Intelligent stone slab layout storage management method based on machine vision

Abstract

The present application relates to the technical field of inventory management, in particular to a stone surface intelligent warehouse management method based on machine vision, comprising the following steps: multi-modal sensor acquisition stone image temperature distance data, gray quantization and time difference feature set generation, synchronous calibration, Kalman filter fusion and dynamic weight adjustment, output positioning result and analysis of reflectance coefficient adjustment light source, optimization of light field parameter collection texture reconstruction three-dimensional model and error correction update, based on model extraction position information correction inventory strategy generates inventory management scheme, in the present application, through multi-source sensor cooperative acquisition fusion, positioning consistency is kept, error diffusion is avoided, illumination monitoring and reflection adjustment optimize light field, ensure stone texture complete and stable, three-dimensional model dynamic reconstruction and correction real-time update feature synchronous storage state, improve stacking level and goods location recognition accuracy, enhance inventory information reliability and consistency.

Description

Technical Field

[0001] This invention relates to the field of inventory management technology, and in particular to a machine vision-based intelligent warehouse management method for stone panels. Background Technology

[0002] Inventory management technology encompasses the technical means of optimizing and managing the storage, classification, scheduling, and transportation of goods and materials. Core aspects of this field include real-time monitoring of inventory data, accurate inventory forecasting, automated and intelligent operation of inventory systems, high efficiency in inventory scheduling, and improved warehouse management efficiency. Its technologies cover multiple areas such as the Internet of Things (IoT), intelligent identification technology, and automated logistics equipment, aiming to achieve refined inventory management to meet the needs of modern production and consumption.

[0003] The traditional intelligent storage management method for stone panels refers to the approach taken to address the positioning and management issues caused by differences in specifications, large weights, and large quantities of stone during the storage process. It relies on manual numbering and registration of each piece or identification through barcode labels, combined with manual counting or semi-automatic scanning equipment for data entry and management.

[0004] Traditional stone storage management relies on manual recording or barcode labeling of each piece. In scenarios with large quantities and dense stacking, it is easy to miss markings or mix up information. Manual counting is time-consuming and it is difficult to ensure stable identification when the ambient light changes or the stacking layers are deep. Semi-automatic scanning is often affected by reflection and obstruction, resulting in data loss or positioning deviation. This makes it difficult to update inventory data in a timely manner, which can easily lead to inaccurate location of goods and extended search time. Under high-frequency inbound and outbound conditions, it has an adverse impact on overall management efficiency. Summary of the Invention

[0005] To address the technical problems of traditional stone warehousing management, which relies on manual recording or barcode labeling of each piece, leading to omissions or information confusion in scenarios with large quantities and dense stacking, manual counting is time-consuming and struggles to maintain stable recognition under varying lighting conditions or deep stacking layers, while semi-automated scanning is often affected by reflections and obstructions, resulting in data loss or positioning errors, causing untimely inventory data updates, inaccurate location tracking, and prolonged search times, thus negatively impacting overall management efficiency under high-frequency inbound and outbound conditions, this invention provides a machine vision-based intelligent warehousing management method for stone panels, comprising the following steps:

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a machine vision-based intelligent warehouse management method for stone slabs, comprising the following steps:

[0007] S1: Collect stone surface image data, infrared temperature distribution data, and ultrasonic distance measurement data through multimodal sensors. Perform grayscale processing on the image data, quantization processing on the temperature data, and time difference calculation on the distance data to generate a multimodal sensing feature dataset.

[0008] S2: Time synchronization calibration is performed based on the multimodal sensing feature dataset. The synchronized data is then input into the Kalman filter algorithm for fusion processing. The weight coefficients are dynamically adjusted according to the shelf positioning accuracy and storage space identification to generate comprehensive positioning result data.

[0009] S3: Based on the comprehensive positioning result data, perform reflection coefficient analysis on the stone stacking level and the warehouse coding area, detect the light intensity value, and adjust the light source parameters when the reflection coefficient exceeds the preset reflection threshold to generate an optimized light field parameter set.

[0010] S4: Based on the optimized light field parameter set, re-collect the stone surface texture data, input the texture data into the three-dimensional reconstruction algorithm for model construction, update the surface feature map through error correction calculation, and generate dynamic three-dimensional model data.

[0011] As a further embodiment of the present invention, the multimodal sensing feature dataset includes texture information features, temperature distribution features, and depth information features; the comprehensive positioning result data includes location coordinate information, stacking hierarchy information, and region identification information; the optimized light field parameter set includes illumination intensity parameters, light source direction parameters, and light source frequency parameters; and the dynamic three-dimensional model data includes a geometric structure model, a surface texture model, and an error correction model.

[0012] As a further aspect of the present invention, the specific steps of S1 are as follows:

[0013] S101: After acquiring the image data of the stone surface, the color channel values ​​are converted into a gray value matrix in sequence using the pixel grayscale conversion method. The brightness is re-encoded according to the channel ratio of each pixel in the matrix, and the grayscale discrimination of the pixels is calculated for the brightness difference to generate a set of grayscale pixel values.

[0014] S102: Based on the collected infrared temperature distribution data, call the pixel coordinates in the grayscale pixel value set, discretize the continuous temperature values ​​according to the temperature interval threshold, and superimpose the discrete values ​​at the coordinate positions. Calculate the temperature level by comparing with the interval threshold to obtain the temperature interval value distribution.

[0015] The temperature range threshold is used to discretize the collected continuous temperature values ​​into different level ranges according to preset or adaptive division rules, so as to realize the standardized hierarchical representation of temperature data.

[0016] S103: For ultrasonic distance measurement data, call the coordinate position in the temperature range numerical distribution, calculate the distance value according to the time difference between the transmitted and received signals, and integrate it with the temperature distribution value and grayscale pixel value of the corresponding coordinates to generate a multimodal sensing feature dataset.

[0017] As a further aspect of the present invention, the specific steps of S2 are as follows:

[0018] S201: Based on the multimodal sensing feature dataset, the difference in the detection time records of the sensor output signals under the same time period is recorded, the offset between the sensor recording times is calculated, and the time base is corrected on the data sequence to generate a time synchronization difference sequence.

[0019] S202: Call the time synchronization difference sequence and input it into the Kalman filter algorithm. Weight the residual between the predicted state and the observed value, adjust the state transition matrix and covariance matrix, update the filtering process, and generate the fused positioning estimate.

[0020] S203: Based on the fused positioning estimate, compare it with the shelf positioning accuracy value and the storage space identifier value, adjust the allocation ratio of different weight coefficients in the Kalman filter, update the corresponding weight values ​​of position, velocity and observation signal, and generate comprehensive positioning result data.

[0021] As a further aspect of the present invention, the specific steps of S3 are as follows:

[0022] S301: Based on the comprehensive positioning result data, traverse the stone stacking levels and the warehouse location coding area, compare the spatial coordinates with the area boundary values, extract the stone surface reflection coefficient, and generate a set of area reflection coefficient values.

[0023] S302: Call the set of regional reflectance coefficient values ​​to detect the light intensity of the cargo location coding area, compare the reflectance coefficient values ​​with the set reflectance threshold, filter out areas that exceed the threshold and record the light intensity, and obtain the over-threshold light intensity value.

[0024] The set reflection threshold is used to determine whether the surface reflectance of the area meets the minimum standard of illumination requirements, so as to screen the target area that needs to be optimized.

[0025] S303: Based on the above-threshold illumination intensity value, the brightness and angle parameters of the light source are jointly adjusted, and the allocation is optimized according to the illumination intensity value range. The adjusted light source parameters are then redistributed to the cargo location coding area to generate an optimized light field parameter set.

