Intelligent positioning method and system of reference seat

Abstract

This invention provides an intelligent positioning method and system for a reference pedestal. The method includes: acquiring a digital image of a region to be positioned, containing the reference pedestal, and simultaneously acquiring corresponding operational data, which includes multiple continuous attribute values; extracting feature points from the digital image and clustering the feature points to obtain multiple visual feature clusters; selecting representative visual feature clusters from these clusters; determining a set of candidate feature points for each representative visual feature cluster; obtaining an operational encoding vector based on the continuous attribute values ​​in the operational data; obtaining a multimodal feature set based on the candidate feature point set and the operational encoding vector; and completing the positioning of the reference pedestal based on the multimodal feature set and a pre-stored multimodal template feature set. This invention improves the accuracy and adaptability of reference pedestal positioning while increasing computational efficiency, meeting the positioning needs of complex industrial scenarios.

Description

Technical Field

[0001] This invention relates to the field of positioning technology, and more specifically, to an intelligent positioning method and system for a reference pedestal. Background Technology

[0002] In the field of precision assembly manufacturing, accurate identification and positioning of the reference base is crucial to ensuring the quality of subsequent assembly. Traditional positioning methods mostly rely on mechanical positioning or visual matching based on fixed templates, which are difficult to adapt to fluctuations in the actual position of the workpiece caused by conveyor line errors, workpiece placement deviations, and environmental factors. Furthermore, existing methods largely process visual information in isolation, failing to effectively integrate real-time operating data, thus limiting their positioning accuracy and robustness in dynamically changing environments. Therefore, there is an urgent need for an intelligent positioning method that can integrate multi-source information and adapt to environmental changes to meet the high-precision and high-efficiency positioning requirements of modern intelligent manufacturing. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent positioning method and system for a reference pedestal to improve the above-mentioned problems.

[0004] To achieve the above objectives, this application provides the following technical solution:

[0005] On one hand, embodiments of this application provide an intelligent positioning method for a reference pedestal, the method comprising:

[0006] Acquire a digital image containing a reference base, and simultaneously acquire the corresponding operating condition data, which includes multiple consecutive attribute values.

[0007] Feature points are extracted from digital images and clustered to obtain multiple visual feature clusters. Representative visual feature clusters are selected from the visual feature clusters. A set of candidate feature points corresponding to each representative visual feature cluster is determined.

[0008] The working condition encoding vector is obtained based on the continuous attribute values ​​in the working condition data; a multimodal feature set is obtained based on the candidate feature point set and the working condition encoding vector; and the positioning of the reference pedestal is completed based on the multimodal feature set and the pre-stored multimodal template feature set.

[0009] Furthermore, feature points are extracted from the digital image and clustered to obtain multiple visual feature clusters. Representative visual feature clusters are selected from these clusters, including:

[0010] The digital image is sequentially processed by grayscale conversion, Gaussian filtering for noise reduction, and histogram equalization to obtain a preprocessed digital image. The ORB algorithm is then used to detect feature points in the preprocessed digital image, and a corresponding descriptor is calculated for each feature point.

[0011] The average linkage clustering algorithm is used to calculate the Hamming distance between any two descriptors and convert it into similarity. Feature points with similarity greater than a preset first similarity threshold are iteratively clustered and merged until the maximum similarity between all clusters is less than the preset first similarity threshold, resulting in multiple visual feature clusters. The number of feature points contained in each visual feature cluster is calculated, and the top N visual feature clusters are selected as representative visual feature clusters in descending order of the number of feature points.

[0012] Furthermore, the set of candidate feature points corresponding to each representative visual feature cluster is determined, including:

[0013] Feature points other than those representing visual feature clusters are denoted as residual feature points. For each residual feature point, the similarity between it and all feature points in each representative visual feature cluster is calculated, and the maximum similarity is taken as the inter-class similarity between the residual feature point and each representative visual feature cluster.

[0014] For each representative visual feature cluster, the remaining feature points whose inter-class similarity is between a preset second similarity threshold and a preset first similarity threshold are selected to form a candidate feature point set corresponding to the representative visual feature cluster. The preset second similarity threshold is less than the preset first similarity threshold.

[0015] Furthermore, based on continuous attribute values ​​in the operating condition data, an operating condition coding vector is obtained, including:

[0016] Randomly select an attribute value from all consecutive attribute values ​​as the first initial cluster center; set the required total number of initial cluster centers K, and repeat the following process until K initial cluster centers are selected, where K is greater than 1:

[0017] Using the latest selected initial cluster center as the baseline value, calculate the absolute difference between each attribute value and the baseline value, and record it as the current distance; square all current distance values ​​and sum them to obtain the sum of squared values; compare the square of each current distance with the sum of squared values ​​to obtain the probability ratio of that attribute value being selected as the next initial cluster center; select the attribute value with the largest probability ratio as the new initial cluster center;

[0018] The operating condition coding vector is determined based on the obtained K initial cluster centers.

[0019] Furthermore, the working condition coding vector is determined based on the obtained K initial cluster centers, including:

[0020] After obtaining K initial cluster centers, the K-means algorithm is used for clustering to obtain clusters. The mean of all attribute values ​​in each cluster is calculated, and the clusters are sorted in ascending order of mean. After sorting, the maximum and minimum values ​​of attribute values ​​in each cluster are determined. After sorting, the boundary value corresponding to two adjacent clusters is calculated. The boundary value is equal to the sum of the maximum value of the previous cluster and the minimum value of the next cluster, divided by two.

