A trackless navigation texture information redundancy analysis method

Abstract

The application provides a trackless navigation texture information redundancy analysis method, comprising: processing original image data of a current wear area by using an edge detection algorithm to obtain wear boundary contour coordinates and gray distribution data; extracting boundary coordinate point sets from a reliable anchor point area and determining an optimized update frequency value according to the position and shape matching degree of the boundary coordinate point sets and historical path map data; incrementally updating the path map by using the optimized update frequency value, integrating the gray distribution data of the reliable anchor point area and the texture detail loss degree value, and generating an enhanced positioning reference data set; updating the enhanced positioning reference data set according to supplementary data and performing periodic matching degree evaluation according to the optimized update frequency value to maintain long-term positioning consistency.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a method for analyzing the redundancy of texture information in trackless navigation. Background Technology

[0002] In the field of modern intelligent navigation, especially in the research and application of trackless navigation technology, ensuring the stability and reliability of long-term positioning is a crucial issue. This technology is widely used in industrial warehousing, logistics transportation, and service robots, directly affecting equipment operating efficiency and safety. Particularly in complex environments, navigation systems need to continuously adapt to changes in ground conditions to maintain the accuracy of path planning, a requirement that places extremely high demands on technological innovation. However, many current navigation solutions often struggle to effectively cope with the dynamic evolution of the ground environment, especially when the ground wears down over time. Existing methods focus more on extracting and matching initial environmental features, neglecting the significant changes in ground features due to wear over time. This neglect makes it difficult for the system to maintain long-term positioning consistency when facing environmental changes, especially when new features gradually form in the wear area, often failing to identify and utilize these changes in a timely manner. Focusing on specific technical challenges, the textural evolution of the ground wear area becomes a core factor affecting long-term positioning stability. Ground wear not only leads to the loss of original texture details, reducing the clarity of feature recognition, but also forms new boundaries and patterns as wear deepens. While these newly formed wear boundaries may possess a certain degree of temporal stability, the uncertainty and complexity of their formation process make it difficult to accurately determine which boundaries are stable and which are still changing. This difficulty in judgment directly leads to navigation systems mistakenly treating unstable areas as reliable references or ignoring truly stable areas when updating path information, thus causing positioning errors. For example, in a large warehouse, the ground is primarily a hardened cement layer. Due to hundreds of daily forklift runs, multiple main wear zones have formed. Some of the tire grooves on these main paths have evolved into stable grooves of uniform width and depth after months of use, maintaining high visibility even under varying lighting conditions, making them ideal positioning references. However, the wear on side paths or temporary stacking areas is dynamic, with boundaries constantly expanding or breaking due to irregular loads. If the navigation system cannot distinguish the stability differences between these areas, it may repeatedly adjust its path planning, leading to inconsistent robot paths, or even deviations or stalls. Therefore, accurately identifying and utilizing the stable boundary features in ground wear areas as reliable references for long-term positioning has become a critical problem that urgently needs to be solved. Summary of the Invention

[0003] This invention provides a method for analyzing the redundancy of texture information in trackless navigation, mainly including:

[0004] An edge detection algorithm is used to process the original image data of the current wear area to obtain the wear boundary contour coordinates and grayscale distribution data.

[0005] The degree of texture detail loss is identified based on the wear boundary contour coordinates and grayscale distribution data. The degree of texture detail loss is then classified and stratified to assess the redundancy of texture information and determine the initial stable pattern region.

[0006] The variation amplitude of the preliminary stable pattern region within the continuous acquisition cycle is evaluated against a preset threshold. When the variation amplitude is lower than the preset threshold, the preliminary stable pattern region is confirmed as a reliable anchor point region.

[0007] Extract the boundary coordinate point set from the reliable anchor point area and determine the optimized update frequency value based on the position and shape matching degree between the boundary coordinate point set and the historical path map data;

[0008] Periodic matching evaluations are performed based on optimized update frequency values ​​to maintain long-term positioning consistency.

[0009] Furthermore, the step of processing the original image data of the current wear area using an edge detection algorithm to obtain the wear boundary contour coordinates and grayscale distribution data includes:

[0010] The original image data of the worn area is acquired and Gaussian filtering is performed. The filtered image is then processed using... The operator calculates the gradient values ​​in the horizontal and vertical directions to determine the edge points; from the set of edge points, the eight-neighborhood tracking method is used to extract a continuous contour chain, record the coordinates of each pixel on the contour chain, and at the same time obtain the gray value corresponding to the position of each pixel on the contour chain. The gray value difference between adjacent pixels is calculated and the difference distribution is statistically analyzed to obtain the wear boundary contour coordinate sequence and gray value distribution data.

[0011] Furthermore, the step of identifying the degree of texture detail loss based on the wear boundary contour coordinates and grayscale distribution data, classifying the degree of texture detail loss into layers, evaluating texture information redundancy, and determining preliminary stable pattern regions includes:

[0012] The wear boundary contour coordinate sequence is obtained and the Euclidean distance between adjacent coordinate points is calculated. The standard deviation of the distance value sequence is used to determine the continuity. The gray values ​​of the contour points are extracted from the gray-scale distribution data and the gray-scale change rate of adjacent points is calculated. The ratio of the number of points with a change rate exceeding a preset threshold to the total number of contour points is used to obtain the texture detail loss value. The texture detail loss value is compared with a preset interval threshold for hierarchical classification. Local texture blocks are extracted using a fixed-size window. By calculating the gray-scale histograms of different texture blocks and comparing the Bach distance, repetitive texture patterns are identified. The proportion of pixels occupied by repetitive texture patterns is used to obtain the texture information redundancy. When the texture information redundancy is higher than the preset threshold and the continuity is high, a preliminary stable pattern region is determined.

[0013] Furthermore, it also includes: obtaining the groove depth and groove width formed by tire marks from the wear boundary contour coordinates, collecting the clarity of the groove edge under different lighting conditions, analyzing the uniformity of the groove depth and the consistency of the groove width along the entire path, evaluating the continuity and integrity of the edge contour and the boundary fracture situation, and determining the stable groove area formed by the main road and the dynamic expansion area generated by the branch road.