[0026] As a further aspect of the present invention, the specific steps of S4 are as follows:

[0027] S401: Obtain the optimized light field parameter set, detect the light intensity value and the stone surface reflectivity value under different incident angles, establish the correspondence between the two through polynomial fitting, adjust the light intensity distribution according to the difference, collect stone surface texture image data and compare the pixel matrix with the light intensity value to generate light texture matching coefficient.

[0028] S402: Call the lighting texture matching coefficient, input the texture image pixel matrix into the three-dimensional reconstruction algorithm, calculate the depth difference based on the mapping relationship between gray value and three-dimensional coordinate point position, combine the depth difference to analyze the three-dimensional point cloud distribution, and then aggregate according to coordinate adjacency to obtain the surface point cloud density value.

[0029] S403: Based on the surface point cloud density value, compare the point cloud density with the reconstruction error threshold in the error correction calculation, filter out the out-of-limit points and adjust the three-dimensional coordinate values, update the surface feature map, and generate dynamic three-dimensional model data.

[0030] As a further aspect of the present invention, the light texture matching coefficient refers to the coefficient value calculated from the gray-scale distribution of the pixel matrix of the stone surface texture image and the corresponding light intensity value under specific incident angle and light intensity conditions.

[0031] The surface point cloud density value refers to the ratio of the number of three-dimensional coordinate points determined by the effective point cloud to the area within a unit surface area;

[0032] The reconstruction error threshold refers to the allowable range of the difference between the depth value of the point obtained from the 3D reconstruction and the reference depth value.

[0033] As a further aspect of the present invention, the method further includes step S5:

[0034] S5: Based on the dynamic three-dimensional model data, the location information of the inventory counting unit and the inbound and outbound record table is extracted and calculated. By calculating the location deviation value, the frequency parameter of the dynamic three-dimensional model data update is adjusted according to the location accuracy to generate an inventory management plan.

[0035] The inventory management solution includes location accuracy strategy, inventory scheduling strategy, and inbound / outbound control strategy.

[0036] As a further aspect of the present invention, the specific steps of S5 are as follows:

[0037] S501: Based on the dynamic three-dimensional model data, extract the coordinate information of the inventory counting unit and the inbound / outbound record table, compare the corresponding material numbers, calculate the coordinate difference vector and summarize it into a unified difference set to generate the position deviation value.

[0038] S502: Call the position deviation value, compare the coordinate difference of the inventory unit at different time nodes with the coordinate accuracy benchmark value, calculate the inventory compliance ratio, set the weight according to the proportion of the number of inventory units, perform weighted average analysis to analyze the overall identification and matching degree, and obtain the position accuracy.

[0039] S503: Based on the location accuracy rate and the benchmark value for adjusting the inventory management strategy, determine whether the inventory frequency and the inbound / outbound calibration mechanism need to be adjusted, reset the measures within the range of difference, and generate an inventory management plan.

[0040] As a further aspect of the present invention, the position deviation value is the difference vector between the coordinates of the inventory unit in the dynamic three-dimensional model data and the corresponding coordinates in the inbound and outbound record table;

[0041] The position accuracy rate is the proportion of the number of inventory units whose position deviation is within the allowable range under a given coordinate accuracy benchmark value, after weighting, to the total number of inventory units.

[0042] The benchmark value for adjusting the inventory management strategy is a pre-set threshold value based on the location accuracy.

[0043] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0044] In this invention, the collaborative acquisition and fusion of multi-source sensor data ensures consistency of positioning results across different measurement dimensions, avoiding error propagation caused by environmental interference. Real-time monitoring of light intensity and adjustment of the reflection coefficient optimize the light field, ensuring the integrity and stability of stone texture information during acquisition. The 3D model, through dynamic reconstruction and error correction, can update surface features in real time, guaranteeing synchronous matching between the model and the actual storage state. This continuous data optimization process improves the accuracy of stacking hierarchy and location identification, making positioning and management in complex scenarios more precise, and effectively enhancing the reliability and consistency of inventory information. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a schematic diagram of the steps of the present invention;

[0047] Figure 2 This is a detailed schematic diagram of S1 of the present invention;

[0048] Figure 3This is a detailed schematic diagram of S2 of the present invention;

[0049] Figure 4 This is a detailed schematic diagram of S3 of the present invention;

[0050] Figure 5 This is a detailed schematic diagram of S4 of the present invention;

[0051] Figure 6 This is a detailed schematic diagram of S5 of the present invention. Detailed Implementation

[0052] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0053] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0054] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0055] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0056] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0057] Please see Figure 1 This invention provides a machine vision-based intelligent warehouse management method for stone slabs, comprising the following steps:

[0058] S1: Collect stone surface image data, infrared temperature distribution data, and ultrasonic distance measurement data through multimodal sensors. Perform grayscale processing on the image data, quantization processing on the temperature data, and time difference calculation on the distance data to generate a multimodal sensing feature dataset.

[0059] S2: Time synchronization calibration is performed based on the multimodal sensor feature dataset. The synchronized data is then input into the Kalman filter algorithm for fusion processing. The weight coefficients are dynamically adjusted according to the shelf positioning accuracy and storage space identification to generate comprehensive positioning result data.

[0060] S3: Based on the comprehensive positioning results data, the reflection coefficient of the stone stacking level and the warehouse coding area is analyzed, the light intensity value is detected, and when the reflection coefficient exceeds the preset reflection threshold, the light source parameters are adjusted to generate an optimized light field parameter set.

[0061] S4: Re-collect stone surface texture data based on the optimized light field parameter set, input the texture data into the three-dimensional reconstruction algorithm for model construction, update the surface feature map through error correction calculation, and generate dynamic three-dimensional model data;

[0062] S5: Based on dynamic 3D model data, the location information of inventory counting units and inbound / outbound record tables is extracted and calculated. By calculating the location deviation value, the frequency parameters of dynamic 3D model data updates are adjusted according to the location accuracy to generate an inventory management plan.

[0063] The multimodal sensing feature dataset includes texture information features, temperature distribution features, and depth information features. The comprehensive positioning result data includes location coordinate information, stacking hierarchy information, and region identification information. The optimized light field parameter set includes illumination intensity parameters, light source direction parameters, and light source frequency parameters. The dynamic 3D model data includes geometric structure model, surface texture model, and error correction model. The inventory management scheme includes location accuracy strategy, inventory scheduling strategy, and inbound / outbound control strategy.

[0064] Please see Figure 2 The specific steps of S1 are as follows:

[0065] S101: After acquiring the image data of the stone surface, the color channel values ​​are converted into a gray value matrix in sequence using the pixel grayscale conversion method. The brightness is re-encoded according to the channel ratio of each pixel in the matrix, and the grayscale discrimination of the pixels is calculated for the brightness difference to generate a set of grayscale pixel values.

