An AGV visual texture navigation path map real-time generation method

Abstract

The application provides an AGV visual texture navigation path map real-time generation method, comprising the following steps: collecting ground glossiness, gray saturation area, comparing the brightness difference between the gray saturation area and the normal area, and identifying the distribution position and coverage range of the high-reflective area; according to the distribution position and coverage range of the high-reflective area, extracting the texture missing part covered by the mirror reflection spot, and determining the boundary coordinates of the texture missing part; according to the optimized light incidence angle distribution, collecting the gray jump amplitude when the AGV drives from the rough cement ground into the polished ceramic tile area, and evaluating the transition intensity of the material change; using the reconstructed complete texture feature, filling the texture missing area in the path map, integrating the texture information of the path map, and generating a complete AGV navigation path map.

Description

Technical Field

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

[0002] In modern industry and logistics, visual navigation technology for Automated Guided Vehicles (AGVs) plays a crucial role, directly impacting their path planning and operational efficiency in complex environments. Especially in generating navigation path maps based on ground texture, accurately capturing ground feature information is key to ensuring navigation accuracy. This technology not only affects the operational stability of AGVs but also has irreplaceable value in improving automated production efficiency. However, existing methods often struggle to adapt to the interference caused by the varying light reflection characteristics of different surfaces when dealing with changes in ground material. Many solutions neglect the impact of material reflectivity on image quality, particularly under complex lighting conditions, failing to effectively balance the extraction of image details with the suppression of interference factors. This limitation makes navigation map generation unstable in certain scenarios, especially when the ground material transitions from rough to smooth. Furthermore, different materials reflect light in vastly different ways. For example, rough surfaces scatter more light, making image details relatively easy to capture, while smooth surfaces easily produce strong specular reflections, forming bright patches. These bright areas obscure the original texture features of the ground, leading to the loss of image information. Moreover, this reflective property is not constant; it is affected by the intensity and angle of external light sources, thus posing a challenge to the stability of image acquisition. Especially in real-world industrial scenarios, ground materials often change gradually or mix, such as a slow transition from rough concrete in a warehouse to a smooth stainless steel platform, or from non-slip rubber flooring to easy-clean tile aisles in a food processing plant. These sudden or gradual changes in material reflectivity can cause interruptions in the continuity of AGV-acquired images. The first few frames, rich in texture for map construction, suffer from information loss due to high brightness saturation in subsequent frames, leading to broken, drifting, or false blank areas in the navigation map. Therefore, accurately capturing texture information and avoiding interference from bright areas in scenarios with dynamically changing ground material reflectivity has become a key issue in improving the real-time generation of AGV visual navigation path maps. Summary of the Invention

[0003] This invention provides a method for real-time generation of AGV visual texture navigation path maps, mainly including:

[0004] Obtain the location and coverage of highly reflective areas in ground images;

[0005] The boundary coordinates of the texture-deficient parts are determined based on the distribution location and coverage of the highly reflective areas;

[0006] The optimized light incident angle distribution is obtained by adjusting the illumination parameters of the controllable light source based on the boundary coordinates of the texture missing parts.

[0007] Ground image sequences were acquired based on the optimized light incident angle distribution, and the transition intensity of material changes was evaluated.

[0008] The intensity of the controllable light source is adjusted according to the transition intensity of the material change, and the adjusted ground image is acquired to restore the texture information covered by the highly reflective area.

[0009] Based on the recovered texture information, identify texture breakpoints and fuse surrounding edge details to generate map detail filling information, reconstructing the complete texture features of highly reflective areas;

[0010] By using the reconstructed complete texture features to fill in the texture-missing areas in the path map, an AGV navigation path map containing complete texture features is generated.

[0011] Furthermore, acquiring the distribution location and coverage area of ​​highly reflective areas in the ground image includes:

[0012] The ratio of reflected light intensity to incident light intensity is collected as the gloss coefficient and combined with the gray-scale saturation area to determine the reflective intensity. Edge detection and connected component analysis are used to extract the closed contour enclosed by the strong reflective boundary. The centroid of the closed contour is calculated as the center coordinate of the highly reflective patch. The controllable light source illumination angle is adjusted according to the center coordinate of the highly reflective patch and the offset distance of the camera optical axis. The standard deviation of the brightness distribution before and after adjustment is compared. The illumination angle with the smallest standard deviation is selected as the current illumination configuration parameter. Based on the illumination configuration parameter, the image is re-acquired and the pixel coordinate range of the closed contour and the size of the circumscribed rectangle are statistically analyzed to obtain the distribution location and coverage of the highly reflective area.

[0013] Furthermore, determining the boundary coordinates of the texture-missing portion based on the distribution location and coverage area of ​​the highly reflective region includes:

[0014] Based on the distribution location of highly reflective areas, the original grayscale values ​​are read to extract the set of pixels exceeding the saturation threshold. Morphological dilation is used to expand the saturation region to obtain the specular reflection coverage area. Edge curves are fitted by edge detection and least squares method, and texture missing boundary feature points are marked according to curvature differences. The angle difference of the main direction of the gradient direction histogram inside and outside the boundary is calculated, and the contour is finely adjusted along the normal direction to minimize the angle difference, thus obtaining the adjusted texture missing boundary. Polygon approximation is performed on the adjusted texture missing boundary to obtain the clockwise vertex coordinate sequence and the Euclidean distance between adjacent vertices as the boundary coordinate set of the texture missing part.

[0015] Furthermore, the step of adjusting the illumination parameters of the controllable light source based on the boundary coordinates of the texture-deficient portion to obtain the optimized light incident angle distribution includes:

[0016] The initial opening parameters of the light shield are determined by calculating the polygon area, centroid, and circumcircle radius of the texture-deficient region based on the boundary coordinate set, and the blade servo motor is controlled to adjust the opening area. The light spot contours under different opening areas are collected, and the occlusion configuration parameters are determined by minimizing the number of overlapping pixels between the light spot edge and the boundary of the texture-deficient region. Based on the occlusion configuration parameters, the incident angle of each point in the light spot is calculated to generate an incident angle distribution histogram, and the peak value is extracted as the main incident angle. The main incident angle distribution is changed by rotating the azimuth angle of the light shield, and the optimized light incident angle distribution is determined by minimizing the angle between the light intensity gradient vector direction and the texture extension direction in the texture-deficient region.

[0017] Furthermore, the step of acquiring ground image sequences based on the optimized light incident angle distribution and evaluating the transition intensity of material changes includes:

[0018] Based on the optimized light incident angle distribution, a controllable light source is configured and ground image sequences are continuously acquired. When the average gray value difference between adjacent frames exceeds a preset threshold, the material transition start point is marked. Gray-level histograms are acquired at fixed intervals and the peak position offset value is calculated. When the offset value of multiple consecutive frames exceeds a preset offset threshold, the transition end point is marked. The gray-level jump amplitude is obtained by accumulating the absolute value of the offset value between the start point and the end point. The ratio of the gray-level jump amplitude to the length of the transition region is calculated as the transition gradient. The transition gradient value is used as the transition intensity evaluation value of the material change.