[0014] Furthermore, the groove depth and width formed by tire marks are obtained from the wear boundary contour coordinates. The clarity of the groove edges under different lighting conditions is collected. The uniformity of the groove depth and the consistency of the groove width along the entire path are analyzed. The continuity and integrity of the edge contour and the boundary breakage are evaluated. The stable groove area formed by the main road and the dynamic expansion area generated by the branch road are determined. This includes: extracting the vertical distance difference between adjacent coordinate points from the wear boundary contour coordinate sequence to obtain the groove depth value; calculating the groove width value using the horizontal coordinate difference between the left and right boundary points of the contour; collecting images under different lighting intensities and calculating the average edge pixel gradient amplitude as the clarity value; calculating the coefficient of variation based on the groove depth value sequence to determine the depth uniformity; calculating the standard deviation of the groove width value sequence to evaluate the width consistency; detecting straight line segments through Hough transform and counting the number of discontinuities to obtain boundary breakage feature data; when the coefficient of variation and standard deviation are lower than the preset threshold, it is determined to be a stable groove area of ​​the main road; otherwise, the dynamic expansion area of ​​the branch road is determined based on the discontinuity density and the clarity fluctuation amplitude.

[0015] Furthermore, the evaluation of the variation amplitude of the preliminary stable pattern region within a continuous acquisition period and a preset threshold, wherein when the variation amplitude is lower than the preset threshold, the preliminary stable pattern region is confirmed as a reliable anchor point region, includes:

[0016] The boundary coordinate sequence and grayscale distribution sequence of the preliminary stable pattern region within a continuous acquisition cycle are obtained. The Euclidean distance between corresponding position coordinates during adjacent cycles is calculated as the position offset. The position change amplitude is obtained by taking the square root of the sum of the squares of the offsets. The grayscale change amplitude is obtained by calculating the root mean square of the difference in grayscale values ​​during adjacent cycles. The comprehensive change amplitude value is obtained by fusing the results. When the comprehensive change amplitude value is lower than a preset threshold, the preliminary stable pattern region is marked as a reliable anchor point region.

[0017] Furthermore, the evaluation of the variation amplitude of the preliminary stable pattern region within a continuous acquisition period and a preset threshold, and the confirmation that the preliminary stable pattern region is a reliable anchor point region when the variation amplitude is lower than the preset threshold, includes:

[0018] The boundary contour is extracted from the reliable anchor point region, and the boundary coordinate point set is obtained using the equal-interval sampling method. Principal component analysis is performed on the boundary coordinate point set to extract the principal axis direction and aspect ratio features. The centroid position and circumscribed rectangle parameters of the boundary coordinate point set are calculated, and a feature vector containing geometric parameters is constructed. The reference contour point set of the corresponding position in the historical path map data is obtained. The rotation matrix and translation vector between the boundary coordinate point set and the reference contour point set are calculated using the iterative nearest point algorithm. The position matching degree is evaluated based on the mean distance of the aligned point pairs. The shape matching degree is obtained by calculating the cosine similarity between the feature vector and the corresponding feature vector in the historical records. The texture detail loss value sequence within the most recent period is extracted, and the coefficient of variation is calculated as a stability index. The update cycle is set according to the stability index, position matching degree, and shape matching degree, or a comprehensive score is obtained by weighted summation. The optimized update frequency value is determined according to the interval to which the comprehensive score belongs.

[0019] Furthermore, after determining the optimized update frequency value based on the interval to which the comprehensive score belongs, it includes:

[0020] The update time is determined based on the optimized update frequency value. The original data of the area to be updated in the path map is obtained. The grayscale histogram and the corresponding texture detail loss value of the reliable anchor point area are extracted. The grayscale histogram is compared with the historical grayscale data in the original map to identify the changed areas. Incremental updates are performed on the changed areas. The boundary coordinate points of the reliable anchor point area are connected to form a closed contour. The mean and variance of the distance between adjacent points are calculated as geometric constraint parameters. The boundary deformation amplitude is limited according to the geometric constraint parameters. The updated map data, the grayscale histogram, the texture detail loss value and the geometric constraint parameters are integrated to construct a feature vector set containing spatial location, grayscale features and texture attributes. A spatial index tree is built on the feature vector set. The timestamp information is fused to form reference data and normalization processing is performed to generate an enhanced positioning reference dataset.

[0021] Furthermore, after generating the enhanced positioning reference dataset, the process includes: retrieving a set of reference feature vectors within the corresponding spatial range from the enhanced positioning reference dataset; calculating the Euclidean distance between the current feature vector and each vector in the set of reference feature vectors; converting the distance values ​​into similarity scores; and averaging the similarity scores to obtain the overall matching degree. When the overall matching degree is higher than a preset threshold, the coordinate information of the corresponding reliable anchor point region is extracted; the coordinate difference between the current image feature point and the anchor point feature point is calculated to obtain an offset vector; and the current position is calibrated based on the offset vector. When the overall matching degree is not higher than a preset threshold, additional image acquisition is triggered and features are extracted; the result with the highest matching degree is selected to determine supplementary data.

[0022] Furthermore, before performing periodic matching evaluation based on the optimized update frequency value to maintain long-term positioning consistency, this includes:

[0023] The feature information in the supplementary data is obtained and compared with the enhanced positioning reference dataset to identify new features. The new features are added to the corresponding spatial location of the dataset and the timestamp is updated to obtain the expanded positioning reference dataset. The evaluation period is set according to the optimized update frequency value. At each period, the current navigation image is extracted and the matching degree is calculated with the expanded positioning reference dataset. The matching degree value and the corresponding time are recorded to form a time series record.

[0024] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0025] This invention discloses a method for analyzing the redundancy of texture information in trackless navigation. Addressing the challenges of texture detail loss in tire wear areas, unstable boundary contours, and dynamic changes in positioning reference data in various scenarios, the method extracts wear boundary contours and grayscale distribution data using edge detection algorithms. It then assesses texture redundancy using geometric continuity and intensity gradients to identify stable pattern regions. Furthermore, it confirms reliable anchor point regions through multi-period variation amplitude evaluation. Finally, it fuses boundary coordinate point sets with historical path maps to optimize update frequency, enabling incremental updates of the path map and the generation of enhanced positioning reference datasets. This invention dynamically calibrates the position during navigation through matching degree evaluation and periodically updates data to maintain long-term consistency, ultimately achieving high-precision positioning and path planning, significantly improving navigation stability and reliability in complex environments. Attached Figure Description

[0026] Figure 1 This is a flowchart of a method for analyzing the redundancy of texture information in trackless navigation according to the present invention.

[0027] Figure 2 This is a schematic diagram of a method for analyzing the redundancy of texture information in trackless navigation according to the present invention.

[0028] Figure 3 This is another schematic diagram of a method for analyzing the redundancy of texture information in trackless navigation according to the present invention. Detailed Implementation

[0029] 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.