[0021] Based on the calculated K-1 boundary values, the range of continuous attribute values ​​is divided into K continuous and non-overlapping numerical intervals. The first interval contains the minimum value to the minimum boundary value among all attribute values, the last interval contains the maximum boundary value to the maximum value among all attribute values, and the intermediate intervals correspond to the numerical range between adjacent boundary values. Each numerical interval is matched with an operating category. The mean of all attribute values ​​is calculated, and the operating category corresponding to the operating condition data is determined according to the numerical interval in which the mean is located. The operating category is converted into a one-hot encoded vector, and this one-hot encoded vector is used as the operating condition encoded vector.

[0022] Secondly, this application provides an intelligent positioning system for a reference pedestal, the system comprising:

[0023] The acquisition module is used to acquire a digital image containing a reference base, and at the same time acquire the corresponding working condition data of the digital image, which includes multiple continuous attribute values.

[0024] The clustering module is used to extract feature points from digital images, cluster the feature points to obtain multiple visual feature clusters, select representative visual feature clusters from the visual feature clusters, and determine the candidate feature point set corresponding to each representative visual feature cluster.

[0025] The positioning module is used to obtain the working condition encoding vector based on the continuous attribute values ​​in the working condition data; based on the candidate feature point set and the working condition encoding vector, a multimodal feature set is obtained; and based on the multimodal feature set and the pre-stored multimodal template feature set, the positioning of the reference pedestal is completed.

[0026] Furthermore, the clustering module includes:

[0027] The processing unit is used to sequentially perform grayscale conversion, Gaussian filtering for noise reduction, and histogram equalization on the digital image to obtain a preprocessed digital image; the ORB algorithm is used to detect feature points in the preprocessed digital image, and the corresponding descriptor is calculated for each feature point;

[0028] The clustering unit is used to calculate the Hamming distance between any two descriptors using the average chaining clustering algorithm and convert it into similarity. Feature points with similarity greater than a preset first similarity threshold are iteratively clustered and merged until the maximum similarity between all clusters is less than the preset first similarity threshold, resulting in multiple visual feature clusters. The number of feature points contained in each visual feature cluster is calculated, and the top N visual feature clusters are selected as representative visual feature clusters in descending order of the number of feature points.

[0029] Furthermore, the clustering module includes:

[0030] The first calculation unit is used to record feature points other than those representing visual feature clusters as remaining feature points; for each remaining feature point, it calculates the similarity between it and all feature points in each representative visual feature cluster, and takes the maximum similarity as the inter-class similarity between the remaining feature point and each representative visual feature cluster.

[0031] The filtering unit is used to filter out the remaining feature points whose inter-class similarity is between a preset second similarity threshold and a preset first similarity threshold for each representative visual feature cluster, thereby forming a candidate feature point set corresponding to the representative visual feature cluster. The preset second similarity threshold is less than the preset first similarity threshold.

[0032] Furthermore, the positioning module includes:

[0033] Select a cell to randomly choose an attribute value from all consecutive attribute values ​​as the first initial cluster center; set the required total number of initial cluster centers K, and repeat the following process until K initial cluster centers are selected, where K is greater than 1:

[0034] Using the latest selected initial cluster center as the baseline value, calculate the absolute difference between each attribute value and the baseline value, and record it as the current distance; square all current distance values ​​and sum them to obtain the sum of squared values; compare the square of each current distance with the sum of squared values ​​to obtain the probability ratio of that attribute value being selected as the next initial cluster center; select the attribute value with the largest probability ratio as the new initial cluster center;

[0035] The determination unit is used to determine the working condition coding vector based on the obtained K initial cluster centers.

[0036] Further, the defined units include:

[0037] The second calculation unit is used to perform clustering using the K-means algorithm after obtaining K initial cluster centers, to obtain clusters, calculate the mean of all attribute values ​​in each cluster, sort the clusters in ascending order of mean, and determine the maximum and minimum attribute values ​​in each cluster after sorting; after sorting, calculate the boundary value corresponding to two adjacent clusters, the boundary value is equal to the sum of the maximum value of the previous cluster and the minimum value of the next cluster, divided by two;

[0038] The conversion unit is used to divide the range of continuous attribute values ​​into K consecutive and non-overlapping numerical intervals based on the calculated K-1 boundary values. The first interval contains the minimum value to the minimum boundary value among all attribute values, the last interval contains the maximum boundary value to the maximum value among all attribute values, and the intermediate intervals correspond to the numerical range between adjacent boundary values. Each numerical interval is matched with an operating category. The mean of all attribute values ​​is calculated, and the operating category corresponding to the operating condition data is determined according to the numerical interval in which the mean is located. The operating category is converted into a one-hot encoded vector, and the one-hot encoded vector is used as the operating condition encoded vector.

[0039] Thirdly, this application provides an intelligent positioning device for a reference pedestal, the device comprising a memory and a processor. The memory stores a computer program; the processor executes the computer program to implement the steps of the intelligent positioning method for the reference pedestal described above.

[0040] Fourthly, this application provides a readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described intelligent positioning method for a reference pedestal.

[0041] The beneficial effects of this invention are as follows:

[0042] 1. This invention employs a two-level screening mechanism for visual features to effectively distinguish and retain highly stable representative features and moderately variable candidate features, constructing a multimodal feature set that combines discriminative power and robustness. This method ensures that a sufficient number of reliable matching points can be obtained even in complex environments. Simultaneously, the offline precisely established calibration benchmark template provides a high-precision spatial reference for online positioning, resulting in a significant improvement in final positioning accuracy compared to traditional methods, meeting the stringent requirements of high-precision assembly.