[0019] Furthermore, the step of adjusting the controllable light source illumination intensity based on the transition intensity of material changes and acquiring the adjusted ground image to recover the texture information covered by highly reflective areas includes:

[0020] Based on the relationship between the transition intensity of material changes and preset high and low thresholds, the power adjustment coefficient is calculated through a mapping table or proportional interpolation. Pulse width modulation is used to control the controllable light source to achieve the target power. Ground images under the target power are acquired and bilateral filtering is performed. The gradient amplitude of the filtered image is calculated to identify the texture edge pixels. The average difference between the gray values ​​on both sides along the edge normal direction is calculated as the texture edge contrast. When the texture edge contrast is higher than the preset contrast threshold, linear extrapolation is used to predict the internal gray values ​​in the high reflective area from the boundary inwards, and the weighted average is taken with the actual gray values ​​to obtain the restored texture information.

[0021] Furthermore, the step of identifying texture breakpoints based on the recovered texture information and fusing surrounding edge details to generate map detail filling information, and reconstructing the complete texture features of the highly reflective area, includes:

[0022] The recovered texture information is labeled with connected components, and texture break points are determined by the distance between adjacent texture regions and the intermediate saturation state. Edge pixel sets are extracted around the break points, and gradient direction histograms are calculated to obtain the main edge direction and edge density. Based on the similarity of the main edge direction, a transition pixel sequence is generated by linear interpolation in the break area, and gray values ​​are assigned by distance weighting to obtain map detail filling information. The saturated pixels in the high reflectivity area are replaced with map detail filling information, and the filling boundary is smoothed by Gaussian filtering to obtain the reconstructed complete texture features of the high reflectivity area.

[0023] Furthermore, the step of filling in the texture-deficient areas in the path map with the reconstructed complete texture features to generate an AGV navigation path map containing complete texture features includes:

[0024] Identify gray-saturated and texture-deficient areas in the path map, map the reconstructed complete texture features to the corresponding positions on the path map according to the original position coordinates, and fuse them with a weighted average of texture clarity to obtain a preliminary path map. Check the gray-level difference and gradient direction difference between the filled area and the surrounding original area, and use bilinear interpolation to generate transition pixels to achieve smooth boundary continuity. Store the coordinate index and gray-level value of all texture areas in raster format, and output an AGV navigation path map file containing complete texture features.

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

[0026] This invention discloses a real-time generation method for AGV visual texture navigation path maps, proposing a complete solution to the problem of texture loss caused by changes in ground material and specular reflection. This invention collects ground gloss and grayscale saturation data using a gloss detection probe, combines this with the intensity of supplementary lighting to accurately identify the distribution of highly reflective areas, and determines the coordinates of missing texture parts through boundary contour analysis. Subsequently, this invention dynamically adjusts the opening of the supplementary lighting shield and the angle of light incidence to optimize lighting conditions, collects material transition intensity, enhances texture edge contrast, and restores the original texture information obscured by reflective patches. Finally, through texture breakpoint analysis and edge detail fusion, complete texture features are reconstructed, filling in the missing areas of the path map and generating a high-precision AGV navigation path map. This invention significantly improves the navigation accuracy and stability of AGVs in complex ground environments, especially in highly reflective scenarios such as polished tiles. Attached Figure Description

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

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

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

[0030] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0031] like Figures 1-3 This embodiment of a method for real-time generation of AGV visual texture navigation path maps may specifically include:

[0032] Step S101: Collect the ground gloss and grayscale saturation area, compare the brightness difference between the grayscale saturation area and the normal area, and identify the distribution location and coverage of the high reflectivity area.

[0033] The ground is scanned by a gloss detection probe, and the ratio of reflected light intensity to incident light intensity is collected as the gloss coefficient. When the gloss coefficient exceeds a first preset threshold, it is determined to be a high-gloss surface. The percentage of pixels with saturated gray values ​​within the probe's scanning range is obtained as the saturation area. The reflectance intensity is determined by multiplying the saturation area by the gloss coefficient. The Sobel edge detection operator is used to process the ground image at the location corresponding to the gloss detection probe, and the gradient value at the boundary between the gray-saturated area and the normal area is extracted as the boundary gradient. When the boundary gradient is greater than a second preset threshold, it is marked as a strong reflective boundary. The closed contour enclosed by the strong reflective boundary is obtained through eight-neighbor connected component analysis, and the centroid position of the pixel coordinates within the closed contour is calculated as the center coordinate of the high reflective patch. Based on the offset distance between the center coordinate of the high reflective patch and the optical axis of the AGV camera, the illumination angle of the supplementary light is adjusted so that the light avoids the area corresponding to the center coordinate of the high reflective patch. The ground brightness distribution map under different illumination angles after adjustment is obtained. The standard deviation change of the original brightness distribution and the adjusted brightness distribution is compared, and the illumination angle configuration with the smallest standard deviation is selected and recorded as the illumination configuration parameter for the current material. Based on the illumination configuration parameters, ground images are re-acquired, and the closed contours are used as the boundaries of the high reflectivity areas. The pixel coordinate range of each closed contour is counted to obtain the distribution location. The length and width of the circumscribed rectangle of each closed contour are calculated as the coverage area. The distribution location and coverage area data of all high reflectivity areas are summarized to form a high reflectivity area distribution map.

[0034] Specifically, in one embodiment, the gloss detection probe employs an integrating sphere gloss sensor, which integrates a stable light source and a photodetector. When the AGV travels in a warehouse environment, the light beam emitted by the probe illuminates the surface of the ground material, and a portion of the light is reflected back to the photodetector inside the probe.

[0035] Specifically, the reflected light intensity value is collected by a photodiode array. Each diode corresponds to a different angle of reflected light. The gloss coefficient is obtained by summing the reflected light intensities at each angle and dividing by the total incident light intensity. When the warehouse floor transitions from a cement area to an epoxy resin coating area, the gloss coefficient rises sharply from around 0.3 to over 0.8. When it exceeds the first preset threshold of 0.6, it is determined to have entered a high-gloss surface area.

[0036] For example, in the stainless steel floor area of ​​a food processing workshop, due to the significant specular reflection characteristics of the surface, large areas of grayscale saturation appear in the images captured by the probe. The saturation area is calculated by scanning pixel by pixel, counting the number of pixels whose grayscale values ​​reach the upper limit of saturation, and then dividing by the total number of pixels in the probe's field of view. When the stainless steel floor is covered with cleaning liquid, the saturation area can reach more than 40% of the field of view. Multiplying the percentage of saturation area by the gloss coefficient yields the reflectivity value, which comprehensively reflects the specular reflection degree of the material and the coverage of the reflective area.

