtrack to the previous stable anchor point position and re-execute texture matching positioning to obtain the suppressed stable trajectory.

[0066] The process involves obtaining the pairing records of the corrected trajectory coordinates and the corresponding real-time image frames, extracting the pixel position of the matching reference texture segment relative to the center of the current image frame as the current position offset, reading the position offset of the previous image frame from the cache, and calculating the difference between the current position offset and the position offset of the previous image frame to obtain the inter-frame offset difference value. Based on this inter-frame offset difference value, it is compared with a preset change threshold. If the inter-frame offset difference value is less than or equal to the preset change threshold, the current trajectory coordinates are determined to be in a stable state, and the currently corrected preferred anchor point is marked as a stable anchor point and stored in the stable anchor point sequence. If the inter-frame offset difference value is greater than the preset change threshold, the current position is determined to have a drift anomaly. For the current position with the drift anomaly, the stable anchor point closest to the current position is obtained from the stable anchor point sequence as the backtracking target anchor point, and the AGV's position state is reverted to the historical coordinate position corresponding to the backtracking target anchor point. Based on the position of the backtracking target anchor point, the current real-time image frame is re-acquired from the image acquisition module, and a preferred set of anchor points is reconstructed around the backtracking target anchor point. The set is traversed to select the anchor point with the highest similarity as the matching benchmark to perform texture feature matching and coordinate correction. The re-corrected coordinates are used as the trajectory coordinates after drift suppression to obtain a stable trajectory.

[0067] In one possible implementation, for each frame of a real-time image after coordinate correction, the center pixel coordinates of the matching reference texture segment are located in the image. The pixel distance between these coordinates and the center point of the image frame is calculated and recorded as the current position offset. The position offset reflects the degree of deviation of the AGV's current actual position from the center of the image acquisition field of view.

[0068] The calculation of the inter-frame offset difference is based on the comparison of the position offsets of two consecutive image frames. The position offset value recorded in the previous image frame is read from the buffer, and the position offset of the current image frame is subtracted from the position offset of the previous image frame. The absolute value of the difference is taken as the inter-frame offset difference. Under normal AGV operation, since the vehicle moves smoothly along a predetermined trajectory, the change in position offset between consecutive frames should remain within a small range. If the inter-frame offset difference suddenly increases, it indicates a discontinuous jump in the positioning result, which may indicate drift. The setting of the preset change threshold needs to comprehensively consider the normal operating speed of the AGV and the image acquisition frame rate.

[0069] For example, if the maximum travel speed of the AGV is 1.5 meters per second and the image acquisition frame rate is 20 frames per second, then the maximum displacement of the AGV between two adjacent frames is about 7.5 centimeters. The preset change threshold can be set to two to three times the maximum displacement to accommodate position fluctuations during normal travel and identify abnormal jumps.

[0070] Specifically, the stable anchor point sequence is established using a dynamic appending method. When the inter-frame offset difference is less than or equal to a preset change threshold, the current trajectory coordinates are determined to be in a stable state. The preferred anchor point used for the current correction is marked as a stable anchor point and appended to the end of the stable anchor point sequence in chronological order. Each anchor point in the stable anchor point sequence records its corresponding historical coordinate position, texture feature vector, and timestamp information when it was marked as stable. As the AGV continues to move, the stable anchor point sequence grows continuously, forming a trajectory skeleton composed of reliable positioning reference points.

[0071] In one embodiment, the determination of drift anomalies triggers a backtracking process. When the inter-frame offset difference value is greater than a preset change threshold, it indicates that the positioning result of the current position is significantly discontinuous from the previous moment, and is determined to be a drift anomaly. Drift anomalies may be caused by various factors, such as water accumulation on the ground causing temporary failure of the image acquisition module, AGV passing through a strongly lit area causing image overexposure, or mismatch during texture matching. Regardless of the specific cause, once a drift anomaly is detected, suppression processing is immediately initiated to prevent erroneous positioning results from affecting subsequent trajectory recording.