[0043] 2. This invention discretizes continuous operating condition data, such as ambient temperature, into operating condition codes with clear physical meaning through a probability-guided clustering algorithm, and then fuses these codes with visual features to form a multimodal descriptor. This allows the feature matching process to not only see the image but also perceive the environmental state. When operating conditions change (such as temperature fluctuations), the system can adaptively adjust the matching strategy and confidence threshold based on the differences in operating condition codes, thereby maintaining high matching accuracy and positioning reliability under different environmental conditions and solving the performance degradation problem of traditional methods when the environment changes.

[0044] 3. By intelligently clustering and filtering visual features, the number of feature points involved in the final matching is reduced, thereby significantly reducing computational complexity, shortening positioning response time, and meeting the cycle time requirements of high-speed production lines. Simultaneously, the combination of offline template creation and online rapid matching ensures both baseline accuracy and real-time efficient computation.

[0045] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing embodiments of the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0046] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a schematic flowchart of the intelligent positioning method for the reference pedestal described in this embodiment of the invention;

[0048] Figure 2 This is a schematic diagram of the intelligent positioning system structure of the reference pedestal described in this embodiment of the invention;

[0049] Figure 3 This is a schematic diagram of the intelligent positioning device structure of the reference base described in this embodiment of the invention. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0051] It should be noted that similar reference numerals or letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0052] Example 1

[0053] like Figure 1 As shown, this embodiment provides an intelligent positioning method for a reference pedestal, which includes steps S1, S2 and S3.

[0054] The process includes step S1: acquiring a digital image containing a reference base, and simultaneously acquiring the corresponding working condition data, which includes multiple consecutive attribute values.

[0055] In this step, acquiring a digital image containing the reference base can be understood as follows: The conveying system (such as a robotic arm) delivers the reference base to a pre-defined assembly station on the assembly platform. The assembly station is an area; the conveying system simply places the reference base into this area for coarse positioning. After a pre-defined delay, a camera mounted directly above the assembly station is triggered to capture an image of the reference base, obtaining a digital image for subsequent precise positioning calculations. The camera's installation position must ensure that the assembly station is within the camera's field of view, and the camera is installed with its optical axis perpendicular to the assembly station.

[0056] In this step, operating condition data refers to the set of continuous or discrete parameters collected within the same or relevant time window as image acquisition, reflecting the system's operating status (e.g., spindle motor current / voltage) or environmental conditions (e.g., light source brightness, ambient temperature). This data is not the visual information of the image itself, but rather auxiliary measurement values ​​directly related to the physical operation of the equipment. Specifically, in this step, ambient temperature data for the 10 minutes prior to image acquisition is synchronously collected from the environmental monitoring system using a temperature sensor. The sampling frequency is 1Hz, resulting in 600 continuous sampled values, forming a continuous attribute value set T={t1,t2,…,t…}. 600}

[0057] Step S2: Extract feature points from the digital image and cluster the feature points to obtain multiple visual feature clusters. Select a representative visual feature cluster from the visual feature clusters and determine the candidate feature point set corresponding to each representative visual feature cluster.

[0058] In this step, feature points are extracted from the digital image and clustered to obtain multiple visual feature clusters. The specific implementation steps for selecting representative visual feature clusters from the visual feature clusters include steps S21 and S22.

[0059] Step S21: Perform grayscale conversion, Gaussian filtering for noise reduction, and histogram equalization on the digital image in sequence to obtain a preprocessed digital image; use the ORB algorithm to detect feature points in the preprocessed digital image and calculate the corresponding descriptor for each feature point.

[0060] In this step, the acquired digital images are processed sequentially as follows:

[0061] Grayscale conversion: Converts an RGB image to an 8-bit grayscale image;

[0062] Gaussian filtering: Noise reduction is performed using a 5×5 Gaussian kernel;

[0063] Histogram equalization: The CLAHE algorithm is used to enhance contrast;

[0064] After preprocessing, the ORB (Oriented FAST and Rotated BRIEF) algorithm is used to detect feature points and calculate descriptors on the preprocessed image. The FAST corner detection threshold can be set to 20, constructing a 4-layer image pyramid. For each detected feature point, its principal direction is calculated, and within a 31×31 pixel neighborhood, the grayscale values ​​of the sampled points are compared using a rotated 256-bit algorithm to generate a 256-bit binary descriptor. Simultaneously, the pixel coordinates of each feature point in the image are recorded.

[0065] Step S22: Using the average linkage clustering algorithm, calculate the Hamming distance between any two descriptors and convert it into similarity; iteratively cluster and merge feature points with similarity greater than a preset first similarity threshold until the maximum similarity between all clusters is less than the preset first similarity threshold, resulting in multiple visual feature clusters; calculate the number of feature points contained in each visual feature cluster, and select the first N visual feature clusters as representative visual feature clusters in descending order of the number of feature points.

[0066] This step can be understood as follows: using the average linkage clustering algorithm, calculate the Hamming distance between feature point descriptors and convert it into similarity: similarity = 1 - Hamming distance / 256; N can be a positive integer greater than 1;

[0067] Suppose we set a preset first similarity threshold T1 = 0.85, and iteratively merge feature points with similarity greater than T1 until the maximum similarity between all clusters is less than T1, resulting in 5 visual feature clusters. We then count the number of feature points contained in each cluster and select the top N (e.g., 4) clusters as representative visual feature clusters.