[0066] After acquiring the image data of the stone surface, the collected color image is decomposed into three independent channels: R (red channel), G (green channel), and B (blue channel). For each channel, the integer values ​​of each pixel within the range of 0 to 255 are read sequentially. The color values ​​are converted to grayscale values ​​by weighting the three channel values ​​of a single pixel. The weights can be set according to the reflectivity of the stone surface under target lighting conditions. For example, in a light-colored granite sample, the weight of red is set to 0.3, the weight of green to 0.59, and the weight of blue to 0.11. Multiplication operations are performed on the three channels of each pixel, and the results are summed to obtain a preliminary grayscale value. Subsequently, brightness re-encoding is performed on each pixel value. This brightness re-encoding is achieved by comparing the preliminary grayscale value with a preset brightness reference value. For example, a brightness reference value is set... When grayscale value When the value is greater than 128, a weighted amplification operation is performed. When grayscale value When the value is less than 128, a weighted reduction operation is performed. The weighting coefficients of 1.1 and 0.9 here are derived from a statistical analysis of the differences in stone brightness under artificial lighting. The statistical method involves selecting 100 sample pixels, measuring the mean, calculating the variance, and selecting the point with the smallest variance as the set value. After the brightness adjustment is completed, a difference calculation is performed on the gray values ​​of adjacent pixel coordinates, and the difference is set to be greater than the brightness contrast threshold. Pixels with significant grayscale differences are considered to be pixels with significant grayscale differences. This threshold is obtained by taking the mean of the ranges of the statistical results of grayscale differences from 10 different types of stone samples. For example, the difference results of adjacent pixels in sample 1... The mean was 15, and the value for sample 2 was 14.6. Therefore, 15 was selected as the standard threshold. After traversing the entire image, the pixel coordinates and their grayscale values ​​determined by brightness differences were added to the grayscale pixel value set and stored in a two-dimensional matrix. Each element in the two-dimensional matrix was... Pixel coordinates and grayscale values ​​are recorded in a format, illustrated with actual sample data:

[0067] Table 1: Grayscale data of sampled pixels on stone surface

[0068]

[0069] As shown in Table 1, the sampling process selected four consecutive horizontal pixels. First, the grayscale values ​​were calculated for each pixel. For example, in the first row, the initial values ​​were R=140, G=150, and B=160. =140×0.3+150×0.59+160×0.11 = 42+88.5+17.6 =148.1 ≈ 148, since =148>128, perform amplification. The difference is then compared with the gray values ​​of adjacent pixels. If the value is less than the threshold of 15, it will not be included in the grayscale set. This process is performed by traversing all pixels in the image to ensure the integrity of the grayscale pixel value set. The final result is a grayscale value matrix organized by coordinates.

[0070] S102: Based on the collected infrared temperature distribution data, call the pixel coordinates in the grayscale pixel value set, discretize the continuous temperature values ​​according to the temperature interval threshold, and superimpose the discrete values ​​at the coordinate positions. Calculate the temperature level by comparing with the interval threshold to obtain the temperature interval value distribution.

[0071] Temperature range thresholds are used to discretize the collected continuous temperature values ​​into different level ranges according to preset or adaptive division rules, so as to achieve standardized hierarchical representation of temperature data;

[0072] Infrared temperature distribution data is obtained. Based on the stored pixel coordinates (x, y) in the grayscale pixel value set, the continuous infrared temperature measurement values ​​at the corresponding locations are retrieved point by point. First, unit conversion and normalization are performed on the temperature data stream collected by the infrared sensor, mapping the temperature values ​​in degrees Celsius (°C) to the [0, 1] interval. The Min-Max normalization method is then used. This is the measured temperature value at that pixel location. and Take the lowest and highest temperatures from the current full-map sampling, for example... , For a pixel at coordinates (101, 200), if the infrared thermometer value is 25.6, then the normalized temperature is... The normalized temperature is then mapped back to a temperature range. The temperature range thresholds are set empirically based on the actual temperature distribution in the storage environment, comprising five ranges: Range 1 is T < 23.5, Range 2 is 23.5 ≤ T < 24.8, Range 3 is 24.8 ≤ T < 26.1, Range 4 is 26.1 ≤ T < 27.4, and Range 5 is T ≥ 27.4. This classification is based on... and The difference is divided into five equal parts and rounded to the nearest 0.1℃. For example, a temperature difference of 6.4℃ divided by 5 gives approximately 1.28℃. Adding this difference to the starting point yields the boundary value. Then, the measured temperature of this pixel is 25.6℃, which, according to the classification interval, belongs to interval 3, and a discretized temperature value of 3 is assigned. The above operations need to be repeated for the temperature value corresponding to each pixel coordinate in the grayscale set; that is, first extracting (x, y), then calling the temperature value at the same location in the sensor's storage array, performing normalization, interval judgment, and number assignment. After numbering and assigning values ​​to the pixel positions, the following steps are performed: The triplet form is superimposed on the grayscale matrix, where For the final discretized values, the analysis of differences between adjacent temperature levels is required in the multi-point continuous determination. If the temperature index difference of a cluster at the same location exceeds 1, it is recorded as a temperature abrupt change point, for example, the temperature difference between adjacent pixel A and B. Pixel B The difference is 2, indicating a mutation. Finally, the discretized matrix of the entire image is checked. The check mechanism is to count the percentage of pixels in the numbered interval. If the percentage of a certain temperature level is less than 1% of the total number of pixels in the entire image, then the level is deleted in the final temperature interval numerical distribution result and its coordinate point temperature is classified into the majority class of the neighboring level. After the above normalization, hierarchical mapping, number assignment and difference check processing, the final temperature interval numerical distribution arranged in pixel coordinates is obtained.

[0073] S103: For ultrasonic distance measurement data, call the coordinate position in the temperature range numerical distribution, calculate the distance value based on the time difference between the transmitted and received signals, and integrate it with the temperature distribution value and grayscale pixel value of the corresponding coordinates to generate a multimodal sensing feature dataset.

[0074] After acquiring ultrasonic distance measurement data, the coordinates (x, y) of the temperature range numerical distribution are retrieved point by point to read the discrete temperature value at that location. With grayscale value Then, the emission time is read from the raw ultrasonic ranging data corresponding to that location. and receiving time First, the time difference is obtained by subtracting the transmission time from the reception time. ,For example , The difference between the two is Then, based on the actual temperature value at that point... Calculate the speed of sound at this temperature by adding a temperature correction to the base speed of sound of 331.4 m / s. Therefore, the speed of sound is 331.4 + (0.6 × 25.6) = 346.76 m / s. Next, multiplying the speed of sound by the time difference gives the round-trip distance of the sound wave: 346.76 × 0.00482 = 1.6694 m. Since this is the round-trip path length, it needs to be divided by 2 to obtain the one-way distance, which is 0.8347 m. This value corresponds to the temperature number at the same coordinate location. and grayscale value Combining to form multimodal triples During the coordinate-by-coordinate traversal of the entire map, the above process needs to be repeated: first, locate the coordinates, then obtain the temperature number and grayscale value, simultaneously read the transmission and reception times and calculate the difference, use temperature to correct the speed of sound, and then combine the time difference to estimate the one-way distance. To avoid introducing outliers, each calculated distance value needs to be judged for its reasonableness within the interval. If the distance is less than the minimum effective stacking spacing of 0.50m set by the warehousing system and greater than zero, it is classified as a valid distance; otherwise, it is marked as an invalid point. After the valid points are calculated, multiple multimodal triples are stored sequentially according to the coordinate index, and multimodal data in different areas will be labeled with the same material category. This yields a complete multimodal sensing feature dataset, containing the visual grayscale, temperature grading information, and spatial distance measurements for each valid pixel.

[0075] Please see Figure 3 The specific steps of S2 are as follows:

[0076] S201: Based on the multimodal sensing feature dataset, the difference in detection time records of sensor output signals within the same time period is calculated, the offset between sensor recording times is calculated, and time base correction is performed on the data sequence to generate a time synchronization difference sequence.