[0037] It should be noted that the Sobel edge detection operator calculates the horizontal and vertical gradients respectively using two 3×3 convolution kernels when processing ground images. The horizontal convolution kernel is [-1,0,1;-2,0,2;-1,0,1], and the vertical convolution kernel is [-1,-2,-1;0,0,0;1,2,1]. These two kernels are convolved with the image to obtain the horizontal gradient. and vertical gradient The boundary gradient value is calculated by... The gradient value at the boundary between polished ceramic tiles and anti-slip mats increases sharply due to the significant difference in reflectivity between the two materials. When the gradient value exceeds a second preset threshold, the pixel is marked as a highly reflective boundary point. Eight-neighbor connected component analysis starts from any boundary point and checks if its eight neighboring pixels are also boundary points. If so, they are included in the same connected component. This process is recursively repeated until all connected boundary points are found, forming a closed contour. The centroid coordinates are obtained by calculating the arithmetic mean of the coordinates of all pixels within the contour, representing the geometric center of the highly reflective patch.

[0038] In one possible implementation, the AGV camera's optical axis is perpendicular to the ground. When a highly reflective patch's center coordinates deviate from the optical axis, the control system calculates the offset distance and direction. For example, near a metal guide rail in a logistics sorting center, the patch's center might deviate from the optical axis by 50 centimeters. In this case, the servo motor of the supplementary lighting adjusts the illumination angle based on the offset, ensuring the main beam avoids the patch's center by 10-15 degrees. During the adjustment process, the system continuously acquires ground images from different angles and calculates the standard deviation of brightness distribution for each image. The standard deviation reflects the dispersion of image brightness; it is smaller when the light is uniformly distributed and larger when there are localized bright areas. By comparing the original image's standard deviation with the adjusted standard deviation, the system selects the angle configuration that reduces the standard deviation the most as the optimal illumination parameter for the current material.

[0039] Preferably, in the environment of a chemical plant's anti-corrosion flooring, the ground material may transition from rough acid-resistant bricks to a smooth epoxy self-leveling coating over a short distance. After re-acquiring ground images based on optimized irradiation configuration parameters, the previously identified closed contours are directly used as the precise boundaries of the high-reflectivity areas. Geometric analysis is performed on each closed contour, extracting the horizontal and vertical coordinates of all pixels on the contour, identifying the maximum and minimum coordinate values, and determining the circumscribed rectangle of the contour. The length and width of the circumscribed rectangle represent the coverage parameters of the high-reflectivity area, expressed in pixels or converted actual distance.

[0040] Understandably, when the AGV needs to construct a navigation map of the entire work area, the system continuously records information on all detected highly reflective areas. The data for each area includes parameters such as center coordinates, boundary contour coordinates, and the size of the circumscribed rectangle. This data is indexed according to the AGV's travel path and timestamps, forming a spatiotemporally correlated database of highly reflective area distribution.

[0041] In one embodiment, the distribution map is generated using a rasterization method, dividing the AGV operating area into 10cm × 10cm grid cells. Each grid cell records the presence of highly reflective areas and the reflectivity level of those areas. In this way, the system can generate a two-dimensional heatmap of highly reflective area distribution, visually displaying the spatial distribution characteristics of reflective areas throughout the operating environment. This distribution map serves as an important reference for AGV path planning, enabling the navigation system to anticipate road conditions ahead and adjust camera exposure parameters and image processing strategies in advance, ensuring stable extraction of ground texture features even in areas of material variation.

[0042] Step S102: Based on the distribution location and coverage of the highly reflective areas, extract the texture missing parts covered by the specular reflection patches and determine the boundary coordinates of the texture missing parts.

[0043] Based on the coordinates of the distribution of highly reflective areas, the grayscale distribution of the re-acquired ground image at the corresponding locations is read. A set of pixels with a grayscale value of 255 (i.e., saturated) is extracted. Morphological dilation is used to expand the boundary of the saturated region. The dilation kernel size is set to the integer value obtained by dividing the mean grayscale gradient of adjacent unsaturated pixels by a preset coefficient, resulting in the expanded specular reflection coverage area. Canny edge detection is performed on the expanded specular reflection coverage area to extract the coordinate sequence of edge pixels. The edge curve equation is fitted using the least squares method, and the curvature value of each point on the curve is calculated. When the difference in curvature values ​​between adjacent points exceeds a preset threshold, that point is marked as a texture missing boundary feature point. Connecting all feature points forms the texture missing region contour. Using the pixel grayscale distribution within a preset width range on both sides of the texture missing region contour, the gradient direction histogram is used to obtain the inner and outer principal direction angle values. If the difference between the two angle values ​​exceeds a preset angle threshold, a texture discontinuity is determined. The contour position is finely adjusted along the contour normal direction to minimize the difference in the inner and outer principal direction angles, resulting in the adjusted texture missing boundary. By performing polygon approximation on the adjusted texture missing boundary, the boundary vertex coordinate sequence is obtained. The vertices are arranged in clockwise order, and the Euclidean distance between adjacent vertices is calculated as the boundary line segment length. All vertex coordinates and line segment length data are summarized to determine the boundary coordinate set of the texture missing part.

[0044] Specifically, the core of morphological dilation lies in the adaptive determination of the dilation kernel size. First, sampling points are selected at the boundaries of the saturated region, and the gray-level gradient of the surrounding unsaturated pixels is calculated for each sampling point. The Sobel operator is used for gradient calculation, obtaining the gradient components in the horizontal and vertical directions, and then the gradient magnitude is calculated. The gradient magnitudes of all sampling points are summed and divided by the number of sampling points to obtain the gradient mean. The dilation kernel size is equal to the integer part of this mean divided by a preset coefficient. When the AGV operates in the polished stainless steel floor area of ​​the warehouse, due to strong specular reflection, the gradient value at the edge of the saturated region is usually very large, reaching over 200. If the preset coefficient is 20, the dilation kernel size is 10×10 pixels. The Canny edge detection algorithm includes multiple processing stages. First, Gaussian filtering is applied to the expanded specular reflection coverage area to reduce noise interference. Then, the magnitude and direction of the image gradient are calculated, and non-maximum suppression is used to retain the points with the largest local gradients. Next, dual threshold detection is used; points above the high threshold are identified as strong edges, and points between the high and low thresholds are identified as weak edges. Edge connectivity preserves weak edges connected to strong edges, while suppressing other weak edges. The extracted edge pixels are arranged into a coordinate sequence based on spatial continuity. During least-squares fitting, the edge point coordinates are substituted into a polynomial equation, and the equation coefficients are solved by minimizing the sum of squared errors. For quadratic curves... The curvature calculation formula is: , where x is the x-coordinate of a point on the curve.