[0030] like Figures 1-3 This embodiment of a method for analyzing the redundancy of texture information in trackless navigation may specifically include:

[0031] S101. The original image data of the current wear area is processed using an edge detection algorithm to obtain the wear boundary contour coordinates and grayscale distribution data.

[0032] The original image data of the worn area is acquired and Gaussian filtering is performed to remove noise interference. The filtered image is then processed using... The operator calculates the gradient values ​​in the horizontal and vertical directions respectively. The gradient magnitude is obtained by taking the square root of the sum of the squares of the two gradient values. When the gradient magnitude exceeds a preset threshold, the pixel position is determined as an edge point. A continuous contour chain is extracted from the edge point set using the eight-neighborhood tracking method. The horizontal and vertical coordinates of each pixel on the contour chain are recorded, and the gray values ​​corresponding to the pixel positions of the contour chain are obtained. The gray value difference between adjacent pixels is calculated and the difference distribution is statistically analyzed to obtain the wear boundary contour coordinate sequence and gray value distribution data.

[0033] In one implementation, edge detection of the worn area achieves precise positioning through a gradient operator.

[0034] Specifically, the Gaussian filter uses a 5×5 convolution kernel with a standard deviation of 1.5 to perform convolution operations on the original image, suppressing salt-and-pepper noise and high-frequency interference in the image.

[0035] It should be noted that, The horizontal convolution kernel of the operator 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]. Convolution operations are performed on each pixel position of the filtered image to obtain the horizontal gradient. and vertical gradient Through calculation Obtain the gradient magnitude. When the gradient magnitude exceeds a preset threshold of 80, the pixel location is marked as an edge point.

[0036] Preferably, the eight-neighborhood tracking method starts from any edge point and searches for the positions of the eight adjacent pixels in a clockwise direction. After finding the next edge point, it continues tracking until a closed contour is formed or the image boundary is reached. The coordinates of each contour point are stored in a sequence according to the tracking order.

[0037] For example, the grayscale difference statistics adopt a sliding window method to calculate the grayscale mean difference of 5 adjacent pixels on the contour. When the difference is less than 20, the contour segment is considered continuous; otherwise, it is marked as a break position, forming complete wear boundary description data.

[0038] S102. Identify the degree of texture detail loss based on the wear boundary contour coordinates and grayscale distribution data, classify the degree of texture detail loss into layers, evaluate the redundancy of texture information, and determine the preliminary stable pattern area.

[0039] The wear boundary contour coordinate sequence is obtained, and the Euclidean distance between adjacent coordinate points is calculated. The standard deviation of the distance value sequence is calculated. When the standard deviation is less than a first preset threshold, it is judged as high continuity; when the standard deviation is between the first and second preset thresholds, it is judged as medium continuity; otherwise, it is judged as low continuity. Simultaneously, the gray values ​​of contour points are extracted from the gray-scale distribution data, and the gray-scale change rate of five adjacent points is calculated. When the change rate exceeds a third preset threshold, it is marked as an intensity gradient abrupt change point. The ratio of the number of abrupt change points to the total number of contour points is calculated to obtain the texture detail loss value. Based on the texture detail loss value and a preset interval threshold, hierarchical classification is performed. A fixed-size window is used to extract local texture blocks within the wear area at a preset step size. Repeated texture patterns are identified by calculating the gray-scale histograms of different texture blocks and comparing the Barcol distance of the histograms. , and There are two histograms. for If the distance is less than 0.1, it is judged as a repeating pattern. The redundancy of texture information is obtained by calculating the ratio of the total number of pixels occupied by the repeating texture pattern to the total number of pixels in the wear area. If the redundancy of texture information is higher than the fourth preset threshold and the continuity determination result is high continuity, it is determined that the region forms a regular pattern. Combining the distribution density of the intensity gradient abrupt change points (calculated as the average interval distance of the abrupt change points on the contour length) and the hierarchical classification result of the texture detail loss degree value, a preliminary stable pattern region is determined.

[0040] In one implementation, the degree of texture detail loss is accurately quantified through multi-dimensional feature analysis. Floor wear in warehouse environments exhibits a gradual deepening from the edges to the center; the geometric features and grayscale changes of the wear boundary become key indicators for judging texture loss. The Euclidean distance is calculated by taking the square root of the sum of the squares of the coordinate differences between adjacent contour points, i.e., for the first... Points and the Points The distance value is Geometric continuity is assessed by calculating the standard deviation of the distance values ​​across the entire contour sequence; the standard deviation reflects the smoothness of the contour. The grayscale change rate is calculated using the five-point difference method, which measures the grayscale values ​​of five consecutive points on the contour. to The rate of change When the rate of change exceeds a preset threshold of 30, it is marked as an intensity gradient abrupt change point. The proportion of abrupt change points directly reflects the degree of loss of texture details; the higher the proportion, the less of the original texture features are preserved.

[0041] Preferably, the layered classification sets three ranges based on the degree of loss of texture details: below 0.3 is light wear, 0.3 to 0.7 is moderate wear, and above 0.7 is heavy wear. Different levels correspond to different subsequent processing strategies.

[0042] For example, the assessment of texture information redundancy is based on the Bach distance metric to measure the similarity between different texture patches. The Bach distance calculates the degree of overlap between two normalized histograms, with a value ranging from 0 to 1; a higher value indicates higher similarity. When the Bach distance between multiple texture patches exceeds 0.8, a duplicate texture pattern is considered to exist.

[0043] Specifically, the determination of a stable pattern region takes into account three factors: texture information redundancy, geometric continuity, and intensity gradient abrupt change point distribution density. Only when all three conditions meet the preset criteria can it be determined that the region has formed a stable pattern that can be used for long-term positioning.

[0044] The groove depth and width formed by tire marks are obtained from the wear boundary contour coordinates. The clarity of the groove edge under different lighting conditions is collected. The uniformity of the groove depth and the consistency of the groove width along the entire path are analyzed. The continuity and integrity of the edge contour and the boundary fracture are evaluated. The stable groove area formed by the main road and the dynamic expansion area generated by the branch road are determined.