[0077] Based on a multimodal sensing feature dataset, the system first reads the output signal timestamp sequence of each sensor within the same time period, considering multiple types of sensors. Each record in the sequence consists of a timestamp value and its corresponding sensor identifier. For example, a set of timestamp data is read from an infrared temperature sensor, an ultrasonic ranging sensor, and a visual acquisition terminal, and converted into a unified time format accurate to milliseconds. After reading all the data, the system first selects the timestamp sequence of one sensor as the baseline sequence, choosing the timestamp of the visual acquisition unit as the reference. The system then extracts the time values ​​of the corresponding sampling points in the baseline sequence and iterates through the data. The recording times of other sensors at the same sampling point number are subtracted from the reference time value to obtain the time difference. This pairwise subtraction calculation yields the time difference for each corresponding sampling point. For example, for sampling point 1, the timestamp for visual acquisition is 15:32:10.125, and for the infrared temperature sensor it is 15:32:10.140, resulting in a difference of 0.015s. For sampling point 2, the difference is -0.009s, and so on, obtaining the entire set of time differences. The average of this entire set of time differences is then calculated as the overall time offset of the current device. For example, the differences for 10 sampling points are as follows:

[0078] ;

[0079] The differences are summed sequentially to obtain 0.051 seconds, then divided by 10 to get an average offset of 0.0051 seconds. This offset is then used to perform addition or subtraction corrections on the timestamps throughout the entire sequence of this type of sensor. If the offset is positive, it is subtracted from the timestamps of the remaining sensors to align forward to the reference time; if negative, it is added to align backward. This correction generates a new time synchronization sequence. Based on this, the corrected sequence is compared point-by-point with the reference sequence again to calculate a new set of time differences, resulting in a time synchronization difference sequence. This sequence stores the sampling values ​​after the time synchronization reference correction. To ensure data quality, a time deviation threshold needs to be set in the generated time synchronization difference sequence to detect residual asynchrony between the two devices. In this embodiment, the threshold is set to 0.002s. This value is based on the maximum time drift allowed by the system. The drift is obtained by testing the variance stability range of the system under multiple synchronization calibrations. When the absolute value of any difference in the time synchronization difference sequence exceeds 0.002s, it is marked as an abnormal difference and the abnormal point number is recorded. The remaining values ​​are used as normal differences for subsequent processing. The time synchronization difference sequence generated in this way will be provided to subsequent steps for filtering and fusion processing.

[0080] Table 2: Sensor Sampling Time Difference Record Table

[0081]

[0082] As shown in Table 2, the time difference of the sampling points is close to zero after correction, and the residual difference is used as the time synchronization difference sequence to be input into the next step.

[0083] S202: Call the time synchronization difference sequence and input it into the Kalman filter algorithm. Weight the residual between the predicted state and the observed value, adjust the state transition matrix and covariance matrix, update the filtering process, and generate the fused positioning estimate.

[0084] After calling the time synchronization difference sequence, the sequence is first read point by point and used as the initial input data for filtering. Each synchronization difference is then read. and the corresponding sensor state observations at that time. Then, the predicted state vector is initialized. The position value can be set to the estimated position at the previous moment, and the velocity value can be calculated based on the position difference and time interval between adjacent moments. For example, if the estimated position at the previous moment is 3.254m, the current moment is 0.10s apart from the previous moment, and the position difference is 0.023m, then the predicted velocity value is 0.23m / s. The predicted state is used as the initial input to the filtering process. Then, the residual between the predicted state and the current observation value is calculated. The residual is calculated by subtracting the predicted value from the current observation value. For example, if the observed position is 3.261m and the predicted position is 3.254m, then the residual is 0.007m. After obtaining the residual, a weighted processing is required. At this time, the weight coefficient of the current step is called. The residuals are scaled; in this embodiment, the initial settings are as follows: =0.68, this value was determined with reference to the optimal point of residual convergence in multiple offline tests of the system. The test method is to... Increasing from 0.1 to 1.0, the weight value corresponding to the minimum output mean square error is taken as the setpoint. The residual 0.007 × 0.68 is used to obtain a weighted residual of 0.00476m. This weighted residual is then superimposed on the predicted state to obtain the updated state estimate of 3.25876m. At the same time, the coefficients of the state transition matrix are adjusted. The position and velocity transition terms in the state transition matrix are adjusted according to time intervals. For example, the velocity coefficient of the position in the matrix is ​​adjusted from 0.10 in the previous period to the actual time interval of 0.093 in the current period. The covariance matrix needs to be dynamically updated according to the noise level. Its diagonal position variance term is compared with the observed noise variance. With prediction noise variance To correct, if Greater than Then, a scaling factor is added to the corresponding position of the covariance. , here =1.15, derived from the average ratio when noise abruptly changes in the original sample, for example when... =0.0005, When the value is 0.0007, the covariance of the location is updated to 0.0005 × 1.15 = 0.000575. The entire weighted and matrix update process is repeated at each sampling point of the time-synchronized difference sequence, so that each step in the continuous sequence is completed by the prediction state, observation value, residual calculation, weighted correction and matrix update in series. The final fused positioning estimate sequence is the location information sequence after iterative update based on the time-synchronized input.

[0085] S203: Based on the fused positioning estimate, compare it with the shelf positioning accuracy value and the storage space identification value, adjust the allocation ratio of different weight coefficients in the Kalman filter, update the corresponding weight values ​​of position, velocity and observation signal, and generate comprehensive positioning result data;

[0086] Based on the fused positioning estimates, the estimated location values ​​for each time step are read sequentially. and the corresponding shelf positioning accuracy benchmark value The baseline value is a reference value collected by a high-precision laser ranging device at the same shelf location. For example, the accuracy baseline value for a certain shelf location is 3.260m, and its accuracy tolerance threshold is set to ±0.005m. When the absolute value of the difference between the fused positioning result and this reference value is less than or equal to 0.005m, the match is considered normal; otherwise, it is recorded as a deviation point. Then, for each estimated point, the corresponding storage space identifier value is read. This identifier value is a code assigned to a specific storage location by the warehouse management system. For example... This refers to the shelf unit in row 205, column B. The offset between the fused positioning estimate and the reference value is rounded to the nearest millimeter. For example, if the estimated value is 3.25876m, the deviation is -0.00124m, which is rounded to -1mm. This deviation value will be used together with the original deviation data corresponding to the storage space identifier in the multi-weight allocation adjustment calculation. In this adjustment process, the three types of weight coefficients in the Kalman filtering process are first read, namely the position update weight... Speed ​​update weight and the weight of the observed signal In this embodiment, the initial values ​​are 0.68, 0.21, and 0.11, respectively. Then, the adjustment ratio for each type of weight is calculated. This ratio is derived from the ratio of the deviation value to the accuracy tolerance ratio, and then multiplied by the adjustment coefficient. ,Should Based on the statistical analysis of the original stable operating data, this setting... =0.35, for example, the current deviation of -1mm accounts for 0.2 of the tolerance of 5mm, multiplied by If we get 0.07, then The value is adjusted to 0.68 + (0.07 × sign(-1)) = 0.61, where sign(-1) indicates that the deviation direction is negative, thus reducing the position weight. The reduced weight value is then evenly distributed among the... and Above, each category increases by 0.035, new =0.245, =0.145, so the dynamically adjusted set of weight coefficients will replace the original weights in the Kalman filter. The new weights will be applied in the next positioning cycle to update the data, so as to ensure that the weight ratio matches the current positioning accuracy requirements. The points of the fused positioning estimate sequence are traversed, and the above weight adjustment process is repeated. The final generated comprehensive positioning result data is a correspondence between the Kalman filter output after dynamic weight adjustment and the storage space identifier value, ensuring that each output location value matches the corresponding cargo location label one by one.

[0087] Please see Figure 4The specific steps of S3 are as follows:

[0088] S301: Based on the comprehensive positioning result data, traverse the stone stacking levels and cargo location coding areas, compare the spatial coordinates with the area boundary values, extract the stone surface reflection coefficient, and generate a set of area reflection coefficient values.