[0045] For example, when an AGV passes through the junction of epoxy resin flooring and ceramic tile flooring, the edge curve will show a significant change in curvature due to the difference in reflective properties between the two materials. The curvature value is calculated point by point along the curve. When the curvature difference between two adjacent points exceeds 0.1, the point is marked as a texture loss boundary feature point. These feature points often correspond to the location of material abrupt change or texture break.

[0046] Preferably, the statistical process of the gradient direction histogram involves direction analysis of local regions. Five-pixel-wide strip regions are selected on both the inner and outer sides of the texture-deficient region's outline, and the gradient direction of each pixel within each of these two regions is calculated. The gradient direction is determined using the arctangent function. Calculate the angle values ​​within the range of -π to π, and convert them to 0 to 360 degrees. Divide the 360 ​​degrees into 36 intervals, each interval being 10 degrees, and count the number of pixels falling into each interval. The angle corresponding to the highest peak of the histogram is the principal direction angle value.

[0047] In one possible implementation, the fine-tuning of the contour position employs an iterative optimization method. The contour point is moved within a range of ±3 pixels along the contour normal, moving 1 pixel at a time. For each new position, the difference in the principal direction angles between the inner and outer sides is recalculated. By comparing the angle differences at different positions, the position with the smallest difference is selected as the optimized position for that contour point. This local optimization allows the adjusted contour to more accurately reflect the true boundaries of texture defects.

[0048] Specifically, polygon approximation employs The algorithm first connects the start and end points of the contour and finds the point on the contour farthest from the straight line. If this distance is greater than a preset tolerance, the point is taken as a new vertex, and the two sub-contours are recursively processed. If the distance is less than the tolerance, the contour segment is replaced with a straight line segment. Through this recursive segmentation, the smooth contour curve is approximated as a polygon composed of several line segments. Furthermore, vertex sorting ensures the consistency of the boundary description. The centroid coordinates of the polygon are calculated, and then the polar angle of each vertex relative to the centroid is calculated. The vertices are sorted in ascending order of polar angle to obtain a clockwise sequence. The Euclidean distance between adjacent vertices is calculated by taking the square root of the sum of the squares of their coordinate differences.

[0049] Understandably, determining the set of boundary coordinates provides a precise spatial range for subsequent texture restoration. Each boundary segment not only records the endpoint coordinates but also includes geometric attributes such as segment length and orientation angle. In the metal shelving area of ​​the logistics center, this boundary information helps to accurately identify the range of shelving markings and guide lines obscured by mirror reflections.

[0050] For example, when dealing with a complex-shaped specular reflection region, polygon approximation may generate 20 to 30 vertices. These vertex coordinates and corresponding line segment information are then organized into structured data.

[0051] Step S103: Analyze the coverage area of ​​the texture missing part defined by the boundary coordinates, adjust the opening size of the adjustable light shield of the fill light to change the light illumination range, and obtain the optimized light incident angle distribution based on the relative position of the light shield opening and the texture missing area.

[0052] The polygon area of ​​the texture-deficient region is calculated using a set of boundary coordinates. The centroid coordinates and circumcircle radius of this region are obtained. The diameter of the basic opening of the light shield is determined based on the ratio of the circumcircle radius to the distance from the supplementary light to the ground. If the ratio is greater than a preset threshold, the opening is set to ellipse; otherwise, it is set to circle, thus obtaining the initial opening parameters of the light shield. The servo motor of the light shield blades is controlled according to the initial opening parameters to adjust the blade opening angle and change the opening area. The ground spot contours under different opening areas are collected, and the number of overlapping pixels between the edge of the spot and the boundary of the texture-deficient region is extracted. When the number of overlapping pixels is less than a preset threshold, the current blade angle is recorded as the occlusion configuration parameter. Based on the spot distribution under the occlusion configuration parameters, the angle between the line connecting each point in the spot to the supplementary light and the ground normal is calculated as the incident angle of each point. The number of pixels in different incident angle intervals is counted to generate an incident angle distribution histogram. The angle corresponding to the peak value of the histogram is extracted as the principal incident angle. By rotating the azimuth angle of the light shield to change the spatial distribution of the main incident angle, the light intensity gradient vector field in the texture-deficient area after rotation is obtained. If the angle between the gradient vector direction and the texture extension direction is less than a preset angle threshold, the current azimuth angle and the main incident angle are saved as the optimized light incident angle distribution.

[0053] Specifically, in one implementation, the polygon area of ​​the texture-deficient region is calculated using a shoelace formula.

[0054] Specifically, the area of ​​the polygon is obtained by arranging the boundary coordinates in order and then summing the cross product of the coordinates of adjacent vertices. The centroid coordinates are obtained by weighted averaging of the coordinates of all vertices, with the weight being the area of ​​the triangle corresponding to each vertex. The circumcircle radius is obtained by calculating the maximum distance from the centroid to each vertex. When the AGV operates in the polished metal floor area of ​​the warehouse, the areas lacking texture are often irregularly shaped, and the circumcircle radius can reach more than 30 centimeters. The mechanical structure of the light shield adopts a blade design similar to the aperture of a camera, containing 6 or 8 rotatable metal blades. Each blade is controlled by an independent stepper motor, and the motor drives the gear to mesh with the rack at the root of the blade, achieving precise adjustment of the blade opening angle. When the ratio of the circumcircle radius to the distance from the supplementary light to the ground exceeds 0.5, the system determines that an elliptical opening is required. The elliptical opening is achieved by differentially controlling the opening angle of each blade, with two sets of blades opening at different angles, forming the major axis and minor axis. The circular opening is achieved by synchronously controlling all blades to open at the same angle. The initial opening parameters include the opening shape type, major axis or diameter, minor axis, and the offset of the opening center relative to the light beam axis.

[0055] For example, the servo motors of the light-shielding blades are controlled according to the initial opening parameters. After receiving the pulse signal, the servo motors drive the blades to rotate according to a preset acceleration curve. During the blade opening process, ground spot images are acquired in real time, and the spot contours are extracted through binarization processing. The overlap detection between the spot edge and the boundary of the texture-deficient area is achieved through pixel-by-pixel comparison, and the number of pixels jointly covered by the two contours is counted. When the AGV passes over the highly reflective epoxy resin floor, the blade opening angle is gradually reduced, causing the spot to gradually shrink until the number of overlapping pixels drops to less than 10% of the total boundary pixels. At this point, the current angle value of each blade is recorded as the shading configuration parameter.