[0045] The vertical distance difference between adjacent coordinate points is extracted from the wear boundary contour coordinate sequence. The groove depth formed by the tire mark is obtained by calculating the height difference between the lowest point on the inner side of the contour and the highest point on the outer side. The groove width is calculated by using the horizontal coordinate difference between the left and right boundary points of the contour. Images of the same groove location are acquired under different light intensities, and the average gradient magnitude of the edge pixels is calculated as the sharpness value. The depth variation coefficient is obtained by calculating the ratio of the mean to the standard deviation of the groove depth value sequence. When the variation coefficient is less than a first preset threshold, the depth is considered uniform. The standard deviation of the groove width value sequence is calculated to evaluate the width consistency. Straight segments in the contour are detected by Hough transform, and the number and spacing of discontinuities between straight segments are counted to obtain boundary breakage feature data. If the depth variation coefficient is less than the first preset threshold and the standard deviation of the width value sequence is less than the second preset threshold, it is determined to be a stable groove area of ​​the main road. Otherwise, if the density of intermediate discontinuities in the boundary breakage feature data exceeds the third preset threshold and the fluctuation amplitude of the sharpness value in the continuous acquisition period is higher than the fourth preset threshold, it is determined to be a dynamic expansion area of ​​the branch road. Otherwise, it is determined to be another area.

[0046] In one implementation, precise extraction of groove features is achieved through multi-dimensional measurements. Tire marks created by repeated forklift traffic in a warehouse exhibit differentiated morphological characteristics across different road sections. High-frequency use on main roads generates grooves with stable depths, while intermittent traffic on side roads causes continuous changes in groove boundaries. Groove depth is measured based on the difference in the vertical components of the profile coordinates.

[0047] Specifically, viewed from the cross-sectional perspective of the wear boundary profile, the groove exhibits a concave shape. The groove depth is determined by identifying the lowest point in the profile sequence as the groove bottom and calculating the height difference between this point and the adjacent highest point. The groove width is obtained by determining the horizontal coordinates of the edge points on both sides of the groove and calculating their lateral spacing. In actual measurements, depth and width data are sampled every 10 centimeters, forming a measurement sequence distributed along the path. This dense sampling method captures subtle changes in the groove morphology, providing a data foundation for subsequent uniformity assessment.

[0048] Preferably, the changes in lighting conditions are achieved by acquiring images at three different times: morning, noon, and evening. Under different lighting angles, the shadow distribution at the groove edge changes, and the sharpness of the edge is quantified by calculating the gradient magnitude of the edge pixels and taking the average value.

[0049] For example, the coefficient of variation, as a measure of relative dispersion, is calculated as the ratio of the standard deviation to the mean, thus eliminating the influence of dimensions. When the coefficient of variation for trench depth is less than 0.15, it indicates a relatively uniform depth distribution. Width consistency is directly measured using the standard deviation; the smaller the standard deviation, the more stable the width variation.

[0050] In one possible implementation, the Hough transform detects straight line segments in the contour through a voting mechanism in the parameter space. Each edge point corresponds to a curve in the parameter space, and curves with multiple collinear points intersect at a single point. Straight line segments are identified by statistically analyzing the cumulative value of the intersection points.

[0051] Specifically, the determination of stable trench regions and dynamically expanding regions takes into account both geometric features and time-varying characteristics. Regions with uniform depth, consistent width, and continuous boundaries are determined to be stable regions.

[0052] S103. Evaluate the variation amplitude of the preliminary stable pattern region extraction within the continuous acquisition cycle and compare it with the preset threshold. When the variation amplitude is lower than the preset threshold, confirm that the preliminary stable pattern region is a reliable anchor point region. When the variation amplitude is not lower than the preset threshold, repeat the acquisition and re-identify the degree of texture detail loss to obtain an updated stable pattern region.

[0053] The process involves acquiring the boundary coordinate sequence and grayscale distribution sequence of a preliminary stable pattern region over a predetermined number of consecutive acquisition cycles. The Euclidean distance between corresponding position coordinates during adjacent cycles is calculated as the position offset. The square root of the sum of the squares of all offsets is used to obtain the position change amplitude. The root mean square of the grayscale value difference between adjacent cycles is calculated to obtain the grayscale change amplitude. A weighted average method is used to fuse the position change amplitude and the grayscale change amplitude to obtain a comprehensive change amplitude value. If the comprehensive change amplitude value is lower than a first preset threshold, the preliminary stable pattern region is marked as a reliable anchor point region, and its boundary coordinate set and texture feature vector are recorded as positioning reference data. If the comprehensive change amplitude value is not lower than the first preset threshold, a new image acquisition is triggered, and the current wear area image is extracted. The degree of texture detail loss is recalculated based on the current wear area image. The comprehensive change amplitude values ​​from multiple cycles are constructed into a time series, and their rate of change is calculated. Defined as , For the first Periodic amplitude value, The amplitude value of the previous period, when the rate of change When it approaches zero, the original boundary range is maintained, and when the rate of change To shrink the boundary range to a region where the change amplitude is less than the second preset threshold, when the change rate When the value is negative, the boundary range is extended to the stable region, resulting in an updated stable pattern region.

[0054] In one implementation, the dynamic assessment of stable pattern regions is achieved through multi-period continuous monitoring. Wear areas in a warehouse environment are affected by changes in cargo stacking and vehicle load, and their texture features exhibit a gradual evolution. By tracking this evolution process, areas that have tended to stabilize are identified. The calculation of the magnitude of change involves a comprehensive assessment of both spatial location and grayscale features.

[0055] Specifically, after acquiring an image of the wear area in each acquisition cycle, the coordinates of equally spaced sampling points on the boundary contour are extracted to form a coordinate sequence. For the first... Coordinate points of each cycle and the Corresponding points of each cycle Calculate the Euclidean distance This serves as the position offset at that point. The square root of the sum of the squares of the offsets of all sampling points yields the overall positional change amplitude for that period. The grayscale change amplitude is obtained by calculating the root mean square of the grayscale value difference between corresponding pixels. A weighted average method is used to fuse the two types of change amplitudes. The weighting coefficients are dynamically adjusted based on the stability of historical data; regions with stable positional changes are assigned higher grayscale weights, and vice versa.

[0056] Preferably, the first preset threshold is set based on the statistical characteristics of historical stable regions, and is usually taken as 1.5 times the average change range of the stable regions, which avoids misjudgment and ensures the accuracy of identification.

[0057] For example, the time series is constructed using a sliding window mechanism, with each window containing the combined change amplitude values ​​of 5 consecutive periods. The slope of the change trend is obtained as the rate of change by fitting the data points within the window using the least squares method.

[0058] In one possible implementation, the boundary range adjustment follows a conservative shrinkage principle. When a positive rate of change is detected, it indicates that the region is still evolving. At this point, the region shrinks pixel by pixel from the boundary inward until a stable core region with a change magnitude less than a second preset threshold is found. This gradual shrinkage strategy can preserve the available stable features to the greatest extent.