[0089] To read the comprehensive positioning results data, the dataset is first traversed point by point, and the spatial coordinate values ​​corresponding to each stacking position are extracted sequentially. The system retrieves the location code information and its associated location code, then calls the mapping table between location codes and physical space boundaries in the warehouse database to obtain the boundary range of the location in a three-dimensional coordinate system. The boundary is defined by the minimum coordinate value... with maximum coordinate value Composition, such as storage location code The spatial range is to During the traversal, each coordinate in the comprehensive positioning data is compared axis by axis with the minimum and maximum values ​​of the corresponding area. If all three axes (X, Y, and Z) are within the closed interval of the area, the point is determined to fall into the designated storage location area. Otherwise, the comparison continues with the boundary values ​​of the next area until a match is achieved. After the coordinates and areas are matched, the brightness value L of the stone surface collected by the visual sensor corresponding to that coordinate point, as well as the pre-calibrated incident light intensity, are read. The two factors are then used to calculate the reflectance coefficient. The luminance value is first normalized by dividing the luminance value L by the maximum luminance that the device can collect. Obtain normalized brightness For example, if the brightness at a certain point is 184 and the maximum sampled brightness is 255, then... Then, the reflection coefficient is obtained by dividing the normalized brightness by the normalized value of the incident light intensity, for example, the incident light intensity. =0.85 (relative unit), then the reflectance coefficient is 0.849. The calculation of this reflectance coefficient needs to be repeated for each matched coordinate, and the results are written to the regional reflectance coefficient set in turn. The storage structure of the regional reflectance coefficient set is a list, and each item is a key-value pair of (location code, reflectance coefficient). In addition, in order to exclude outliers, a reasonable range for the reflectance coefficient needs to be set. The normal range of reflectance coefficient obtained in the process of stone surface characteristic calibration is 0.40≤R≤0.95. If the calculated reflectance coefficient exceeds this range, the data of that point is discarded to avoid environmental noise or measurement failure affecting the overall accuracy of the set. After performing the above steps, the final regional reflectance coefficient set will contain each location code and its corresponding reflectance coefficient value.

[0090] Table 3: Sampling Data of Stone Cargo Location Boundary and Reflection Coefficient

[0091]

[0092] As shown in Table 3, the two sampling points were compared at the boundary, normalized for brightness, and calculated for reflectance coefficient, resulting in reflectance coefficient values ​​of 0.849 and 0.799, respectively. The data will be used in the next step for light intensity detection and threshold comparison.

[0093] S302: Call the set of regional reflectance coefficient values ​​to detect the light intensity of the cargo location coding area, compare the reflectance coefficient values ​​with the set reflectance threshold, filter out areas that exceed the threshold and record the light intensity, and obtain the over-threshold light intensity value.

[0094] A reflection threshold is set to determine whether the surface reflectance of a region meets the minimum standard required for illumination, in order to screen the target regions that need to be optimized.

[0095] The system retrieves the set of area reflectance coefficient values, sequentially reads each data item (location code, reflectance coefficient), and uses the location code as the key to retrieve the stored current illuminance measurement value for that location. This value is obtained through the output of a light sensor fixed to the top of the shelf and has been converted into a relative light intensity unit [0,1] interval, such as the storage location. The measured value was 0.83, and the cargo space... The measured value was 0.88. Subsequently, for each storage location, its reflectance R was compared with the set reflectance threshold. For comparison, this threshold value was determined based on the statistical analysis results of reflectance coefficients of the same material under different lighting conditions. After taking the mean of 0.78 from 100 stone samples and adding twice the standard deviation of 0.06 to obtain 0.90, this value was further corrected to 0.85 based on the spectral characteristics of the warehouse lighting fixtures, and thus set as the final threshold. =0.85, when comparing, if If it is, it is determined to be a high-reflectivity area; otherwise, it is a normal area, such as a cargo space. The reflectance coefficient R=0.849 differs from the threshold by -0.001, which does not exceed the threshold and is therefore considered normal. The cargo location... If the reflectance coefficient R = 0.799, the difference is -0.051, which is normal. If the reflectance coefficient of a certain storage location is 0.870, the difference from the threshold is 0.020, and it is judged as an area exceeding the threshold. After completing the reflectance coefficient judgment, the light intensity value of the area exceeding the threshold needs to be further detected. Corresponding to the system's light intensity classification standard, which is classified as: low-illuminance area <0.60, 0.60≤ in the medium illumination zone <0.80, high illuminance area ≥0.80, determine the illumination range so that the light source power can be allocated according to different intensities in subsequent light field parameter adjustments, such as in a cargo location. If R = 0.872 and =0.79, therefore the area is a medium illuminance exceeding the threshold region, and the illuminance value recorded is 0.79, while the cargo location If R = 0.868 and =0.81, then it is determined to be a high-illuminance over-threshold area, and the illuminance value is recorded as 0.81. After traversal, the over-threshold area (location code, R, ...) is then... The data is appended to the over-threshold illumination intensity list. This list only saves the precise values ​​of illumination intensity in the classification of areas that are identified as high reflectivity areas. This data is used to perform allocation optimization processing according to the light intensity conditions of the area when adjusting the brightness and angle of the light source.

[0096] S303: Based on the light intensity value exceeding the threshold, the brightness and angle parameters of the light source are jointly adjusted, and the distribution is optimized according to the range of light intensity values. The adjusted light source parameters are then redistributed to the cargo location coding area to generate an optimized light field parameter set.

[0097] Read the list of light intensities exceeding the threshold, and iterate through each data item (location code, R, ...) in sequence. ), and the light intensity value Extract the data as a reference for light source adjustment, and simultaneously read the original light source brightness parameters corresponding to the location code. (Unit: cd / m) 2 ) and angle parameters (Unit: degrees) The factory-rated luminance of the light source in its initial state is... =1200cd / m 2 And the angle of the light direction is =45°, then according to Calculate the required brightness correction amount within the interval of the light intensity classification. and angle correction amount The brightness correction amount is derived from the difference between the illumination intensity and the standard target intensity. The difference multiplied by the brightness scaling factor ,in The value is set to 0.75 (relative value) to ensure balanced lighting in the middle section; this is the scaling factor. Take 800 cd / m 2 This value comes from the test results of the lighting response curve, such as the location of the storage unit. of =0.79, differing from the target by 0.04, therefore the brightness correction is 0.04 × 800 = 32 cd / m². 2 New brightness parameters =1200-32=1168 cd / m 2 This indicates that brightness needs to be reduced, and the angle correction amount is required. The excess is calculated based on the magnitude of the reflection coefficient R exceeding the threshold, and the excess portion is denoted as... ,coefficient The angle is set to 25°, derived from measured results of the sensitivity of the incident light direction to the intensity of reflection, for example, R=0.872. If the value exceeds 0.022, the corresponding angle correction is 0.022 × 25 ≈ 0.55°. When the light source is too bright and the threshold exceeds 0.015, the adjustment strategy is to reduce the brightness and increase the angle value to deviate from the reflection direction. Conversely, if If the value is too low and the R threshold is only slightly exceeded (between 0 and 0.010), then the brightness is increased and the angle is slightly adjusted to a negative value, for example, in the cargo position. like =0.81, a difference of 0.06 corresponds to a brightness correction of 48 cd / m². 2 Lower, and Corresponding to a 0.45° increasing angle, after traversing the over-threshold region, (location code, , A new light field parameter list is written, which stores all optimized combinations of light source brightness and angle, and is indexed by the cargo location code. Finally, an optimized light field parameter set is generated, which provides a data basis for the subsequent system to control the power output and direction of the light source according to the set.