[0068] In step S2, the set of candidate feature points corresponding to each representative visual feature cluster is determined, including steps S23 and S24;

[0069] Step S23: Record the feature points other than the clusters representing visual features as the remaining feature points; for each remaining feature point, calculate the similarity between it and all feature points in each cluster representing visual features, and take the maximum similarity as the inter-class similarity between the remaining feature point and each cluster representing visual features.

[0070] This step can be understood as follows: For the remaining feature points that do not belong to any representative visual feature cluster, calculate the similarity between each remaining feature point and all feature points in each representative visual feature cluster, and take the maximum value as the inter-class similarity.

[0071] Step S24: For each representative visual feature cluster, select the remaining feature points whose inter-class similarity is between a preset second similarity threshold and a preset first similarity threshold, and form a candidate feature point set corresponding to the representative visual feature cluster. The preset second similarity threshold is less than the preset first similarity threshold.

[0072] This step can be understood as follows: Assuming a preset second similarity threshold T2 = 0.80, for each representative visual feature cluster, filter the remaining feature points whose inter-class similarity falls within the (T2, T1) interval. Finally:

[0073] Cluster 1 representing visual features: 42 remaining feature points were selected;

[0074] Cluster 2 representing visual features: 38 remaining feature points were selected;

[0075] Cluster 3 representing visual features: 45 remaining feature points were selected;

[0076] Representing visual feature cluster 4: 40 remaining feature points were selected;

[0077] Step S3: Obtain the working condition encoding vector based on the continuous attribute values ​​in the working condition data; obtain the multimodal feature set based on the candidate feature point set and the working condition encoding vector; and complete the positioning of the reference pedestal based on the multimodal feature set and the pre-stored multimodal template feature set.

[0078] In this step, the specific implementation steps for obtaining the working condition coding vector based on the continuous attribute values ​​in the working condition data include steps S31 and S32.

[0079] Step S31: Randomly select an attribute value from all consecutive attribute values ​​as the first initial cluster center; set the required total number of initial cluster centers K, and repeat the following process until K initial cluster centers are selected, where K is greater than 1:

[0080] Using the latest selected initial cluster center as the baseline value, calculate the absolute difference between each attribute value and the baseline value, and record it as the current distance; square all current distance values ​​and sum them to obtain the sum of squared values; compare the square of each current distance with the sum of squared values ​​to obtain the probability ratio of that attribute value being selected as the next initial cluster center; select the attribute value with the largest probability ratio as the new initial cluster center;

[0081] In this step, K equals the total number of running categories. For example, in this step, K is set to 4, corresponding to four temperature categories: low temperature, normal, high temperature, and high temperature.

[0082] Step S32: Determine the working condition coding vector based on the obtained K initial cluster centers.

[0083] The specific implementation steps in this step include step S321;

[0084] Step S321: After obtaining K initial cluster centers, the K-means algorithm is used to perform clustering to obtain clusters. The mean of all attribute values ​​in each cluster is calculated, and the clusters are sorted in ascending order of mean. After sorting, the maximum and minimum values ​​of attribute values ​​in each cluster are determined. After sorting, the boundary value corresponding to two adjacent clusters is calculated. The boundary value is equal to the sum of the maximum value of the previous cluster and the minimum value of the next cluster, divided by two.

[0085] Step S322: Based on the calculated K-1 boundary values, divide the range of continuous attribute values ​​into K continuous and non-overlapping numerical intervals. The first interval contains the minimum value to the minimum boundary value among all attribute values, the last interval contains the maximum boundary value to the maximum value among all attribute values, and the intermediate intervals correspond to the numerical range between adjacent boundary values. Match each numerical interval with an operating category. Calculate the mean of all attribute values, determine the operating category corresponding to the operating condition data based on the numerical interval in which the mean is located, convert the operating category into a one-hot encoded vector, and use this one-hot encoded vector as the operating condition encoded vector.

[0086] This step can be understood as:

[0087] If K=4, three boundary values ​​are calculated: b1=24.0, b2=25.5, b3=26.7. The range of the temperature measurement value is [22.8, 28.2].

[0088] According to the above classification rules:

[0089] The first interval is [22.8, 24.0), which matches the low-temperature operation category;

[0090] The second interval: [24.0, 25.5), matches the normal running category;

[0091] The third interval: [25.5, 26.7), matches a higher category;

[0092] The fourth interval: [26.7, 28.2], matches the high-temperature running category;

[0093] The mean of all temperature values ​​is calculated as μ = 25.3. Since 25.3 ∈ [24.0, 25.5), the current operating category is determined to be normal.

[0094] The normal category is converted into a 4-dimensional one-hot encoded vector. According to the mapping rule that low temperature is the 1st bit, normal is the 2nd bit, higher is the 3rd bit, and high temperature is the 4th bit, the operating condition encoded vector [0,1,0,0] is generated.

[0095] In step S3, a multimodal feature set is obtained based on the candidate feature point set and the working condition encoding vector. The specific implementation steps for locating the reference pedestal based on the multimodal feature set and the pre-stored multimodal template feature set include steps S33 and S34:

[0096] Step S33: Concatenate the descriptors of the feature points in each candidate feature point set with the working condition coding vector to obtain multimodal features; collect all multimodal features to obtain a multimodal feature set; obtain a multimodal template feature set, which contains multiple multimodal template features, including the descriptors of template feature points and the working condition coding vector.

[0097] In this step, the descriptors of the feature points in each candidate feature point set are concatenated with the working condition coding vector to obtain multimodal features, i.e., the first 256 bits are the visual descriptors and the last 4 bits are the working condition codes.