[0072] Preferably, the selection of the backtracking target anchor point is based on the principle of proximity. All recorded stable anchor points are traversed from the most recent 20 stable anchor point sequences. The Euclidean distance between the historical coordinates of each anchor point and the current anomalous position is calculated, and the anchor point with the smallest distance is selected as the backtracking target anchor point. The reason for selecting the closest stable anchor point instead of directly selecting the anchor point at the end of the sequence is that when a drift anomaly occurs, the current position record may have deviated significantly from the true position. Using the closest stable anchor point as the backtracking target can minimize the magnitude of position regression and reduce the time cost required for repositioning.

[0073] Understandably, the position state rollback operation resets the AGV's positioning reference point to the historical coordinates of the target anchor point. This operation temporarily abandons the trajectory record during the drift anomaly, using the target anchor point as the new positioning starting point. Position state rollback does not control the physical movement of the AGV, but rather resets the coordinates at the positioning data level, allowing subsequent texture matching positioning to restart based on a reliable historical position. Further, a real-time image frame is acquired from the image acquisition module, and the grayscale distribution feature vector of this image frame is extracted. Using the texture feature vector stored at the target anchor point as a matching template, a normalized cross-correlation method is used to calculate the similarity between the two and verify the matching validity. If the match is successful, the corrected coordinates are calculated based on the known coordinates of the target anchor point and the pixel offset relationship in the image frame. These coordinates are then output as the drift-suppressed trajectory coordinates, restoring the continuous recording of trajectory coordinates and obtaining a stable trajectory.

[0074] For example, in a long, straight aisle in a warehouse, when an AGV travels to a certain point, ground reflection causes a sudden change in texture matching across three consecutive image frames. The inter-frame offset difference suddenly increases to more than five times the normal value, triggering a drift suppression process. The nearest stable anchor point in the stable anchor point sequence is located three meters prior, corresponding to a tile seam feature. After reverting to this anchor point, the current image frame is re-acquired and texture matching is performed. Since the AGV has now left the reflective area, the matching process returns to normal, resulting in stable trajectory coordinates after drift suppression, allowing for continuous trajectory recording.

[0075] Step S107: Update the historical coordinate library based on the stable trajectory coordinates, send the updated historical coordinate library back to the image acquisition module, evaluate the improvement in overall positioning reliability, and obtain the final visual texture navigation AGV running trajectory monitoring results.

[0076] Based on the stable trajectory coordinates, the coordinate positions and corresponding texture features of each anchor point in the stable anchor point sequence are written into the historical coordinate database, and the updated historical coordinate database is sent back to the image acquisition module. The deviation values ​​between each corrected position and the mileage position recorded in the updated historical coordinate database are obtained, and the arithmetic mean of the deviation values ​​is calculated as the positioning deviation mean. The positioning deviation mean before and after the update is compared, and the improvement in overall positioning reliability is evaluated based on the change in the positioning deviation mean. The visual texture navigation AGV running trajectory monitoring results, including the stable trajectory coordinate sequence and the positioning reliability evaluation results, are output.

[0077] In one possible implementation, during the historical coordinate database update process, the verified anchor point data from the stable anchor point sequence is written into the historical coordinate database. The data written for each anchor point includes its physical coordinate position, the corresponding texture feature vector, and the timestamp information indicating it is a stable anchor point. The updated historical coordinate database is then transmitted back to the image acquisition module via a data communication interface, providing subsequent texture matching and localization processes with richer reference anchor point resources.

[0078] The mean positioning deviation is calculated based on the statistical difference between the corrected position and the corresponding mileage position recorded in the historical coordinate database. The mean positioning deviation is obtained by arithmetically averaging all deviation values. By comparing the change in the mean positioning deviation before and after the update, the improvement in overall positioning reliability within the current operating cycle is assessed. The final output of the visual texture navigation AGV trajectory monitoring results includes two parts: a stable trajectory coordinate sequence and a positioning reliability assessment result, providing data support for the AGV's autonomous navigation control and operational status monitoring.