[0098] Please see Figure 5 The specific steps of S4 are as follows:

[0099] S401: Obtain the optimized light field parameter set, detect the light intensity value and the stone surface reflectance value under different incident angles, establish the correspondence between the two through polynomial fitting, adjust the light intensity distribution according to the difference, collect stone surface texture image data and compare the pixel matrix with the light intensity value to generate light texture matching coefficient.

[0100] Obtain the optimized light field parameter set, and iterate through the entries in the set (location code, ... , ), to adjust the light intensity parameters With angle parameters Applied to a lighting controller, it enables the light source output to operate according to a new set value. Then, under this lighting condition, using angle parameters as input, it controls a stepping rotation structure to set multiple incident angle measurement sequences, selecting... Based on the above, three sets of incident conditions were established: decreasing the angle by 2°, keeping the original value, and increasing the angle by 2°. For each set of conditions, the actual light intensity was obtained using a surface light sensor. (Normalized to the 0-1 range) and the surface reflectance of the stone obtained through the reflectance detection module. This involves creating one-to-one data pairs between light intensity and reflectivity, and establishing a temporary array for recording these pairs, such as storage locations. The three sets of measurement results are as follows: at 43.55°, I=0.79, R=0.842; at 45.55°, I=0.83, R=0.849; and at 47.55°, I=0.78, R=0.846. After establishing the relationship between light intensity and reflectivity, the adjustment amount of light intensity distribution is calculated for the measurement points. The method is to compare the light intensity at each angle with the preset target light intensity. Subtract 0.80 from the sum, then multiply by the proportionality factor. Determine the brightness adjustment value. For example, a difference of 0.03 at 45.55° corresponds to an adjustment value of 27 cd / m². A positive value increases the light intensity, while a negative value decreases it. Simultaneously, fine-tune the incident direction. If the absolute value of the reflectivity difference between two adjacent angles is greater than 0.004, adjust the angle direction by taking the sign of the difference. For example... If the light intensity is less than the negative threshold of -0.004, the initial angle reference for the next measurement will be shifted 0.5° to the left. After adjusting the light intensity distribution, a high-resolution texture camera will be activated to acquire texture images of the stone surface at the location. The acquired images will be quantized into a grayscale matrix by pixels. Each pixel is located at a known spatial coordinate mapping position of the storage location. The pixel grayscale value is normalized and compared with the corresponding coordinate illumination intensity value to calculate the illumination-texture matching coefficient. This coefficient is the mean of the absolute values ​​of the pixel-wise differences between the normalized grayscale value and the normalized illumination intensity value. For example, in the comparison of 200 randomly selected pixels, the cumulative sum of the absolute values ​​of the grayscale and illumination differences is 12.4. The average value is taken to obtain a matching coefficient of 0.062. The same calculation is performed by traversing the sampled data of the storage location, and the final result is written into the illumination-texture matching coefficient set.

[0101] Table 4: Example Table of Light Intensity and Reflectivity Measurement and Matching Coefficient Calculation

[0102]

[0103] As shown in Table 4, the light intensity and reflectivity were measured for three different incident angles at the same storage location, and the light intensity difference and brightness adjustment value were calculated. The light and texture matching coefficient results obtained from the measurement data were used to measure the accuracy of the correspondence between light and texture.

[0104] S402: Call the lighting texture matching coefficient, input the texture image pixel matrix into the 3D reconstruction algorithm, calculate the depth difference based on the mapping relationship between gray values ​​and 3D coordinate points, combine the depth difference to analyze the 3D point cloud distribution, and then aggregate according to coordinate adjacency to obtain the surface point cloud density value.

[0105] To call the lighting texture matching coefficient, first, for each storage location code, call its matching coefficient value. With the acquired texture image pixel matrix The matrix is ​​a two-dimensional array, where each element is a normalized grayscale value and corresponds to the index of the specific spatial location of the cargo location in the three-dimensional scanning coordinate system. The pixel matrix is ​​then used as input into the 3D reconstruction process. During processing, the grayscale value and corresponding spatial coordinates of each pixel are used to call the grayscale-depth mapping table established during light source calibration, and the grayscale value is matched with an approximate depth reference. Then compare the depth values ​​acquired at the same coordinate position in adjacent frames or under different incident angles. By subtracting, we obtain the depth difference. For example, if a pixel's grayscale value of 0.65 corresponds to a reference depth of 2.812m, and the actual measured depth is 2.804m, then the difference is... =-0.008m, record the depth difference of the pixels in the matrix. Next, a combinatorial analysis is performed on the depth difference matrix of the entire texture image. The spatial 3D coordinates of each pixel are added to its depth difference to obtain corrected coordinate points. These points are then combined to form a 3D point cloud dataset of the cargo location surface. Each point in this point cloud dataset carries both location coordinates and grayscale information. After the combination is completed, the coordinate adjacency of the point cloud is calculated. Adjacency is determined by calculating the Euclidean distance between two points. When the distance between any two points is less than the aggregation threshold, the adjacency is determined. In this embodiment, the surfaces are considered to belong to the same local surface structure. The value is 0.006m, derived from the weighted average of the scanning resolution and noise level of the stone surface microstructure. When traversing the point cluster, it is first sorted by X coordinate, and then the three-dimensional distance is calculated with adjacent points in turn. For example, the coordinates of point A are... Coordinates of point B The distance is:

[0106] ;

[0107] Slightly above the threshold, therefore not aggregated; while with the coordinates of another point C The distance is:

[0108] ;

[0109] The points are then aggregated into a local surface unit. After determining the adjacency of all points and clustering them, the number of points within each class is counted. The aggregated count value is the surface point cloud density data. For example, if the surface of a certain cargo location is aggregated to obtain 135 class units with a total of 5400 points, then the average point cloud density is 5400 / 135≈40 points / unit. The cargo location is traversed to generate complete surface point cloud density values, which are used as input data for error correction in the next step.

[0110] S403: Based on the surface point cloud density value, compare the point cloud density with the reconstruction error threshold in the error correction calculation, screen out-of-limit points and adjust the three-dimensional coordinate values, update the surface feature map, and generate dynamic three-dimensional model data.

[0111] To retrieve the surface point cloud density value, first iterate through the point cloud density values ​​of each storage location. and the total number of corresponding aggregated units, this value is compared with the preset reconstruction error threshold. For comparison, the threshold value is determined based on the correlation curve between point cloud density and reconstruction accuracy statistically analyzed in multiple batches of surface texture scanning experiments. In this embodiment, the threshold value is selected as... =35 points / cell, indicating that when the average number of points per local cell is lower than this threshold, the local accuracy of the reconstructed model is insufficient. Less than If the condition is met, the cell is identified as an out-of-limit point; otherwise, it is considered a normal point. Then, for each marked out-of-limit point, all the corresponding 3D coordinates of that point in the point cloud dataset are read. With texture grayscale value Next, the coordinate error correction amount is calculated. The method for determining this point is to calculate the difference between the coordinates of the point and its nearest n=5 neighboring points, that is, first calculate the average coordinates of the neighboring points. Then, subtract the coordinates of the point from each of the three-dimensional correction components to obtain the corrected components in the three-dimensional direction. For example, if the current point coordinates are... The average of its 5 neighboring points is Therefore, the correction amount in the X direction is -0.003m, in the Y direction it is -0.001m, and in the Z direction it is -0.003m. Multiply these three correction components by the direction adjustment coefficient respectively. , , The original coordinates are then added back, where the direction adjustment coefficient is selected based on the scanning system accuracy calibration results. In this embodiment, we take... =0.85, =0.90, =0.80, then the final corrected coordinates are , , The operation is repeated for each out-of-limit point to generate a new set of corrected 3D coordinates. The new coordinates replace the corresponding coordinates in the original point cloud dataset. After the replacement is completed, the point cloud and grayscale information of all cargo locations are reorganized to reconstruct the 3D model structure. The local geometric morphology information is updated in the surface feature data of the model, and dynamic 3D model data is output. This data retains the mapping relationship with the original texture image and can continue to be used in conjunction with the lighting information.