[0059] S104. Extract the boundary coordinate point set from the reliable anchor point area and determine the optimized update frequency value based on the position and shape matching degree between the boundary coordinate point set and the historical path map data.

[0060] Boundary contours are extracted from reliable anchor point regions, and boundary coordinate point sets are obtained using an equal-interval sampling method. Principal component analysis is performed on the coordinate point sets to extract the principal axis direction and aspect ratio features. Simultaneously, the centroid position and circumscribed rectangle parameters of the coordinate point sets are calculated, constructing a feature vector containing four types of geometric parameters. A reference contour point set for the corresponding position in historical path map data is obtained. The rotation matrix and translation vector between the boundary coordinate point set and the reference contour point set are calculated using an iterative nearest-point algorithm. The position matching degree is evaluated based on the mean distance between aligned point pairs. The shape matching degree is obtained by calculating the cosine similarity between each component of the feature vector and the corresponding feature vector in the historical record. A sequence of texture detail loss values ​​within the most recent preset period is extracted, and its coefficient of variation is calculated as a stability index. When the stability index is lower than a first preset threshold and the position matching degree is higher than a second preset threshold, the update period is set to a first duration. When the stability index is higher than the first preset threshold or the shape matching degree is lower than a third preset threshold, the update period is set to a second duration. Optionally, the texture detail loss value is calculated by comparing the local variance difference between the current image and historical images. The comprehensive score is obtained by multiplying the position matching degree, shape matching degree and stability index by their respective weight coefficients and summing them. The comprehensive score is divided into a preset number of intervals, and each interval corresponds to an update frequency value. The optimized update frequency value is determined by judging the interval to which the comprehensive score belongs.

[0061] In one implementation, feature extraction and matching evaluation of reliable anchor point regions constitute the core of update frequency optimization. The stable features formed by the wear and tear areas of the warehouse center after long-term use directly affect the balance between positioning accuracy and computational resources through their update frequency. An adaptive update strategy is achieved through precise feature matching and stability evaluation. Principal component analysis (PCA) in feature extraction reduces the high-dimensional set of boundary coordinate points to a feature space with clear physical meaning.

[0062] Specifically, sampling points on the boundary contour are acquired at equal intervals, with the interval dynamically adjusted according to the total length of the contour, typically 100 points per meter. A covariance matrix is ​​constructed for these coordinate points, and eigenvalues ​​and eigenvectors are calculated. The eigenvector corresponding to the largest eigenvalue represents the principal axis direction of the shape, characterizing the main extension direction of the wear area. The ratio of the second largest eigenvalue to the largest eigenvalue reflects the aspect ratio of the shape; a ratio closer to 1 indicates a shape closer to a circle, while a smaller ratio indicates a more elongated shape. The centroid position is obtained by the arithmetic mean of the coordinates of all sampling points, representing the geometric center of the anchor point region. The circumscribed rectangle parameters include the rectangle's length, width, and rotation angle, and the minimum area circumscribed rectangle is obtained using a caliper rotation algorithm. These four types of geometric parameters together constitute the eigenvector, comprehensively describing the morphological characteristics of the anchor point region.

[0063] Preferably, the iterative nearest neighbor algorithm achieves accurate registration between the current anchor point region and the historical map. The core of the algorithm is to find the spatial transformation that minimizes the distance between two point sets through iterative optimization. In each iteration, for each point in the current boundary coordinate point set, the nearest neighbor point is found in the reference contour point set, forming a set of point pairs. Based on these point pairs, the rotation matrix for the current iteration is calculated using singular value decomposition. Translation vector After applying the transformation to the current point set, the mean Euclidean distance between all point pairs is calculated as the registration error. The algorithm terminates when the change in registration error is less than the convergence threshold or when the maximum number of iterations is reached. The registration error directly reflects the degree of positional matching; the smaller the error, the closer the current anchor point position is to the historical record. This registration method can handle minor deformations and positional shifts caused by ground wear, achieving robust positional matching.

[0064] For example, shape matching quantifies the similarity between the current feature vector and historical feature vectors using cosine similarity. Cosine similarity calculates the cosine of the angle between two vectors, ranging from -1 to 1; the closer the value is to 1, the more similar the shapes. For feature vectors containing parameters such as principal axis direction, aspect ratio, centroid position, and circumscribed rectangle, normalization is first performed to eliminate the influence of dimensions, and then the inner product is calculated and divided by the product of the magnitudes.

[0065] In one possible implementation, the stability metric is obtained by analyzing the temporal variation of texture detail loss values. Texture detail loss values ​​from the most recent 10 acquisition periods are extracted from historical records, and the mean and standard deviation of this sequence are calculated. The coefficient of variation, equal to the standard deviation divided by the mean, reflects the relative degree of change.

[0066] Specifically, a tiered strategy is adopted to determine the update cycle. When the stability index is below 0.1 and the position matching degree is above 0.9, it indicates that the anchor point area is highly stable, and the update cycle is set to 24 hours to reduce unnecessary computational overhead. When the stability index is above 0.1 or the shape matching degree is below 0.7, it indicates that the area is still changing, and the update cycle is set to 2 hours to ensure timely capture of feature changes. For cases in between, the update cycle is set to 8 hours to achieve a balance between accuracy and efficiency. Furthermore, a comprehensive scoring mechanism integrates multiple evaluation dimensions. The weight coefficients for position matching degree, shape matching degree, and stability index are set to 0.4, 0.3, and 0.3, respectively, reflecting the importance of position accuracy in localization. The three indicators are multiplied by their respective weights and then summed to obtain a comprehensive score between 0 and 1.

[0067] For example, the mapping relationship between the overall score range and the update frequency is pre-defined: a score of 0.8-1.0 corresponds to 1 update per day, 0.6-0.8 corresponds to 3 updates per day, 0.4-0.6 corresponds to 6 updates per day, 0.2-0.4 corresponds to 12 updates per day, and 0-0.2 corresponds to 1 update per hour. By looking up the table to determine the update frequency value, the conversion from continuous scoring to discrete frequency is realized.

[0068] S105. The path map is incrementally updated using an optimized update frequency value. The grayscale distribution data of reliable anchor point areas and the degree of loss of texture details are integrated to generate an enhanced positioning reference dataset.