[0098] The multimodal template feature set serves as the benchmark for system localization. It is not established in real-time on the production line, but rather precisely established offline and persistently stored in a controlled calibration environment before system deployment. The establishment process is as follows:

[0099] Coordinate system establishment: A two-dimensional rectangular coordinate system is established with the upper surface (assembly plane) of the assembly platform as the XY plane and its geometric center as the origin;

[0100] Preset theoretical position: The theoretical coordinates of the center of the reference pedestal in the two-dimensional coordinate system are preset by the user;

[0101] Establish the calibration reference position: Under standard ambient temperature (25.0℃), after placing the reference base on the assembly station, measure the two-dimensional coordinates (X,Y) of the center of the reference base on the assembly plane, and fine-tune it to bring the center of the reference base to the calibration reference position. This position must satisfy: the distance d between the actual coordinates (X,Y) and the theoretical coordinates (X0,Y0) of the center of the reference base satisfies:

[0102] d= <0.005mm

[0103] When the reference base is in the above-mentioned calibration reference position, perform the following operations:

[0104] Acquire template image: Use the same image acquisition method as in step S1 to capture an image containing the reference pedestal;

[0105] Extracting and constructing template features: Extract feature points and their 256-bit descriptors from the template image (method as in step S21), and record the pixel coordinates of each feature point in the template image. Based on the acquisition environment (25.0℃, normal state), determine the template temperature condition code as [0,1,0,0]. Concatenate the descriptor of each template feature point with this condition code to form a 260-bit multimodal template feature.

[0106] The fundamental purpose of establishing the aforementioned calibration reference position is to provide a visual reference for subsequent visual positioning. At this precise location, a deterministic mapping relationship is established between all visual features (including the pixel coordinate distribution of feature points) of the template image and the theoretical physical position of the workpiece, under a given imaging geometry (vertical camera mounting). Therefore, during real-time positioning on the production line, it is only necessary to match the features of the real-time acquired image with the features of the template image and calculate the difference in the statistical characteristics (such as the mean) of the pixel coordinates of the feature points between the two. Since the physical position represented by the template image is precisely known, this difference in pixel coordinates at the image level can be linearly and reliably mapped to the physical position deviation of the real-time workpiece relative to this high-precision calibration reference position. In short, high-precision physical calibration in the offline stage is the prerequisite and guarantee for achieving high-precision positioning using efficient image comparison algorithms in the online stage.

[0107] Step S34: Match the multimodal feature set with the preset multimodal template feature set. For each multimodal feature, calculate the Hamming distance between the multimodal feature and each multimodal template feature. Divide the second smallest Hamming distance by the smallest Hamming distance to obtain the distance ratio. When the distance ratio is greater than the preset distance ratio threshold, match the multimodal feature with the multimodal template feature corresponding to the smallest Hamming distance to obtain a preliminary matching pair. Filter all the preliminary matching pairs again and complete the positioning of the reference pedestal based on the filtering results.

[0108] In this step, the second smallest Hamming distance and the smallest Hamming distance are the second nearest neighbor and the nearest neighbor, respectively. The second smallest Hamming distance is also the second smallest Hamming distance. The preset distance ratio threshold can be 1.5, but it can be customized according to your needs.

[0109] In this step, the specific implementation steps for filtering all the preliminary matching pairs again and completing the positioning of the reference base based on the filtering results include steps S341 and S342.

[0110] Step S341: Filter all preliminary matching pairs again. Calculate the first similarity between the descriptors in the multimodal features and the descriptors in the multimodal template features, the first Hamming distance between the working condition coding vectors in the multimodal features and the working condition coding vectors in the multimodal template features, and the second similarity for each preliminary matching pair. Sum the first and second similarities using weighted averages to obtain a comprehensive matching score. Simultaneously, determine whether the working condition coding vectors in the multimodal features and the working condition coding vectors in the multimodal template features are the same in each preliminary matching pair. If they are the same, set the comprehensive matching score threshold to the first threshold A. If they are not the same, divide the first Hamming distance by the total number of running categories to obtain the influence factor B. Calculate the comprehensive matching score threshold: Comprehensive matching score threshold = A × (1 - preset coefficient × B). Compare the comprehensive matching score with the comprehensive matching score threshold. If the comprehensive matching score is greater than the comprehensive matching score threshold, retain the preliminary matching pair; otherwise, delete it. Record the finally retained preliminary matching pairs as correct matching pairs.

[0111] In this step, the first similarity is calculated as 1 - Hamming distance / 256, and the second similarity is calculated as 1 - Hamming distance / 4. The weights of the first and second similarities can be 0.95 and 0.05, respectively, and the preset coefficient can be 0.2.

[0112] Step S342: In all correctly matched pairs, calculate the average of the x and y coordinates of all feature point pixels in all multimodal features to obtain the first average x coordinate and the first average y coordinate; calculate the average of the x and y coordinates of all template feature point pixels in all multimodal template features to obtain the second average x coordinate and the second average y coordinate; subtract the second average x coordinate from the first average x coordinate to obtain the x coordinate offset; subtract the second average y coordinate from the first average y coordinate to obtain the y coordinate offset; multiply the x and y coordinate offsets by a preset pixel equivalent to obtain the horizontal axis position adjustment and the vertical axis position adjustment; use the horizontal axis position adjustment and the vertical axis position adjustment to adjust the reference base to complete the positioning of the reference base.