[0112] Please see Figure 6 The specific steps of S5 are as follows:

[0113] S501: Based on dynamic 3D model data, extract coordinate information of inventory counting units and inbound / outbound record tables, compare them according to material number, calculate coordinate difference vector and summarize them into a unified difference set to generate position deviation value;

[0114] When accessing dynamic 3D model data, the system first filters out inventory count units that exist within the model. Each inventory count unit contains a unique material number and its spatial coordinates within the current 3D model. Then, the original coordinates corresponding to the same material number in the inventory inbound / outbound record table are read from the inventory database. The original coordinates are the reference position coordinates recorded by the positioning system after the most recent outbound or inbound operation. Each inventory unit is traversed, and the material number is first used as the query key to compare the number in the dynamic 3D model with that in the outbound / inbound table. If they match, the current coordinates and reference coordinates are extracted for difference calculation. The difference vector is calculated by subtracting the reference coordinates from the current coordinates in multiple axes. For example, the coordinates of a stone slab material number S-009 in the 3D model are... m, the reference coordinate in the inbound / outbound record table is If the difference in the X direction is 12.304 - 12.297 = 0.007m, the difference in the Y direction is -0.005m, and the difference in the Z direction is -0.005m. The triplet of these differences is used as the coordinate difference vector for the material. After calculation, the difference vector is stored in a temporary difference set. The material number is recorded for summary mapping. This operation is repeated until all inventory units are processed. After traversal, the modulus of each difference vector is calculated as the position deviation value. The modulus calculation steps are to sum the squares of the direction differences and then take the square root. For example, for S-009, the modulus is:

[0115] ;

[0116] That is, the deviation value is approximately 9.95mm. This value directly represents the difference between the actual position of the material and the last record in the warehouse. The deviation values ​​corresponding to the material number will be uniformly summarized to form a set of position deviation values.

[0117] Table 5: Example Table for Calculating the Difference Between Inventory Count Unit and Inbound / Outbound Record Locations

[0118]

[0119] As shown in Table 5, the difference vector is obtained by subtracting the current coordinates and reference coordinates of the two inventory counting units along each axis, and the modulus is further calculated as the position deviation value. The deviation values ​​directly constitute the set of position deviation values ​​for use in the next step of analysis.

[0120] S502: Call the position deviation value, compare the coordinate difference of the inventory unit at different time nodes with the coordinate accuracy benchmark value, calculate the inventory compliance ratio, set the weight according to the proportion of the number of inventory units, perform weighted average analysis to analyze the overall identification and matching degree, and obtain the position accuracy.

[0121] To retrieve the set of position deviation values, first read all material numbers and their corresponding deviation values ​​from the set. At the same time, it calls the coordinate precision reference value set in the configuration. This benchmark value is derived from the maximum permissible error threshold of the scanning and positioning equipment and the warehousing operation specifications. This embodiment sets... =0.010m indicates that when the position deviation is less than or equal to this value, the inventory unit is considered to meet the accuracy requirements. Then, each inventory unit in the set is iterated through, and its deviation value is compared with the accuracy benchmark value. If the deviation is 0, the material is counted as "compliant"; otherwise, it is counted as "non-compliant". For example, in Table 5, the deviation value of S-009 is 0.00995m, and the difference from the benchmark is -0.00005m, which is compliant. The deviation value of S-010 is 0.00640m, and the difference from the benchmark is -0.00360m, which is also compliant. The entire set is then iterated to count the number of compliant materials. With total material quantity Calculate the inventory compliance rate The steps involve dividing the number of matching materials by the total number of materials. For example, when there are 200 inventory units and 184 matching materials, the ratio is 184 / 200 = 0.92. After calculating this ratio, a weight is set based on the proportion of inventory unit quantity. The matching ratio of different types of areas or different levels of importance of storage locations is adjusted by weighting. In this embodiment, the weighting is based on three categories of zones: high-priority storage location weight. Medium priority cargo location weight Low-priority storage location weight The weights are derived from the average turnover rate of this type of storage location in warehouse operations. The reasonableness of the weights is calculated based on the annual inbound and outbound statistics. For example, if the matching ratio for high-priority storage locations is 0.95, for medium-priority locations it is 0.90, and for low-priority locations it is 0.88, then the weighted average overall identification and matching degree is... Calculated as First, calculate the weighted values ​​of the high-optimal part (0.475), the medium-optimal part (0.27), and the low-optimal part (0.176) separately, and then add them together to get 0.912. This result is the position accuracy, accurate to within ±0.001. Finally, output the position accuracy as the overall coordinate accuracy performance index for the current inventory cycle.

[0122] S503: Based on the location accuracy and inventory management strategy, adjust the benchmark value to determine whether the inventory frequency and inbound / outbound calibration mechanism need to be adjusted, reset the measures within the range of difference, and generate an inventory management plan.

[0123] To retrieve the location accuracy rate, first read the baseline accuracy value set in the inventory management strategy. This value is derived from the minimum guarantee requirement for inventory accuracy by the warehousing operations department. This embodiment sets... This indicates that when the actual location accuracy is lower than this value, the inventory and inbound / outbound management mechanism needs to be adjusted, and then the difference between the current accuracy and the benchmark value is calculated. For example, the accuracy of the location obtained in the previous step is The difference is 0.921 - 0.930 = -0.009. A negative difference indicates that the current accuracy is lower than the baseline, triggering the adjustment strategy process. First, the inventory counting frequency is determined by reading the current inventory counting frequency. (Unit: times / week), compared with the strategy reference frequency In comparison, the reference frequency in this example is 1.0 times / week. If its absolute value is greater than 0.005, the inventory frequency will be increased by one frequency step. The step size is derived from the statistical median of the original adjustment records; in this example, it is set to 0.5 times / week. Therefore, the new frequency... The two are added together; for example, if the current frequency is 1.0 times / week, it is adjusted to 1.5 times / week. At the same time, the sampling ratio of the inbound and outbound calibration mechanism is evaluated. This percentage represents the proportion of materials sampled in each inbound / outbound operation. In this example, the initial value is 15%, referencing the benchmark percentage. If the accuracy rate is more than 0.010 lower than the benchmark value, then use the standard directly. If the value is lower than the baseline but the range is between 0.005 and 0.010, the sampling ratio is increased linearly. That is, the new sampling ratio equals the current sampling ratio plus the product of the coefficient k and the absolute value of the position accuracy difference |ΔP|. The coefficient k comes from the correlation experiment between the pass rate and the sampling rate; in this example, k = 2.5 (unit: % / 0.001 accuracy difference). Therefore, for |ΔP| = 0.009, the sampling ratio increases by 0.009 × 2.5 ≈ 0.0225, which is an increase of 2.25%. After completing the adjustment calculations for frequency and sampling ratio, the measures within the difference range are reset. The steps are as follows: In the inventory management system parameter table, update the inventory frequency field to... Update the inbound / outbound sampling ratio field to The changes are recorded in the strategy adjustment log, including the values ​​before and after the adjustment, the threshold conditions that triggered the adjustment, the execution timestamp, and the operator identification code. Finally, the new inventory management strategy file is exported as a data object with version number incrementing sequentially, and a new inventory management plan is generated corresponding to this adjustment.