[0079] The above description is merely a specific implementation of this specification. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the scope of protection of this specification is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this specification, and these modifications or substitutions should all be covered within the scope of protection of this specification.

Claims

1. A method for real-time monitoring of the running trajectory of a visual texture-guided AGV, characterized in that, The method includes: The image acquisition module and position calculation module on the AGV chassis acquire real-time ground image frames and current driving distance, and combine them with pre-stored historical trajectory coordinates to obtain initial positioning data; Based on the initial positioning data, the similarity of patterns between adjacent image frames is evaluated and compared with a preset threshold. Locations that exceed the threshold are marked as high-risk areas for texture confusion, thus identifying potential misjudgment locations. Based on the coordinate range of the high-risk texture confusion area, a real-time image frame sequence is acquired from the image acquisition module. Historical coordinates and texture features of the corresponding range are extracted from the historical coordinate library. The interval distance between adjacent texture patterns is counted. The difference between the interval distance and the standard texture period length is calculated. The periodic deviation ratio is obtained by dividing the difference by the standard texture period length. Image frame positions that exceed a preset deviation threshold are selected. For the selected positions, the gray-level difference sequence of adjacent pixels is extracted as a local gray-level gradient feature vector. The vector distance between the local gray-level gradient feature vector and the historical texture features is calculated. Positions with a minimum vector distance greater than a preset discrimination threshold are determined to be unique positions and marked as candidate anchor points. A set of candidate anchor points is generated by traversing all positions. For candidate anchor points, the screening threshold is dynamically adjusted in combination with the cumulative error of driving mileage. Anchor points with threshold values ​​lower than the average length of texture period are retained to obtain the preferred anchor points. From the preferred anchor points, the texture segment with the highest similarity between the texture features of each anchor point and the grayscale distribution of the current image frame is selected as the matching benchmark. The pixel distance of the matching benchmark texture segment in the current image frame relative to the center of the image frame is calculated and converted into a physical distance offset. The known coordinates of the matching benchmark are superimposed with the physical distance offset to obtain the corrected coordinates. The deviation between the corrected coordinates and the current mileage coordinates is taken as the coordinate correction deviation. The magnitude of the coordinate correction deviation and the allowable range are evaluated to determine the corrected trajectory coordinates. Obtain the offset of the corresponding position of the corrected trajectory coordinate record and the real-time image frame, evaluate the offset difference between the previous and next frames and the magnitude of the preset change threshold, and trigger backtracking drift suppression based on historical stable anchor points for offsets exceeding the threshold, that is, backtrack to the previous stable anchor point position and re-perform texture matching localization to obtain the suppressed stable trajectory.

2. The method for real-time monitoring of the running trajectory of a visual texture-guided AGV according to claim 1, characterized in that, The process of acquiring real-time ground image frames and current mileage from the image acquisition module and position calculation module on the AGV chassis, and fusing them with pre-stored historical trajectory coordinates to obtain initial positioning data includes: The system acquires real-time ground image frames from the image acquisition module, obtains the current driving mileage from the location calculation module, matches the real-time ground image frames with the pre-stored historical trajectory coordinates, extracts the texture grayscale distribution features and the period length of the texture pattern from the real-time ground image frames, and generates initial positioning data containing the texture grayscale distribution features and the period length.

3. The method for real-time monitoring of the running trajectory of a visual texture-guided AGV according to claim 1, characterized in that, The step of evaluating the repetition degree of similar patterns between adjacent image frames based on initial positioning data and comparing it with a preset threshold, marking locations exceeding the threshold as high-risk areas for texture confusion, and determining potential misjudgment locations includes: A sliding detection window is set up, and grayscale distribution feature vectors of adjacent image frames within the sliding detection window are extracted. The inter-frame correlation coefficient of the grayscale distribution feature vectors is calculated, and the ratio of the number of image frames exceeding a preset similarity threshold to the total number of frames is used as the similarity pattern repetition. The similarity pattern repetition is compared with the preset repetition threshold, and the path intervals exceeding the preset repetition threshold are marked as high-risk areas for texture confusion. The acquisition positions within the high-risk areas for texture confusion are determined as potential misjudgment positions.