[0069] The current update time is determined based on the optimized update frequency value. The original data of the area to be updated in the path map is obtained and timestamped. The grayscale histogram and corresponding texture detail loss value of the reliable anchor point area are extracted. The grayscale histogram is compared with the historical grayscale data stored in the original map to identify areas where changes exceed a preset threshold. Incremental updates are performed on the changed areas, while retaining the original data of the unchanged areas. The boundary coordinate points of the reliable anchor point area are connected sequentially to form a closed contour. The mean and variance of the distances between adjacent points are calculated as geometric constraint parameters, and the maximum deformation amplitude of the boundary is limited based on these parameters. The updated map data, the grayscale histogram, the texture detail loss value, and the geometric constraint parameters are integrated to construct a feature vector set containing spatial location, grayscale features, and texture attributes. A spatial index tree is built on the feature vector set to accelerate retrieval. Based on the feature vector set and the spatial index tree, timestamp information is fused to form reference data with time-series labels. Normalization processing is performed on the reference data to eliminate the influence of dimensions, generating an enhanced positioning reference dataset.

[0070] In one implementation, the incremental update mechanism of the route map achieves a dynamic balance between computing resources and update accuracy. The wear characteristics of the warehouse floor evolve at different rates depending on the intensity of business operations. By adaptively adjusting the update frequency, the resource waste caused by frequent updates is avoided, while ensuring the timeliness of the positioning reference is maintained.

[0071] It should be noted that determining the update time depends on the optimized update frequency and the coordination of the system clock. When the system clock reaches the predetermined update time, the update process is triggered. First, the storage address of the area to be updated in the map is read, and the original data of the area is extracted, including the coordinate range, grayscale distribution, and the timestamp of the last update. This data is used as a comparison benchmark.

[0072] Specifically, the difference comparison process achieves accurate change detection through multi-level feature comparison. The grayscale histogram comparison uses the Bhattacharyya distance metric, comparing the current grayscale histogram of reliable anchor point regions with historical grayscale data. right The comparison is as follows: A grayscale histogram is typically divided into 256 gray levels, with each gray level recording the frequency of that gray value within a region. The Bach distance is calculated by comparing the corresponding values ​​of the two normalized histograms. The sum of the square roots of the product values ​​ranges from 0 to 1. When the Bach distance is less than 0.8, the grayscale distribution is considered essentially unchanged; when the distance is between 0.8 and 0.6, it indicates a slight change; and when the distance is less than 0.6, it is considered a significant change requiring updating. The comparison of texture detail loss values ​​uses a direct difference method; a difference exceeding 0.2 between the current and historical values ​​is marked as a change in texture features. By comprehensively considering changes in both grayscale and texture dimensions, the boundaries of regions requiring updating are accurately identified.

[0073] Preferably, the calculation of geometric constraint parameters provides a quantitative basis for the stability of the boundary morphology. A closed profile is formed by connecting the boundary coordinate points sequentially according to the sampling order, with the last point connected to the first point to achieve closure. The mean distance between adjacent points reflects the sampling density, while the variance characterizes the smoothness of the boundary; a smaller variance indicates a more regular boundary.

[0074] For example, the construction of the feature vector set integrates information from three dimensions: spatial, grayscale, and texture. Each feature vector contains five components: two-dimensional spatial coordinates represent location information, the grayscale mean and standard deviation describe local grayscale characteristics, and the texture detail loss value quantifies the degree of texture degradation. For an anchor region containing 1000 sampling points, 1000 five-dimensional feature vectors are formed. The spatial index tree uses... A tree structure is used to efficiently organize these 1000 five-dimensional feature vectors. A tree is a balanced index tree specifically designed for multidimensional spatial data. Its input consists of the first two spatial coordinates of all feature vectors as the location basis. The construction process is as follows: vectors are inserted from bottom to top. First, the minimum bounding rectangle is generated using spatial coordinates. Leaf nodes store the actual feature vectors, and non-leaf nodes aggregate the minimum bounding rectangles of their child nodes until a balanced tree is formed. The reason for using trees lies in their support for dynamic indexing of high-dimensional spatial data and their logarithmic query complexity. The purpose of this structure is to partition feature vectors hierarchically according to their spatial location, thereby optimizing subsequent searches from linear scans to range queries with logarithmic time complexity.

[0075] In one possible implementation, the spatial index tree is constructed using a bottom-up batch loading method. First, all feature vectors are sorted according to their spatial location, then grouped into leaf nodes. Each leaf node contains several adjacent feature vectors, and this process is repeated layer by layer upwards until the root node. Furthermore, temporal labeling is implemented by appending a timestamp field to each feature vector. The timestamp records the moment the feature was acquired with millisecond-level precision, enabling the system to trace the evolution of the features.

[0076] For example, normalization eliminates dimensional differences between different feature components. Spatial coordinates divided by the maximum span of the region are mapped to the [0,1] interval, grayscale values ​​are normalized by dividing by 255, and the degree of texture detail loss is already within the [0,1] range and does not require further processing. The normalized feature vectors facilitate distance calculation and similarity comparison. The enhanced localization reference dataset provides a reliable and easily accessible localization benchmark for navigation systems by fusing multi-source information and establishing efficient indexes.

[0077] S106. During navigation, evaluate the matching degree between the current image data and the enhanced positioning reference dataset. When the matching degree is higher than a preset threshold, directly use the reliable anchor point area for position calibration. When the matching degree is not higher than the preset threshold, trigger additional image acquisition to obtain supplementary data.

[0078] During navigation, real-time image data of the current location is acquired and its feature vectors are extracted. A set of reference feature vectors within the corresponding spatial range is retrieved from the enhanced positioning reference dataset. The Euclidean distance between the current feature vector and each vector in the reference feature vector set is calculated. The distance values ​​are converted into similarity scores using a negative exponential function, and the average of these similarity scores is used to obtain the overall matching degree. If the overall matching degree is higher than a preset threshold, the coordinate information of the corresponding reliable anchor point region is extracted from the enhanced positioning reference dataset. The coordinate difference between the current image feature point and the anchor point feature point is calculated to obtain an offset vector, and the current navigation position is calibrated based on the offset vector. If the overall matching degree is not higher than the preset threshold, an additional image acquisition process is triggered to acquire multiple supplementary images around the current location. Features are extracted from these supplementary images and matched with the reference dataset. The result with the highest matching degree is selected to determine the positioning location and obtain supplementary data.

[0079] In one implementation, a matching evaluation mechanism during real-time navigation dynamically ensures positioning accuracy. The navigation system continuously evaluates the degree of matching between the current location and reference data during runtime, adaptively selecting a positioning strategy based on the matching results. It performs rapid calibration when the matching degree is high and triggers supplementary data acquisition when the matching degree is low. The calculation of feature matching degree involves two key steps: distance measurement and similarity conversion.