[0113] In this step, the pixel equivalent coefficient is obtained through camera calibration experiments, which characterizes the actual physical size corresponding to a single pixel in the image plane. Specifically, the camera installed in step S1 is installed with its optical axis perpendicular to the assembly station, so that the pixel coordinate offset and the physical adjustment satisfy a linear proportional relationship. Specifically, a standard calibration plate with known physical size is placed on the assembly station; the camera takes an image of the calibration plate; the number of pixels corresponding to the calibration features in the image is measured; and the pixel equivalent is calculated as: pixel equivalent = physical size / number of pixels.

[0114] By multiplying the horizontal and vertical coordinate offsets by the preset pixel equivalents, the position adjustment of the reference base in the horizontal and vertical directions within the assembly plane can be obtained. After calculating the position adjustment in the horizontal and vertical directions, the robot or actuator is guided to compensate for the assembly position, thereby completing the positioning of the reference base. This information can then be sent to the operator, who can adjust the reference base according to the horizontal and vertical position adjustment to complete its positioning.

[0115] Example 2

[0116] like Figure 2 As shown, this embodiment provides an intelligent positioning system for a reference pedestal, the system including an acquisition module 1, a clustering module 2, and a positioning module 3.

[0117] The acquisition module 1 is used to acquire a digital image containing a reference base, and at the same time acquire the working condition data corresponding to the digital image, which includes multiple continuous attribute values.

[0118] Clustering module 2 is used to extract feature points from digital images and cluster the feature points to obtain multiple visual feature clusters. It selects representative visual feature clusters from the visual feature clusters and determines the candidate feature point set corresponding to each representative visual feature cluster.

[0119] The positioning module 3 is used to obtain the working condition encoding vector based on the continuous attribute values ​​in the working condition data; to obtain the multimodal feature set based on the candidate feature point set and the working condition encoding vector; and to complete the positioning of the reference pedestal based on the multimodal feature set and the pre-stored multimodal template feature set.

[0120] In one specific embodiment of this disclosure, the clustering module 2 further includes a processing unit 21 and a clustering unit 22.

[0121] Processing unit 21 is used to sequentially perform grayscale conversion, Gaussian filtering noise reduction, and histogram equalization on the digital image to obtain a preprocessed digital image; and to use the ORB algorithm to detect feature points in the preprocessed digital image, and to calculate the corresponding descriptor for each feature point.

[0122] Clustering unit 22 is used to calculate the Hamming distance between any two descriptors using the average chaining clustering algorithm and convert it into similarity; feature points with similarity greater than a preset first similarity threshold are iteratively clustered and merged until the maximum similarity between all clusters is less than the preset first similarity threshold, resulting in multiple visual feature clusters; the number of feature points contained in each visual feature cluster is calculated, and the top N visual feature clusters are selected as representative visual feature clusters in descending order of the number of feature points.

[0123] In one specific embodiment of this disclosure, the clustering module 2 further includes a first calculation unit 23 and a filtering unit 24.

[0124] The first calculation unit 23 is used to record feature points other than those representing visual feature clusters as remaining feature points; for each remaining feature point, it calculates the similarity between it and all feature points in each representative visual feature cluster, and takes the maximum similarity as the inter-class similarity between the remaining feature point and each representative visual feature cluster.

[0125] The filtering unit 24 is used to filter out the remaining feature points whose inter-class similarity is between a preset second similarity threshold and a preset first similarity threshold for each representative visual feature cluster, thereby forming a candidate feature point set corresponding to the representative visual feature cluster. The preset second similarity threshold is less than the preset first similarity threshold.

[0126] In one specific embodiment of this disclosure, the positioning module 3 further includes a selection unit 31 and a determination unit 32.

[0127] Select unit 31 to randomly select an attribute value from all consecutive attribute values ​​as the first initial cluster center; set the required total number of initial cluster centers K, and repeat the following process until K initial cluster centers are selected, where K is greater than 1:

[0128] Using the latest selected initial cluster center as the baseline value, calculate the absolute difference between each attribute value and the baseline value, and record it as the current distance; square all current distance values ​​and sum them to obtain the sum of squared values; compare the square of each current distance with the sum of squared values ​​to obtain the probability ratio of that attribute value being selected as the next initial cluster center; select the attribute value with the largest probability ratio as the new initial cluster center;

[0129] The determination unit 32 is used to determine the working condition coding vector based on the obtained K initial cluster centers.

[0130] In one specific embodiment of this disclosure, the determining unit 32 further includes a second calculation unit 321 and a conversion unit 322.

[0131] The second calculation unit 321 is used to perform clustering using the K-means algorithm after obtaining K initial cluster centers, to obtain clusters, calculate the mean of all attribute values ​​in each cluster, sort the clusters in ascending order of mean, determine the maximum and minimum values ​​of attribute values ​​in each cluster after sorting, and calculate the boundary value corresponding to two adjacent clusters. The boundary value is equal to the sum of the maximum value of the previous cluster and the minimum value of the next cluster, divided by two.

[0132] The conversion unit 322 is used to divide the range of continuous attribute values ​​into K consecutive and non-overlapping numerical intervals based on the calculated K-1 boundary values. The first interval contains the minimum value to the minimum boundary value among all attribute values, the last interval contains the maximum boundary value to the maximum value among all attribute values, and the intermediate intervals correspond to the numerical range between adjacent boundary values. Each numerical interval is matched with an operating category. The mean of all attribute values ​​is calculated, and the operating category corresponding to the operating condition data is determined according to the numerical interval in which the mean is located. The operating category is converted into a one-hot encoded vector, and the one-hot encoded vector is used as the operating condition encoded vector.