4. The method for real-time monitoring of the running trajectory of a visual texture-guided AGV according to claim 1, characterized in that, The process involves dynamically adjusting the screening threshold for candidate anchor points based on accumulated mileage error, retaining anchor points with threshold values ​​lower than the average length of the texture period, thus obtaining preferred anchor points, including: For the candidate anchor points, the corresponding mileage value is obtained, the cumulative mileage error value is calculated as the screening threshold, the average length of the texture period is obtained from the historical coordinate library, the screening threshold is compared with the average length of the texture period, and candidate anchor points with a length lower than the average length of the texture period are retained to generate a preferred anchor point set.

5. The method for real-time monitoring of the running trajectory of a visual texture-guided AGV according to claim 1, characterized in that, The process of selecting the texture segment with the highest similarity between the texture features of each anchor point and the grayscale distribution of the current image frame from the preferred anchor points as the matching benchmark, evaluating the magnitude of the deviation after coordinate correction and the allowable range, and determining the corrected trajectory coordinates includes: For the preferred anchor points, the correlation coefficient between the texture feature vector of each anchor point and the grayscale distribution feature vector of the current image frame is calculated. The texture segment corresponding to the anchor point with the highest correlation coefficient is selected as the matching reference. The known coordinates of the matching reference are obtained. The pixel distance of the matching reference texture segment in the current image frame relative to the center of the image frame is calculated and converted into a physical distance offset. This offset is then superimposed on the known coordinates to generate corrected coordinates. The deviation between the corrected coordinates and the current mileage coordinates is compared with the allowable deviation range to determine the corrected trajectory coordinates.

6. The method for real-time monitoring of the running trajectory of a visual texture-guided AGV according to claim 1, characterized in that, The process involves obtaining the offset of the corresponding position between the corrected trajectory coordinate record and the real-time image frame, evaluating the offset difference between consecutive frames and the magnitude of a preset change threshold, and triggering backtracking drift suppression based on historical stable anchor points for offsets exceeding the threshold. This involves reverting to the previous stable anchor point position and re-performing texture matching localization to obtain a suppressed stable trajectory, including: Extract the current position offset corresponding to the corrected trajectory coordinates, calculate the difference between the position offset and the previous image frame position offset as the inter-frame offset difference value, compare the inter-frame offset difference value with a preset change threshold, and mark anchor points whose offset difference value is less than or equal to the preset change threshold as stable anchor points; for offsets exceeding the preset change threshold, select the nearest stable anchor point from the stable anchor point sequence, backtrack to the stable anchor point coordinates, and re-execute texture feature matching and coordinate correction to generate a stable trajectory.

7. The method for real-time monitoring of the running trajectory of a visual texture-guided AGV according to claim 1, characterized in that, The method further includes: updating the historical coordinate library based on the stable trajectory coordinates, sending the updated historical coordinate library back to the image acquisition module, evaluating the degree of improvement in overall positioning reliability, and obtaining the final visual texture navigation AGV running trajectory monitoring results.

8. The method for real-time monitoring of the running trajectory of a visual texture-guided AGV according to claim 7, characterized in that, The process of updating the historical coordinate database based on stable trajectory coordinates, sending the updated historical coordinate database back to the image acquisition module, evaluating the improvement in overall positioning reliability, and obtaining the final visual texture navigation AGV trajectory monitoring results includes: Based on the stable trajectory, the coordinates and texture features of the stable anchor points are written into the historical coordinate database and sent back to the image acquisition module. The average deviation between the corrected position and the mileage position in the updated historical coordinate database is calculated and compared with the average deviation before the update. The monitoring results containing the stable trajectory coordinate sequence are then output.

Images