[0056] Preferably, after determining the light spot distribution based on the occlusion configuration parameters, the incident angle distribution is calculated. Calculating the incident angle requires determining the spatial coordinates of the supplementary light and the coordinates of each pixel on the ground. The supplementary light is typically installed at the front of the AGV body at a fixed height above the ground. For each pixel within the light spot, the spatial vector from that point to the supplementary light is calculated, and then the angle between this vector and the ground normal vector is calculated. On a flat warehouse floor, the normal vector is vertically upward; in a sloping loading / unloading area, the normal vector will be inclined accordingly. The incident angle distribution histogram divides the angle range from 0 to 90 degrees into several intervals, and the number of pixels falling into each interval is counted. The peak value of the histogram reflects the main incident direction of the light, typically occurring between 30 and 60 degrees, an angle that provides sufficient illumination while reducing specular reflection.

[0057] In one possible implementation, the azimuth angle of the light shield is adjusted by a rotary motor on the base. The rotary motor drives the entire light shield assembly to rotate around a vertical axis, changing the orientation of the elliptical or irregularly shaped opening. Every 5 degrees of rotation, an image of the texture-deficient region is captured, and the light intensity gradient of the image is calculated. The light intensity gradient is obtained by the difference between the gray values ​​of adjacent pixels, forming a two-dimensional vector field. The gradient vector at each location points in the direction of the fastest increase in light intensity.

[0058] Understandably, the texture extension direction is obtained by analyzing the intact texture around the texture-deficient area. Texture samples are extracted outside the boundary of the deficient area, and the main direction of the texture is analyzed using Fourier transform. When the angle between the light intensity gradient vector and the main direction of the texture is less than 30 degrees, it indicates that the lighting direction is basically consistent with the texture direction, which is beneficial for highlighting texture features. Furthermore, different azimuth angle configurations are traversed, and the average angle between the gradient vector and the texture direction is calculated for each configuration. When the AGV operates in the metal rack area of ​​the logistics center, the regular arrangement of the racks forms a clear texture direction. The system adjusts the orientation of the light shield to make the lighting direction form an appropriate angle with the rack arrangement direction, enhancing the contrast of the rack edges.

[0059] Specifically, the optimized light incident angle distribution includes multiple parameters: the value of the primary incident angle, the azimuth angle corresponding to the primary incident angle, the secondary incident angle and its distribution range, and the light intensity weighting coefficient for each angle interval. These parameters together describe the three-dimensional light field distribution characteristics after adjustment by the light shield.

[0060] Step S104: Based on the optimized light incident angle distribution, collect the grayscale jump amplitude when the AGV drives from the rough cement floor into the polished tile area, and evaluate the transition intensity of the material change.

[0061] Based on the optimized light incident angle distribution, the supplementary lighting parameters are configured. During AGV movement, a series of ground images are continuously acquired. The average grayscale value of each frame is extracted. When the difference between the average grayscale values ​​of two adjacent frames exceeds a preset threshold, this position is marked as the material transition start point. Starting from the material transition start point, grayscale histograms of the ground images are extracted at fixed intervals in the continuous acquisition sequence. The peak position of the histogram is calculated, and the difference between the current peak position and the previous peak position is used as the grayscale distribution offset value. When the offset values ​​of three consecutive frames exceed a preset offset threshold, it is marked as the transition end point. The absolute values ​​of all offset values ​​between the start and end points are accumulated to obtain the grayscale jump amplitude. Based on the grayscale jump amplitude, the number of image frames acquired from the material transition start point to the transition end point is counted. The number of frames is multiplied by the AGV's moving speed and the sampling time interval to obtain the transition region length. The ratio of the jump amplitude to the transition region length is calculated as the transition gradient. When the transition gradient exceeds a preset gradient threshold, it is determined to be a rapid transition; otherwise, it is determined to be a gradual transition. The transition gradient value is used as the transition intensity evaluation value for material change.

[0062] Specifically, in one implementation, the AGV is equipped with a camera that captures ground images at a rate of 30 frames per second. As the AGV moves from a concrete floor area to a polished tile area in the warehouse, the average grayscale value of the images changes significantly. Due to its rough surface, the concrete floor exhibits predominantly diffuse reflection, resulting in an average grayscale value typically between 80 and 120; while the polished tile surface is smooth with strong specular reflection, leading to an average grayscale value exceeding 180. The average grayscale value of each frame is calculated using a sliding window method. When the grayscale difference between adjacent frames exceeds a preset threshold of 30, it is determined that the area has entered a material transition region.

[0063] It's important to note that the grayscale histogram reflects the pixel distribution of each gray level in an image. In areas with uniform material, the histogram typically exhibits a single-peak distribution, with the peak position corresponding to the dominant grayscale value of the material. When the AGV moves from a rough surface to a smooth surface, the peak of the histogram shifts towards higher grayscale levels. An image is captured every 0.1 seconds, and its 256-level grayscale histogram is calculated. The grayscale level with the most pixels is identified as the peak position. By comparing the difference between the current peak and the initial peak, the degree of change in the ground's reflective properties is quantified.

[0064] For example, in logistics sorting, AGVs need to frequently traverse ground areas with different materials. The peak position of the material transition starting point is recorded as 85. As the AGV moves forward, the peak position of the continuously acquired images gradually increases: 90, 98, 115, 142, and 165. When the offset value of three consecutive frames exceeds 20, the system determines that the transition process has ended. All offset values ​​are summed: 5 + 8 + 17 + 27 + 23 = 80, yielding the total grayscale transition amplitude.

[0065] Preferably, the calculation of the transition region length takes into account the actual motion parameters of the AGV. Assuming the AGV moves at a constant speed of 0.5 m / s, with a sampling interval of 0.1 seconds, and a total of 10 frames are collected from the starting point to the ending point, the transition region length is 10 × 0.5 × 0.1 = 0.5 meters. The transition gradient is calculated as 80 / 0.5 = 160. When this value exceeds the preset gradient threshold of 100, it is considered a sharp transition, indicating that the ground material has changed significantly within a short distance, requiring rapid adjustment of image processing parameters to adapt to the new ground conditions.

[0066] Step S105: Adjust the illumination intensity of the fill light according to the transition intensity of the material change, acquire the adjusted ground image, extract the grayscale difference between the texture edge and the background as the enhanced texture edge contrast, and restore the original texture information covered by the specular reflection patch.