[0133] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0134] Example 3

[0135] Corresponding to the above method embodiments, this disclosure also provides an intelligent positioning device for a reference pedestal. The intelligent positioning device for a reference pedestal described below and the intelligent positioning method for a reference pedestal described above can be referred to in correspondence with each other.

[0136] Figure 3 This is a block diagram of a smart positioning device 300 for a reference pedestal, according to an exemplary embodiment. Figure 3 As shown, the intelligent positioning device 300 of the reference base may include: a processor 301 and a memory 302. The intelligent positioning device 300 of the reference base may also include one or more of a multimedia component 303, an I / O interface 304, and a communication component 305.

[0137] The processor 301 controls the overall operation of the intelligent positioning device 300 of the reference pedestal to complete all or part of the steps in the aforementioned intelligent positioning method of the reference pedestal. The memory 302 stores various types of data to support the operation of the intelligent positioning device 300 of the reference pedestal. This data may include, for example, instructions for any application or method operating on the intelligent positioning device 300 of the reference pedestal, as well as application-related data, such as contact data, sent and received messages, pictures, audio, video, etc. The memory 302 can be implemented using any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The multimedia component 303 may include a screen and an audio component. The screen may be, for example, a touchscreen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signals may be further stored in the memory 302 or transmitted via the communication component 305. The audio component also includes at least one speaker for outputting audio signals. I / O interface 304 provides an interface between processor 301 and other interface modules, such as keyboards, mice, and buttons. These buttons can be virtual or physical. Communication component 305 is used for wired or wireless communication between the smart positioning device 300 of this reference base and other devices. Wireless communication includes, for example, Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, or 4G, or a combination thereof. Therefore, the corresponding communication component 305 may include a Wi-Fi module, a Bluetooth module, and an NFC module.

[0138] In an exemplary embodiment, the intelligent positioning device 300 of the reference pedestal may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the intelligent positioning method of the reference pedestal described above.

[0139] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, which, when executed by a processor, implement the steps of the intelligent positioning method for the reference pedestal described above. For example, the computer-readable storage medium may be the memory 302 including program instructions described above, which may be executed by the processor 301 of the intelligent positioning device 300 of the reference pedestal to complete the intelligent positioning method for the reference pedestal described above.

[0140] Example 4

[0141] Corresponding to the above method embodiments, this disclosure also provides a readable storage medium. The readable storage medium described below can be referred to in conjunction with the intelligent positioning method of the reference pedestal described above.

[0142] A readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the intelligent positioning method for the reference pedestal described in the above method embodiments.

[0143] Specifically, the readable storage medium can be a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, or any other readable storage medium capable of storing program code.

[0144] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent positioning of a reference pedestal, characterized in that, include: Acquire a digital image containing a reference base, and simultaneously acquire the corresponding operating condition data, which includes multiple consecutive attribute values. Feature points are extracted from digital images and clustered to obtain multiple visual feature clusters. A representative visual feature cluster is selected from the visual feature clusters. Determine the set of candidate feature points corresponding to each cluster representing visual features; The working condition encoding vector is obtained based on the continuous attribute values ​​in the working condition data; the descriptors of the feature points in each candidate feature point set are concatenated with the working condition encoding vector to obtain multimodal features; All multimodal features are collected to obtain a multimodal feature set. Based on the multimodal feature set and the pre-stored multimodal template feature set, the positioning of the reference pedestal is completed. The working condition coding vector is obtained based on continuous attribute values ​​in the working condition data, including: Randomly select an attribute value from all consecutive attribute values ​​as the first initial cluster center; set the required total number of initial cluster centers K, and repeat the following process until K initial cluster centers are selected, where K is greater than 1: Using the latest selected initial cluster center as the baseline value, calculate the absolute difference between each attribute value and the baseline value, and record it as the current distance; square all current distance values ​​and sum them to obtain the sum of squared values; compare the square of each current distance with the sum of squared values ​​to obtain the probability ratio of that attribute value being selected as the next initial cluster center; select the attribute value with the largest probability ratio as the new initial cluster center; The working condition coding vector is determined based on the obtained K initial cluster centers; The determination of the working condition coding vector based on the obtained K initial cluster centers includes: After obtaining K initial cluster centers, the K-means algorithm is used for clustering to obtain clusters. The mean of all attribute values ​​in each cluster is calculated, and the clusters are sorted in ascending order of mean. After sorting, the maximum and minimum values ​​of attribute values ​​in each cluster are determined. After sorting, the boundary value corresponding to two adjacent clusters is calculated. The boundary value is equal to the sum of the maximum value of the previous cluster and the minimum value of the next cluster, divided by two. Based on the calculated K-1 boundary values, the range of continuous attribute values ​​is divided into K continuous and non-overlapping numerical intervals. The first interval contains the minimum value to the minimum boundary value among all attribute values, the last interval contains the maximum boundary value to the maximum value among all attribute values, and the intermediate intervals correspond to the numerical range between adjacent boundary values. Each numerical interval is matched with an operating category. The mean of all attribute values ​​is calculated, and the operating category corresponding to the operating condition data is determined according to the numerical interval in which the mean is located. The operating category is converted into a one-hot encoded vector, and this one-hot encoded vector is used as the operating condition encoded vector.