[0124] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A machine vision-based intelligent warehouse management method for stone slabs, characterized in that, Includes the following steps: S1: Collect stone surface image data, infrared temperature distribution data, and ultrasonic distance measurement data through multimodal sensors. Perform grayscale processing on the image data, quantization processing on the temperature data, and time difference calculation on the distance data to generate a multimodal sensing feature dataset. S2: Time synchronization calibration is performed based on the multimodal sensing feature dataset. The synchronized data is then input into the Kalman filter algorithm for fusion processing. The weight coefficients are dynamically adjusted according to the shelf positioning accuracy and storage space identification to generate comprehensive positioning result data. S3: Based on the comprehensive positioning result data, perform reflection coefficient analysis on the stone stacking level and the warehouse coding area, detect the light intensity value, and adjust the light source parameters when the reflection coefficient exceeds the preset reflection threshold to generate an optimized light field parameter set. S4: Re-collect stone surface texture data according to the optimized light field parameter set, input the texture data into the three-dimensional reconstruction algorithm for model construction, update the surface feature map through error correction calculation, and generate dynamic three-dimensional model data; S5: Based on the dynamic three-dimensional model data, the location information of the inventory counting unit and the inbound and outbound record table is extracted and calculated. By calculating the location deviation value, the frequency parameter of the dynamic three-dimensional model data update is adjusted according to the location accuracy to generate an inventory management plan. The inventory management solution includes location accuracy strategy, inventory scheduling strategy, and inbound / outbound control strategy. The specific steps of S5 are as follows: S501: Based on the dynamic three-dimensional model data, extract the coordinate information of the inventory counting unit and the inbound / outbound record table, compare the corresponding material numbers, calculate the coordinate difference vector and summarize it into a unified difference set to generate the position deviation value. S502: Call the position deviation value, compare the coordinate difference of the inventory unit at different time nodes with the coordinate accuracy benchmark value, calculate the inventory compliance ratio, set the weight according to the proportion of the number of inventory units, perform weighted average analysis to analyze the overall identification and matching degree, and obtain the position accuracy. S503: Based on the location accuracy rate and the benchmark value for adjusting the inventory management strategy, determine whether the inventory frequency and the inbound / outbound calibration mechanism need to be adjusted, reset the measures within the range of difference, and generate an inventory management plan.

2. The intelligent warehouse management method for stone slabs based on machine vision according to claim 1, characterized in that, The multimodal sensing feature dataset includes texture information features, temperature distribution features, and depth information features. The comprehensive positioning result data includes location coordinate information, stacking hierarchy information, and region identification information. The optimized light field parameter set includes illumination intensity parameters, light source direction parameters, and light source frequency parameters. The dynamic three-dimensional model data includes a geometric structure model, a surface texture model, and an error correction model.

3. The intelligent warehouse management method for stone slabs based on machine vision according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: After acquiring the image data of the stone surface, the color channel values ​​are converted into a gray value matrix in sequence using the pixel grayscale conversion method. The brightness is re-encoded according to the channel ratio of each pixel in the matrix, and the grayscale discrimination of the pixels is calculated for the brightness difference to generate a set of grayscale pixel values. S102: Based on the collected infrared temperature distribution data, call the pixel coordinates in the grayscale pixel value set, discretize the continuous temperature values ​​according to the temperature interval threshold, and superimpose the discrete values ​​at the coordinate positions. Calculate the temperature level by comparing with the interval threshold to obtain the temperature interval value distribution. S103: For ultrasonic distance measurement data, call the coordinate position in the temperature range numerical distribution, calculate the distance value according to the time difference between the transmitted and received signals, and integrate it with the temperature distribution value and grayscale pixel value of the corresponding coordinates to generate a multimodal sensing feature dataset.

4. The intelligent warehouse management method for stone slabs based on machine vision according to claim 3, characterized in that, The specific steps of S2 are as follows: S201: Based on the multimodal sensing feature dataset, the difference in the detection time records of the sensor output signals under the same time period is recorded, the offset between the sensor recording times is calculated, and the time base is corrected on the data sequence to generate a time synchronization difference sequence. S202: Call the time synchronization difference sequence and input it into the Kalman filter algorithm. Weight the residual between the predicted state and the observed value, adjust the state transition matrix and covariance matrix, update the filtering process, and generate the fused positioning estimate. S203: Based on the fused positioning estimate, compare it with the shelf positioning accuracy value and the storage space identifier value, adjust the allocation ratio of different weight coefficients in the Kalman filter, update the corresponding weight values ​​of position, velocity and observation signal, and generate comprehensive positioning result data.

5. The intelligent warehouse management method for stone slabs based on machine vision according to claim 4, characterized in that, The specific steps for S3 are as follows: S301: Based on the comprehensive positioning result data, traverse the stone stacking levels and the warehouse location coding area, compare the spatial coordinates with the area boundary values, extract the stone surface reflection coefficient, and generate a set of area reflection coefficient values. S302: Call the set of regional reflectance coefficient values ​​to detect the light intensity of the cargo location coding area, compare the reflectance coefficient values ​​with the set reflectance threshold, filter out areas that exceed the threshold and record the light intensity, and obtain the over-threshold light intensity value. S303: Based on the above-threshold illumination intensity value, the brightness and angle parameters of the light source are jointly adjusted, and the allocation is optimized according to the illumination intensity value range. The adjusted light source parameters are then redistributed to the cargo location coding area to generate an optimized light field parameter set.

6. The intelligent warehouse management method for stone slabs based on machine vision according to claim 5, characterized in that, The specific steps of S4 are as follows: S401: Obtain the optimized light field parameter set, detect the light intensity value and the stone surface reflectivity value under different incident angles, establish the correspondence between the two through polynomial fitting, adjust the light intensity distribution according to the difference, collect stone surface texture image data and compare the pixel matrix with the light intensity value to generate light texture matching coefficient. S402: Call the lighting texture matching coefficient, input the texture image pixel matrix into the three-dimensional reconstruction algorithm, calculate the depth difference based on the mapping relationship between gray value and three-dimensional coordinate point position, combine the depth difference to analyze the three-dimensional point cloud distribution, and then aggregate according to coordinate adjacency to obtain the surface point cloud density value. S403: Based on the surface point cloud density value, compare the point cloud density with the reconstruction error threshold in the error correction calculation, filter out the out-of-limit points and adjust the three-dimensional coordinate values, update the surface feature map, and generate dynamic three-dimensional model data.

7. The intelligent warehouse management method for stone slabs based on machine vision according to claim 6, characterized in that, The light texture matching coefficient refers to the coefficient value calculated from the gray-level distribution of the pixel matrix of the stone surface texture image and the corresponding light intensity value under specific incident angle and light intensity conditions. The surface point cloud density value refers to the ratio of the number of three-dimensional coordinate points determined by the effective point cloud to the area within a unit surface area; The reconstruction error threshold refers to the allowable range of the difference between the depth value of the point obtained from the 3D reconstruction and the reference depth value.

8. The intelligent warehouse management method for stone slabs based on machine vision according to claim 1, characterized in that, The position deviation value is the difference vector between the coordinates of the inventory unit in the dynamic three-dimensional model data and the corresponding coordinates in the inbound and outbound record table; The position accuracy rate is the proportion of the number of inventory units whose position deviation is within the allowable range under a given coordinate accuracy benchmark value, after weighting, to the total number of inventory units. The benchmark value for adjusting the inventory management strategy is a pre-set threshold value based on the location accuracy.

Images