[0067] Based on the transition intensity values ​​of material changes, a mapping table between transition intensity and supplementary lighting power is established. If the transition intensity is greater than a preset high threshold, the supplementary lighting power is reduced to a preset proportion of the original power; if the transition intensity is less than a preset low threshold, the original power is maintained. Otherwise, the corresponding power adjustment coefficient is calculated based on the proportion of transition intensity between the high and low thresholds. The supplementary lighting is controlled by pulse width modulation signals to achieve the adjusted target power. The ground is illuminated using the target power, and the adjusted ground image is acquired. The image undergoes bilateral filtering to preserve edge information while smoothing noise. The horizontal and vertical gradients of each pixel in the filtered image are calculated. Texture edge pixels are identified by gradient magnitudes greater than a preset edge threshold. For each texture edge pixel, its gradient direction is calculated as the edge normal direction. A preset number of pixels are sampled along this normal direction and in its opposite direction. The grayscale mean of the pixels on both sides is calculated. The absolute value of the difference between the two means is used as the local contrast of the edge point. The average value of the local contrasts of all edge points is used as the texture edge contrast value. When the texture edge contrast value is higher than the preset contrast threshold, in the specular reflection patch area, scan line by line from the patch boundary inward. Utilize the texture grayscale change pattern of the same row outside the boundary, predict the grayscale value of the corresponding position inside the patch through linear extrapolation, and then weight the predicted grayscale value and the actual collected grayscale value at that position according to the preset weight coefficient to obtain the restored texture information.

[0068] Specifically, in one implementation, the mapping relationship between transition intensity and fill light power is achieved through a piecewise linear function.

[0069] Specifically, three power adjustment ranges are preset. When the transition intensity value is less than 50, the rated power of the supplementary light remains unchanged; when the transition intensity value is greater than 150, the power is reduced to 40% of the rated power; when the transition intensity value is between 50 and 150, the power adjustment coefficient is calculated according to a linear relationship. Assuming the rated power is 10 watts and the transition intensity value is 100, the power adjustment coefficient is 0.7, and the adjusted target power is 7 watts. The pulse width modulation signal achieves power adjustment by changing the duty cycle, which is directly proportional to the power. Bilateral filtering is an edge-preserving smoothing filtering algorithm that considers the similarity of the spatial domain and the gray-level domain. When processing highly reflective ground images, bilateral filtering can remove noise while maintaining the sharpness of texture edges. During the filtering process, for each pixel, the algorithm finds pixels with similar spatial distance and gray-level values ​​in its neighborhood, and calculates the filtered gray-level value of that point by weighted averaging. The spatial weights use a Gaussian function, and the gray-level weights also use a Gaussian function; the two weights are multiplied to obtain the final weight. This dual-weighting mechanism ensures that pixels on both sides of the edge do not interfere with each other, thus preserving edge features. When the AGV passes over a stainless steel floor, bilateral filtering can effectively remove noise caused by reflections while retaining important texture features such as floor seams and scratches.

[0070] For example, gradient calculation is implemented using the Sobel or Prewitt operator. For each pixel in the filtered image, its gray-level change rate in the horizontal and vertical directions is calculated separately. The horizontal gradient Gx reflects the change in pixel gray-level in the horizontal direction, and the vertical gradient Gy reflects the change in the vertical direction. The gradient magnitude is obtained by calculating the square root of the sum of the squares of Gx and Gy, and the gradient direction is obtained by calculating the arctangent of Gy and Gx. When the gradient magnitude exceeds a preset threshold, the pixel is identified as an edge point.

[0071] Preferably, the edge normal direction is determined based on the gradient direction. The gradient direction points in the direction of the fastest increase in grayscale, while the tangent direction of the edge is perpendicular to the gradient direction, and the normal direction is either the same as or opposite to the gradient direction. When sampling along the normal direction, starting from the edge point, extend 5 pixels in both the positive and negative gradient directions, and sample the pixel grayscale values ​​in these two directions respectively.

[0072] In one possible implementation, the calculation of local contrast considers the grayscale distribution characteristics on both sides of the edge. For each edge point, five pixels are sampled in the positive direction of its normal, and the average grayscale value of these five points is calculated as the bright side mean; five points are also sampled in the negative direction of the normal, and the dark side mean is calculated. The absolute value of the difference between the two means is the local contrast of that edge point. When AGVs operate on the epoxy resin floor of a warehouse, the edges of the markings on the floor usually have high local contrast, with the contrast value between yellow markings and dark gray ground reaching over 100.

[0073] Understandably, texture edge contrast value is a statistical characteristic of the local contrast of all edge points. The arithmetic mean of the local contrast of all edge points is calculated as the texture edge contrast index of the entire image. This index reflects the recognizability of texture features under the current lighting conditions. Furthermore, the texture information restoration process employs an interpolation method based on boundary extrapolation. When the texture edge contrast value is higher than the threshold of 60, it indicates that the texture features around the patch are clearly discernible and suitable for texture restoration. Starting from the boundary of the specular reflection patch, the process proceeds line by line into the patch. For each line within the patch, the texture grayscale variation pattern outside the patch is first analyzed. By calculating the grayscale difference between consecutive pixels on the outer side, the periodic or gradient features of the texture are extracted. Then, based on these features, the grayscale value of the corresponding position inside the patch is predicted through linear extrapolation.

[0074] Specifically, the weighted fusion process comprehensively considers both predicted and actual acquired values. Although specular reflection causes grayscale saturation, some locations still retain some original texture information. Weighting coefficients are determined based on the pixel's saturation level: fully saturated pixels are assigned a higher weight of 0.9 for the predicted value, partially saturated pixels have a weight of 0.6, and unsaturated pixels retain their original values. Through weighted averaging, the recovered texture grayscale values ​​are obtained, achieving texture reconstruction of the areas obscured by specular reflection.

[0075] Step S106: Based on the enhanced texture edge contrast and the restored original texture information, evaluate the number and location of texture breakpoints in the currently acquired ground image, fuse the edge details around the texture breakpoints to obtain map detail filling information, and reconstruct the complete texture features of the area covered by the specular reflection patch.

[0076] Based on the enhanced texture edge contrast value and the restored original texture information, connected component labeling is performed on the ground image, and the distance between adjacent texture regions is calculated. If the distance exceeds a preset breakage threshold and the grayscale value of the middle region is saturated, the location is determined to be a texture breakage point. The pixel coordinates of all breakage points are counted to obtain the number of breakage points. For the coordinate location of the texture breakage point, an edge pixel set within a preset radius is extracted around each breakage point, and the gradient direction histogram of the edge pixels is calculated. The direction with the highest frequency in the histogram is taken as the main edge direction of the region, and the angle value of the main edge direction and the edge pixel density are recorded as the edge feature parameters of the breakage point. Based on the edge feature parameters, edge segments with similar direction angles on both sides of the breakage point are identified. A transition pixel sequence connecting the two sides of the edge is generated in the breakage region through linear interpolation. The grayscale value of the transition pixel is weighted according to its distance to the two sides of the edge to obtain the set of filling grayscale values ​​of the breakage region as map detail filling information. The saturated pixel values ​​in the specular reflection patch area are replaced by the map detail filling information. Gaussian filtering is used to smooth the boundary between the filled area and the surrounding original texture to eliminate grayscale abrupt changes and reconstruct the complete texture features of the area covered by the specular reflection patch.