2. The intelligent positioning method for the reference pedestal according to claim 1, characterized in that, Feature points are extracted from digital images and clustered to obtain multiple visual feature clusters. Representative visual feature clusters are selected from these clusters, including: The digital image is sequentially processed by grayscale conversion, Gaussian filtering for noise reduction, and histogram equalization to obtain a preprocessed digital image. The ORB algorithm is then used to detect feature points in the preprocessed digital image, and a corresponding descriptor is calculated for each feature point. The average linkage clustering algorithm is used to calculate the Hamming distance between any two descriptors and convert it into similarity. Feature points with similarity greater than a preset first similarity threshold are iteratively clustered and merged until the maximum similarity between all clusters is less than the preset first similarity threshold, resulting in multiple visual feature clusters. The number of feature points contained in each visual feature cluster is calculated, and the top N visual feature clusters are selected as representative visual feature clusters in descending order of the number of feature points.

3. The intelligent positioning method for the reference pedestal according to claim 2, characterized in that, Determine the set of candidate feature points corresponding to each cluster representing visual features, including: Feature points other than those representing visual feature clusters are denoted as residual feature points. For each residual feature point, the similarity between it and all feature points in each representative visual feature cluster is calculated, and the maximum similarity is taken as the inter-class similarity between the residual feature point and each representative visual feature cluster. For each representative visual feature cluster, the remaining feature points whose inter-class similarity is between a preset second similarity threshold and a preset first similarity threshold are selected to form a candidate feature point set corresponding to the representative visual feature cluster. The preset second similarity threshold is less than the preset first similarity threshold.

4. An intelligent positioning system for a reference pedestal, characterized in that, include: The acquisition module is used to acquire a digital image containing a reference base, and at the same time acquire the corresponding working condition data of the digital image, which includes multiple continuous attribute values. The clustering module is used to extract feature points from digital images and cluster the feature points to obtain multiple visual feature clusters. A representative visual feature cluster is selected from the visual feature clusters. Determine the set of candidate feature points corresponding to each cluster representing visual features; The localization module is used to obtain the working condition encoding vector based on the continuous attribute values ​​in the working condition data; the descriptors of the feature points in each candidate feature point set are concatenated with the working condition encoding vector to obtain multimodal features; All multimodal features are collected to obtain a multimodal feature set. Based on the multimodal feature set and the pre-stored multimodal template feature set, the positioning of the reference pedestal is completed. The positioning module includes: Select a cell to randomly choose an attribute value from all consecutive attribute values ​​as the first initial cluster center; set the required total number of initial cluster centers K, and repeat the following process until K initial cluster centers are selected, where K is greater than 1: Using the latest selected initial cluster center as the baseline value, calculate the absolute difference between each attribute value and the baseline value, and record it as the current distance; square all current distance values ​​and sum them to obtain the sum of squared values; compare the square of each current distance with the sum of squared values ​​to obtain the probability ratio of that attribute value being selected as the next initial cluster center; select the attribute value with the largest probability ratio as the new initial cluster center; The determining unit is used to determine the working condition coding vector based on the obtained K initial cluster centers; The determining unit includes: The second calculation unit is used to perform clustering using the K-means algorithm after obtaining K initial cluster centers, to obtain clusters, calculate the mean of all attribute values ​​in each cluster, sort the clusters in ascending order of mean, and determine the maximum and minimum attribute values ​​in each cluster after sorting; after sorting, calculate the boundary value corresponding to two adjacent clusters, the boundary value is equal to the sum of the maximum value of the previous cluster and the minimum value of the next cluster, divided by two; The conversion unit is used to divide the range of continuous attribute values ​​into K consecutive and non-overlapping numerical intervals based on the calculated K-1 boundary values. The first interval contains the minimum value to the minimum boundary value among all attribute values, the last interval contains the maximum boundary value to the maximum value among all attribute values, and the intermediate intervals correspond to the numerical range between adjacent boundary values. Each numerical interval is matched with an operating category. The mean of all attribute values ​​is calculated, and the operating category corresponding to the operating condition data is determined according to the numerical interval in which the mean is located. The operating category is converted into a one-hot encoded vector, and the one-hot encoded vector is used as the operating condition encoded vector.

5. The intelligent positioning system for the reference pedestal according to claim 4, characterized in that, The clustering module includes: The processing unit is used to sequentially perform grayscale conversion, Gaussian filtering for noise reduction, and histogram equalization on the digital image to obtain a preprocessed digital image; the ORB algorithm is used to detect feature points in the preprocessed digital image, and the corresponding descriptor is calculated for each feature point; The clustering unit is used to calculate the Hamming distance between any two descriptors using the average chaining clustering algorithm and convert it into similarity. Feature points with similarity greater than a preset first similarity threshold are iteratively clustered and merged until the maximum similarity between all clusters is less than the preset first similarity threshold, resulting in multiple visual feature clusters. The number of feature points contained in each visual feature cluster is calculated, and the top N visual feature clusters are selected as representative visual feature clusters in descending order of the number of feature points.

6. The intelligent positioning system for the reference pedestal according to claim 5, characterized in that, The clustering module includes: The first calculation unit is used to record feature points other than those representing visual feature clusters as remaining feature points; for each remaining feature point, it calculates the similarity between it and all feature points in each representative visual feature cluster, and takes the maximum similarity as the inter-class similarity between the remaining feature point and each representative visual feature cluster. The filtering unit is used to filter out the remaining feature points whose inter-class similarity is between a preset second similarity threshold and a preset first similarity threshold for each representative visual feature cluster, thereby forming a candidate feature point set corresponding to the representative visual feature cluster. The preset second similarity threshold is less than the preset first similarity threshold.

Images