[0080] Specifically, the feature vector of the current image contains multiple dimensions such as spatial coordinates, grayscale features, and texture attributes. When calculating the Euclidean distance with the feature vectors in the reference dataset, the squared differences of each dimension are summed and then the square root is taken. The distance value reflects the degree of difference between features; the smaller the distance, the more similar the features. (Negative exponential function) Distance The similarity is converted to a similarity score, where a distance of 0 results in a similarity score of 1, and the similarity score decays exponentially to near zero as the distance increases. This conversion method avoids the singularity problem of the reciprocal of distance being zero. For multiple feature vectors in the reference set, the similarity scores are calculated separately, and then the average is taken to obtain the overall matching score. The overall matching score comprehensively reflects the degree of agreement between the current position and historical stable features.

[0081] Preferably, position calibration achieves precise positioning by calculating and applying an offset vector. When the overall matching degree exceeds a preset threshold of 0.8, it is determined that the current position is near a reliable anchor point area. The accurate coordinates of the anchor point are extracted from the reference data, and the coordinate differences between the current feature point and the anchor point feature point in the horizontal and vertical directions are calculated to form a two-dimensional offset vector.

[0082] For example, the supplementary acquisition strategy provides additional location information when the matching degree is insufficient. When the overall matching degree is below a threshold, one image frame is acquired 2 meters in front, 2 meters behind, 1 meter to the left, and 1 meter to the right of the current position to expand the feature search range. Feature extraction and matching calculation are performed independently on each supplementary image frame, and the position with the highest matching degree is selected as the corrected location result.

[0083] In one possible implementation, this adaptive matching evaluation and positioning calibration mechanism can achieve centimeter-level positioning accuracy in areas with stable wear characteristics, and maintain decimeter-level positioning reliability in transitional areas with changing characteristics.

[0084] S107. Update the enhanced positioning reference dataset based on the supplementary data and perform periodic matching degree evaluation based on the optimized update frequency value to maintain long-term positioning consistency.

[0085] Feature information from supplementary data is acquired and compared with an enhanced positioning reference dataset. Supplementary data refers to real-time sensor data such as navigation images or point clouds. New features not included in the dataset are identified in the supplementary data, added to their corresponding spatial locations, and timestamps are updated to obtain the expanded positioning reference dataset. An evaluation period is set based on an optimized update frequency. At each period, the current navigation image is extracted and its matching degree is calculated against the expanded positioning reference dataset. The matching degree calculation employs… The algorithm extracts image features and calculates similarity scores. The range is 0 to 1. The matching degree value and the corresponding time are recorded to form a time sequence record. By judging whether the matching degree value is stably greater than 0.8 within 5 consecutive cycles, long-term positioning consistency is maintained.

[0086] In one implementation, the supplementary data fusion and update mechanism enables the dynamic improvement of the positioning reference dataset. By continuously incorporating new feature information, the dataset can adapt to the gradual changes in ground wear. New features are identified through differential comparison; feature vectors extracted from the supplementary data are compared one-to-one with those in the existing dataset. When the minimum distance between a feature vector and all vectors in the dataset exceeds a preset threshold, it is determined to be a new feature. The new feature, carrying the current timestamp information, is inserted into the corresponding spatial location in the dataset, achieving incremental expansion of the dataset.

[0087] Preferably, the evaluation period is dynamically adjusted based on the update frequency. When the update frequency is once per hour, the evaluation period is set to 30 minutes; when the update frequency is once per day, the evaluation period is set to 12 hours. At each cycle point, the matching degree is automatically calculated, forming a continuous time-series record.

[0088] For example, long-term consistency is achieved through statistical analysis. When the matching degree value remains within a preset range of 0.7 to 0.9 for five consecutive periods and the standard deviation is less than 0.05, the system is considered to be in a stable state and the positioning consistency is effectively maintained.

[0089] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. The present invention has been described in detail with reference to preferred embodiments. Those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications and substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for analyzing the redundancy of texture information in trackless navigation, characterized in that, The method includes: An edge detection algorithm is used to process the original image data of the current wear area to obtain the wear boundary contour coordinates and grayscale distribution data. The text describes a process for identifying the degree of texture detail loss based on wear boundary contour coordinates and grayscale distribution data. This involves classifying these values ​​hierarchically, assessing texture information redundancy, and determining preliminary stable pattern regions. The process includes: acquiring a wear boundary contour coordinate sequence and calculating the Euclidean distance between adjacent coordinate points; determining continuity by statistically analyzing the standard deviation of the distance value sequence; extracting grayscale values ​​from contour points from grayscale distribution data and calculating the grayscale change rate of adjacent points; calculating the ratio of the number of points with a change rate exceeding a preset threshold to the total number of contour points to obtain the texture detail loss value; classifying the texture detail loss value hierarchically by comparing it with a preset interval threshold; extracting local texture blocks using a fixed-size window; identifying repetitive texture patterns by calculating the grayscale histograms of different texture blocks and comparing their distances; and obtaining texture information redundancy by statistically analyzing the proportion of pixels occupied by repetitive texture patterns. When the texture information redundancy is higher than a preset threshold and the continuity is high, a preliminary stable pattern region is determined. The variation amplitude of the preliminary stable pattern region within the continuous acquisition cycle is evaluated against a preset threshold. When the variation amplitude is lower than the preset threshold, the preliminary stable pattern region is confirmed as a reliable anchor point region. Extract the boundary coordinate point set from the reliable anchor point area and determine the optimized update frequency value based on the position and shape matching degree between the boundary coordinate point set and the historical path map data; Periodic matching evaluations are performed based on optimized update frequency values ​​to maintain long-term positioning consistency.

2. The method for analyzing the redundancy of texture information in trackless navigation according to claim 1, characterized in that, The process employs an edge detection algorithm to process the original image data of the current wear area, obtaining the wear boundary contour coordinates and grayscale distribution data, including: The original image data of the wear area is acquired and Gaussian filtering is performed. The Sobel operator is used to calculate the horizontal and vertical gradient values ​​of the filtered image to determine the edge points. A continuous contour chain is extracted from the edge point set using the eight-neighborhood tracking method. The coordinates of each pixel on the contour chain are recorded, and the gray value corresponding to each pixel position of the contour chain is obtained. The gray value difference between adjacent pixels is calculated and the difference distribution is statistically analyzed to obtain the wear boundary contour coordinate sequence and gray value distribution data.