[0077] Specifically, a four-connected or eight-connected labeling algorithm is used to label connected components. The ground image is scanned line by line. When an unlabeled texture pixel is encountered, a new label value is assigned, and all connected pixels of the same type are recursively labeled. After labeling, each connected component represents an independent texture region. The distance between adjacent texture regions is obtained by calculating the minimum Euclidean distance between the boundary pixels of the two connected components. When the AGV moves in the warehouse, texture features such as guide lines, markings, and seams on the ground will form different connected components. If the minimum distance between two connected components exceeds 20 pixels, and more than 90% of the pixels in the middle region reach saturation, a texture break is identified. After identifying the break region, samples are taken at equal intervals along the central axis of the break region; each sample point is a texture break point. For each break point, the system records its two-dimensional coordinate position and counts the total number of break points in the entire image. On the metal floor of the logistics center, strong specular reflection often causes multiple texture breaks, with the number of break points reaching dozens.

[0078] For example, the extraction process of edge feature parameters involves gradient analysis of local regions. A circular region with a radius of 15 pixels is defined around each breakpoint, and all edge pixels within this region are extracted. Edge pixels are determined by a gradient magnitude threshold, and the gradient direction is obtained by calculating the arctangent of the horizontal and vertical gradients. The directional range from 0 to 180 degrees is divided into 36 intervals, each with a width of 5 degrees. The number of edge pixels falling into each interval is counted, and a gradient direction histogram is constructed. The direction corresponding to the peak of the histogram is the main edge direction of the local region. Simultaneously, the edge pixel density is obtained by calculating the ratio of the number of edge pixels to the total number of pixels in the region, reflecting the richness of the texture in the region.

[0079] Preferably, the identification of edge segments on both sides of the break point is based on the directional similarity criterion. Edge segments with similar main edge directions are searched on both sides of the break point, and edge segments with directional differences within 15 degrees are considered to be likely to belong to the continuation of the same texture feature. After finding the corresponding edge segments, the endpoint coordinates of the two edges are extracted as control points for interpolation.

[0080] In one possible implementation, the linear interpolation process is dynamically adjusted based on the width of the fracture region. For narrow fractures less than 10 pixels wide, linear interpolation is performed by directly connecting the two endpoints; for wider fracture regions, multiple interpolation paths are generated at equal intervals within the fracture region, with each path connecting the corresponding points on both sides. The grayscale value calculation of the interpolation points takes distance weighting into account: let the distance from a certain interpolation point to the left edge be... The distance to the right edge is The average gray level of the left edge is The average gray level of the right edge is The interpolated gray value at that point is... This inverse distance weighting ensures a smooth transition of grayscale values.

[0081] Specifically, map detail fill information is a collection of all interpolated points and their grayscale values. This fill information is organized into a two-dimensional array according to spatial location, with each element containing pixel coordinates and the corresponding fill grayscale value. When an AGV passes over a polished tile floor, specular reflections may obscure the texture of the tile seams; the fill information can restore the continuity of these seams. Furthermore, Gaussian filtering is applied primarily at the boundary between the filled area and the original texture. A 5×5 Gaussian kernel is used to convolve the boundary region, with the standard deviation of the Gaussian kernel set to 1.5. During the filtering process, the filled pixels and the original pixels are weighted and averaged according to Gaussian weights, eliminating visual abrupt changes caused by differences in grayscale values.

[0082] Understandably, reconstructing complete texture features involves not only restoring grayscale values ​​but also restoring the continuity of the texture structure. By analyzing the texture patterns around a patch, the expected texture direction and density within the patch can be inferred. For regular textures such as tile seams, the surrounding geometric patterns are continued; for random textures such as frosted surfaces, the texture features of similar surrounding areas are copied. The reconstructed texture features enable the AGV to continuously track ground features in highly reflective environments, ensuring navigation stability.

[0083] Step S107: Using the reconstructed complete texture features, fill in the missing texture areas in the path map, integrate the path map texture information, and generate a complete AGV navigation path map.

[0084] Using the reconstructed complete texture features, regions in the path map that are saturated with grayscale values ​​and lack texture features are identified. The start and end coordinates of these regions are recorded. The reconstructed texture features are mapped to their corresponding positions on the path map according to the original acquisition coordinates. If the mapped positions overlap, a weighted average fusion is performed based on different weights according to texture clarity to obtain a preliminary filled path map. For the preliminary filled path map, the boundary continuity between the filled area and the surrounding original area is checked. The grayscale difference and gradient direction difference between adjacent pixels on both sides of the boundary are calculated. When the difference exceeds a preset threshold, a bilinear interpolation method is used to generate transition pixels at the boundary to determine the smoothed texture connection state. Based on the texture connection state, the spatial location information and grayscale feature information of all texture regions are integrated. The texture data is stored in a raster format, with each raster cell containing a coordinate index and the corresponding texture grayscale value. The output is an AGV navigation path map file containing complete texture features.

[0085] Specifically, the identification of texture-deficient regions is achieved by scanning every pixel of the path map. All pixels in the map are traversed, and when a continuous saturated pixel region is detected with a texture feature value of zero within that region, it is marked as a missing region. Each missing region is represented by a rectangle, and the coordinates of its top-left and bottom-right corners are recorded. The mapping of reconstructed texture features is based on coordinate correspondence, placing each pixel in the reconstructed data at its original acquisition coordinates to its corresponding position on the path map. The weight allocation for weighted average fusion is based on texture sharpness scores. Sharpness is obtained by calculating the standard deviation of the gradient magnitude of a local region; a larger standard deviation indicates richer texture details, and a higher weight is assigned. When two or more reconstructed textures overlap at the same location, they are assigned corresponding preset weights according to their respective sharpness scores, and the sum of the weights is normalized to 1. The fused grayscale value is equal to the sum of the products of the grayscale values ​​of each reconstructed texture and their corresponding weights.

[0086] For example, the boundary continuity check uses a 3×3 sliding window that moves along the boundary of the filling region. When the window contains both the filled pixels and the original pixels, the difference in grayscale mean and gradient direction between the two types of pixels is calculated. The difference in grayscale mean reflects the degree of abrupt change in brightness, and the difference in gradient direction reflects the change in texture direction. When the difference exceeds a preset threshold, it indicates the existence of a discontinuity.

[0087] Preferably, bilinear interpolation takes two control points on each side of the boundary, and weights the values ​​based on the distances from the point to be interpolated to the four control points. Let the gray values ​​of the four control points be... , , , The normalized distance from the interpolation points to them is , , , The interpolated grayscale value is The weighted average method allows for a smooth transition of gray values ​​at the boundaries.

[0088] Specifically, rasterized storage divides the entire path map into regular grid cells, each cell being 5×5 pixels in size. Each grid cell stores the average texture grayscale value and texture type identifier for that area. The map file uses a binary format, with the file header containing map dimensions, grid size, and coordinate system information, and the file body storing the grid data in row and column order. This format facilitates fast reading and positioning by the AGV.