3. The method for analyzing the redundancy of texture information in trackless navigation according to claim 1, characterized in that, Also includes: The groove depth and width formed by tire marks are obtained from the wear boundary contour coordinates. The clarity of the groove edge under different lighting conditions is collected. The uniformity of the groove depth and the consistency of the groove width along the entire path are analyzed. The continuity and integrity of the edge contour and the boundary fracture are evaluated. The stable groove area formed by the main road and the dynamic expansion area generated by the branch road are determined.

4. The method for analyzing the redundancy of texture information in trackless navigation according to claim 3, characterized in that, The groove depth and width formed by tire marks are obtained from the wear boundary contour coordinates. The clarity of the groove edges under different lighting conditions is collected. The uniformity of the groove depth and the consistency of the groove width along the entire path are analyzed. The continuity and integrity of the edge contour and the boundary breakage are evaluated. The stable groove area formed by the main road and the dynamic expansion area generated by the branch road are determined. This includes: extracting the vertical distance difference between adjacent coordinate points from the wear boundary contour coordinate sequence to obtain the groove depth value; calculating the groove width value using the horizontal coordinate difference between the left and right boundary points of the contour; collecting images under different lighting intensities and calculating the average gradient amplitude of edge pixels as the clarity value; calculating the coefficient of variation based on the groove depth value sequence to determine the depth uniformity; calculating the standard deviation of the groove width value sequence to evaluate the width consistency; detecting straight segments through Hough transform and counting the number of discontinuities to obtain boundary breakage feature data; when the coefficient of variation and standard deviation are lower than the preset threshold, it is determined to be a stable groove area of ​​the main road; otherwise, the dynamic expansion area of ​​the branch road is determined based on the density of discontinuities and the amplitude of clarity fluctuation.

5. The method for analyzing the redundancy of texture information in trackless navigation according to claim 1, characterized in that, The evaluation of the variation amplitude of the preliminary stable pattern region within a continuous acquisition cycle and a preset threshold, wherein when the variation amplitude is lower than the preset threshold, the preliminary stable pattern region is confirmed as a reliable anchor point region, includes: The boundary coordinate sequence and grayscale distribution sequence of the preliminary stable pattern region within a continuous acquisition cycle are obtained. The Euclidean distance between corresponding position coordinates during adjacent cycles is calculated as the position offset. The position change amplitude is obtained by taking the square root of the sum of the squares of the offsets. The grayscale change amplitude is obtained by calculating the root mean square of the difference in grayscale values ​​during adjacent cycles. The comprehensive change amplitude value is obtained by fusing the results. When the comprehensive change amplitude value is lower than a preset threshold, the preliminary stable pattern region is marked as a reliable anchor point region.

6. The method for analyzing the redundancy of texture information in trackless navigation according to claim 1, characterized in that, The process of evaluating the variation amplitude of the preliminary stable pattern region within a continuous acquisition cycle and comparing it with a preset threshold, and confirming the preliminary stable pattern region as a reliable anchor point region when the variation amplitude is lower than the preset threshold, includes: The boundary contour is extracted from the reliable anchor point region, and the boundary coordinate point set is obtained using the equal-interval sampling method. Principal component analysis is performed on the boundary coordinate point set to extract the principal axis direction and aspect ratio features. The centroid position and circumscribed rectangle parameters of the boundary coordinate point set are calculated, and a feature vector containing geometric parameters is constructed. The reference contour point set of the corresponding position in the historical path map data is obtained. The rotation matrix and translation vector between the boundary coordinate point set and the reference contour point set are calculated using the iterative nearest point algorithm. The position matching degree is evaluated based on the mean distance of the aligned point pairs. The shape matching degree is obtained by calculating the cosine similarity between the feature vector and the corresponding feature vector in the historical records. The texture detail loss value sequence within the most recent period is extracted, and the coefficient of variation is calculated as a stability index. The update cycle is set according to the stability index, position matching degree, and shape matching degree, or a comprehensive score is obtained by weighted summation. The optimized update frequency value is determined according to the interval to which the comprehensive score belongs.

7. The method for analyzing the redundancy of texture information in trackless navigation according to claim 6, characterized in that, After determining the optimized update frequency value based on the interval to which the comprehensive score belongs, it includes: The update time is determined based on the optimized update frequency value. The original data of the area to be updated in the path map is obtained. The grayscale histogram and the corresponding texture detail loss value of the reliable anchor point area are extracted. The grayscale histogram is compared with the historical grayscale data in the original map to identify the changed areas. Incremental updates are performed on the changed areas. The boundary coordinate points of the reliable anchor point area are connected to form a closed contour. The mean and variance of the distance between adjacent points are calculated as geometric constraint parameters. The boundary deformation amplitude is limited according to the geometric constraint parameters. The updated map data, the grayscale histogram, the texture detail loss value and the geometric constraint parameters are integrated to construct a feature vector set containing spatial location, grayscale features and texture attributes. A spatial index tree is built on the feature vector set. The timestamp information is fused to form reference data and normalization processing is performed to generate an enhanced positioning reference dataset.

8. The method for analyzing the redundancy of texture information in trackless navigation according to claim 7, characterized in that, After generating the enhanced positioning reference dataset, the process includes: retrieving a set of reference feature vectors within the corresponding spatial range from the enhanced positioning reference dataset; calculating the Euclidean distance between the current feature vector and each vector in the set of reference feature vectors; converting the distance values ​​into similarity scores; and averaging the similarity scores to obtain the overall matching degree. When the overall matching degree is higher than a preset threshold, the coordinate information of the corresponding reliable anchor point region is extracted; the coordinate difference between the current image feature point and the anchor point feature point is calculated to obtain an offset vector; and the current position is calibrated based on the offset vector. When the overall matching degree is not higher than a preset threshold, additional image acquisition is triggered and features are extracted; the result with the highest matching degree is selected to determine supplementary data.

9. The method for analyzing the redundancy of texture information in trackless navigation according to claim 8, characterized in that, Before performing periodic matching evaluation based on optimized update frequency values ​​to maintain long-term positioning consistency, the following steps are included: The feature information in the supplementary data is obtained and compared with the enhanced positioning reference dataset to identify new features. The new features are added to the corresponding spatial location of the dataset and the timestamp is updated to obtain the expanded positioning reference dataset. The evaluation period is set according to the optimized update frequency value. At each period, the current navigation image is extracted and the matching degree is calculated with the expanded positioning reference dataset. The matching degree value and the corresponding time are recorded to form a time series record.

Images