[0089] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

Claims

1. A method for real-time generation of AGV visual texture navigation path maps, characterized in that, include: Obtain the location and coverage of highly reflective areas in ground images; The boundary coordinates of the texture-deficient parts are determined based on the distribution location and coverage of the highly reflective areas; The optimized light incident angle distribution is obtained by adjusting the illumination parameters of the controllable light source based on the boundary coordinates of the texture missing parts. Ground image sequences were acquired based on the optimized light incident angle distribution, and the transition intensity of material changes was evaluated. The intensity of the controllable light source is adjusted according to the transition intensity of the material change, and the adjusted ground image is acquired to restore the texture information covered by the highly reflective area. Based on the recovered texture information, identify texture breakpoints and fuse surrounding edge details to generate map detail filling information, reconstructing the complete texture features of highly reflective areas; The reconstructed complete texture features are used to fill in the texture-missing areas in the path map, generating an AGV navigation path map containing complete texture features; The step of determining the boundary coordinates of the texture-deficient portion based on the distribution location and coverage of the highly reflective area includes: Based on the distribution location of highly reflective areas, the original grayscale values ​​are read to extract the set of pixels exceeding the saturation threshold. Morphological dilation is used to expand the saturation region to obtain the specular reflection coverage area. Edge curves are fitted by edge detection and least squares method, and texture missing boundary feature points are marked according to curvature differences. The angle difference of the main direction of the gradient direction histogram inside and outside the boundary is calculated, and the contour is finely adjusted along the normal direction to minimize the angle difference, thus obtaining the adjusted texture missing boundary. Polygon approximation is performed on the adjusted texture missing boundary to obtain the clockwise vertex coordinate sequence and the Euclidean distance between adjacent vertices as the boundary coordinate set of the texture missing part.

2. The method as described in claim 1, characterized in that, The acquisition of the distribution location and coverage area of ​​highly reflective areas in the ground image includes: The ratio of reflected light intensity to incident light intensity is collected as the gloss coefficient and combined with the gray-scale saturation area to determine the reflective intensity. Edge detection and connected component analysis are used to extract the closed contour enclosed by the strong reflective boundary. The centroid of the closed contour is calculated as the center coordinate of the highly reflective patch. The controllable light source illumination angle is adjusted according to the center coordinate of the highly reflective patch and the offset distance of the camera optical axis. The standard deviation of the brightness distribution before and after adjustment is compared. The illumination angle with the smallest standard deviation is selected as the current illumination configuration parameter. Based on the illumination configuration parameter, the image is re-acquired and the pixel coordinate range of the closed contour and the size of the circumscribed rectangle are statistically analyzed to obtain the distribution location and coverage of the highly reflective area.

3. The method as described in claim 1, characterized in that, The step of adjusting the illumination parameters of the controllable light source based on the boundary coordinates of the texture-deficient portion to obtain the optimized light incident angle distribution includes: The initial opening parameters of the light shield are determined by calculating the polygon area, centroid, and circumcircle radius of the texture-deficient region based on the boundary coordinate set, and the blade servo motor is controlled to adjust the opening area. The light spot contours under different opening areas are collected, and the occlusion configuration parameters are determined by minimizing the number of overlapping pixels between the light spot edge and the boundary of the texture-deficient region. Based on the occlusion configuration parameters, the incident angle of each point in the light spot is calculated to generate an incident angle distribution histogram, and the peak value is extracted as the main incident angle. The main incident angle distribution is changed by rotating the azimuth angle of the light shield, and the optimized light incident angle distribution is determined by minimizing the angle between the light intensity gradient vector direction and the texture extension direction in the texture-deficient region.

4. The method as described in claim 1, characterized in that, The process of acquiring ground image sequences based on the optimized light incident angle distribution and evaluating the transition intensity of material changes includes: Based on the optimized light incident angle distribution, a controllable light source is configured and ground image sequences are continuously acquired. When the average gray value difference between adjacent frames exceeds a preset threshold, the material transition start point is marked. Gray-level histograms are acquired at fixed intervals and the peak position offset value is calculated. When the offset value of multiple consecutive frames exceeds a preset offset threshold, the transition end point is marked. The gray-level jump amplitude is obtained by accumulating the absolute value of the offset value between the start point and the end point. The ratio of the gray-level jump amplitude to the length of the transition region is calculated as the transition gradient. The transition gradient value is used as the transition intensity evaluation value of the material change.

5. The method as described in claim 1, characterized in that, The process of adjusting the controllable light source intensity based on the transition intensity of material changes and acquiring the adjusted ground image to restore the texture information obscured by highly reflective areas includes: Based on the relationship between the transition intensity of material changes and preset high and low thresholds, the power adjustment coefficient is calculated through a mapping table or proportional interpolation. Pulse width modulation is used to control the controllable light source to achieve the target power. Ground images under the target power are acquired and bilateral filtering is performed. The gradient amplitude of the filtered image is calculated to identify the texture edge pixels. The average difference between the gray values ​​on both sides along the edge normal direction is calculated as the texture edge contrast. When the texture edge contrast is higher than the preset contrast threshold, linear extrapolation is used to predict the internal gray values ​​in the high reflective area from the boundary inwards, and the weighted average is taken with the actual gray values ​​to obtain the restored texture information.

6. The method as described in claim 1, characterized in that, The process of identifying texture breakpoints based on the recovered texture information and fusing surrounding edge details to generate map detail filling information, reconstructing the complete texture features of highly reflective areas, includes: The recovered texture information is labeled with connected components, and texture break points are determined by the distance between adjacent texture regions and the intermediate saturation state. Edge pixel sets are extracted around the break points, and gradient direction histograms are calculated to obtain the main edge direction and edge density. Based on the similarity of the main edge direction, a transition pixel sequence is generated by linear interpolation in the break area, and gray values ​​are assigned by distance weighting to obtain map detail filling information. The saturated pixels in the high reflectivity area are replaced with map detail filling information, and the filling boundary is smoothed by Gaussian filtering to obtain the reconstructed complete texture features of the high reflectivity area.

7. The method as described in claim 1, characterized in that, The process of filling in missing texture areas in the path map using reconstructed complete texture features to generate an AGV navigation path map containing complete texture features includes: Identify gray-saturated and texture-deficient areas in the path map, map the reconstructed complete texture features to the corresponding positions on the path map according to the original position coordinates, and fuse them with a weighted average of texture clarity to obtain a preliminary path map. Check the gray-level difference and gradient direction difference between the filled area and the surrounding original area, and use bilinear interpolation to generate transition pixels to achieve smooth boundary continuity. Store the coordinate index and gray-level value of all texture areas in raster format, and output an AGV navigation path map file containing complete texture features.

Images