Intelligent settlement robot system and method

Abstract

The invention relates to the field of intelligent robots, and relates to an intelligent settlement robot system and method, and the method comprises the steps: detecting the feature point deviation degree between a commodity tag code and a tag code recognition model through a feature point detection method, and carrying out the recognition of a commodity tag code based on the feature point deviation degree; the closeness degree between the commodity information after label code image recognition and settlement features set in the intelligent settlement robot is analyzed, so that according to the closeness degree, rotation adjustment is further performed on deviation conditions of different numbers of feature points, accurate adjustment of the feature points of the label code image is realized, and the user experience is improved. And the settlement accuracy and the settlement efficiency of the intelligent settlement robot are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent robots, more particularly, to an intelligent settlement robot system and method. BACKGROUND

[0002] In the current commodity settlement scene, the traditional settlement method often relies on manual code scanning or manual input of commodity information, which is low in efficiency and prone to errors. With the development of technology, intelligent settlement robots have gradually emerged, but the existing intelligent settlement robots have many problems in commodity label code recognition, such as low recognition accuracy, large external interference, and low recognition efficiency, which are difficult to meet the demand for fast and accurate settlement.

[0003] In the prior art, due to different angles of goods, the intelligent settlement robot has low recognition accuracy when recognizing the commodity label code, thereby causing the problem of low settlement efficiency for subsequent commodity settlement operation, therefore, the present application constructs a label code recognition model, detects the feature point deviation degree between the commodity label code and the label code recognition model by using the feature point detection method when the angle of the commodity is different, and analyzes the closeness between the commodity information after label code image recognition and the settlement features set in the intelligent settlement robot according to the feature point deviation degree, so as to further adjust the rotation according to the closeness of the different number of feature point deviations, realize the accurate adjustment of the feature points of the label code image, and improve the accuracy and efficiency of the intelligent settlement robot settlement.

[0004] In order to solve the above problems, an intelligent settlement robot system and method are proposed. SUMMARY

[0005] To achieve the above purpose, the present application provides the following technical solutions: First aspect: an intelligent settlement robot method, comprising: constructing a label code recognition model through the rectangular recognition area of the intelligent settlement robot; image feature recognition module: converting the commodity label code into a label code image and inputting it into the label recognition model, comparing the feature point deviation number and the feature point deviation degree to determine whether the feature points of the label code image and the label recognition model are locally coincident, wherein the local coincidence includes single-point coincidence or multi-point coincidence; missing feature analysis module: if the local coincidence is true, the settlement feature missing degree after the label code image recognition is obtained, and it is judged whether there is a correlation between the settlement feature missing and the feature point missing; classification adjustment module: if there is a correlation between the settlement feature missing and the feature point missing, the adjustment angle is obtained for single-point coincidence and multi-point coincidence respectively, and the rotation adjustment is performed according to the adjustment angle.

[0006] As a further scheme of the present application: the construction process of the label code recognition model is as follows: Taking the width of the rectangular recognition area as the X axis and the length of the rectangular recognition area as the Y axis, a label code recognition model is constructed.

[0007] As a further scheme of the present application: the process of comparing from the number of feature point deviations is as follows: The feature points that coincide in the label code image and the label recognition model are taken as the coinciding feature points, and otherwise, the feature points that do not coincide are taken as the non-coinciding feature points. The proportion of the number of non-coinciding feature points in the total number of feature points in the label recognition model is calculated to obtain the feature deviation number.

[0008] As a further scheme of the present application: the coordinate distance between the coordinates of any selected non-coinciding feature points and the corresponding reference coordinates in the label recognition model is obtained, and the proportion of the total perimeter of the label recognition model is calculated and then averaged to obtain the feature deviation degree.

[0009] As a further scheme of the present application: the process of judging whether the feature points of the label code image and the label recognition model are completely coinciding or partially coinciding is as follows: The feature deviation number and the feature deviation degree are input into a geometric mean model to output a feature comparison value. If the feature comparison value is greater than a feature comparison threshold, a partial coincidence signal is generated. If the feature comparison value is less than or equal to the feature comparison threshold, a complete coincidence signal is generated.

[0010] As a further scheme of the present application: the process of obtaining the settlement feature missing degree after the label code image recognition is as follows: The settlement features after the label code image recognition are compared with the set settlement features to extract the settlement missing features, and the proportion of the number of settlement missing features in the number of set settlement features is calculated to output the settlement feature missing degree.

[0011] As a further scheme of the present application: the process of analyzing the correlation between the settlement feature missing degree and the feature point missing degree is as follows: The settlement feature missing degree and the feature point missing degree of the intelligent settlement robot in multiple historical settlement periods are extracted respectively. Based on the settlement feature missing degree and the feature point missing degree in multiple historical settlement periods, a historical missing set and a historical comparison set are constructed respectively, and the elements in the historical missing set and the historical comparison set are correspondingly combined to obtain multiple suspected association groups. The elements in any one suspected association group are input into a coefficient of variation model to obtain a suspected association coefficient. If the suspected correlation coefficient is less than the preset correlation coefficient threshold, a correlation closeness signal is generated.

[0012] As a further scheme of the present application: for the single-point coincidence case, an adjustment angle is obtained, and the process is as follows: If there is only one coincident feature point between the label code image and the label recognition model, a single-point deviation degree corresponding to any non-coincident feature point is obtained, input into the vector angle formula, and mean value calculation is performed, and an adjustment angle is output. The coincident feature points in the label code image and the label recognition model are taken as the rotation origin, and rotation adjustment is performed according to the adjustment angle.

[0013] As a further scheme of the present application: for the multi-point coincidence case, an adjustment angle is obtained, and the process is as follows: If there are two or more coincident feature points between the label code image and the label recognition model, the non-coincident feature points and the coincident feature points are integrated respectively to obtain a matching point set and a non-matching point set. The barycentric coordinates of the matching point set and the barycentric coordinates of the non-matching point set are obtained respectively, input into the vector angle formula, and mean value calculation is performed to obtain an adjustment angle, the original point in the label recognition model is taken as the rotation origin, and rotation is performed around the rotation axis according to the adjustment angle.

[0014] Secondly, an intelligent settlement robot system comprises: A recognition model construction module: a label code recognition model is constructed through a rectangular recognition area of the intelligent settlement robot. An image feature recognition module: a commodity label code is converted into a label code image and input into the label recognition model, and feature point missing degree is compared from the number of feature point deviations and the degree of feature point deviations to determine whether the feature points of the label code image and the label recognition model are locally coincident, wherein local coincidence includes single-point coincidence or multi-point coincidence. A missing feature analysis module: if local coincidence exists, the settlement feature missing degree after label code image recognition is obtained, and the correlation between settlement feature missing and feature point missing is determined. A classification adjustment module: if the correlation exists, adjustment angles are obtained for single-point coincidence and multi-point coincidence respectively, and rotation adjustment is performed according to the adjustment angles.

[0015] (Three) beneficial effects Compared with the prior art, the present application provides an intelligent settlement robot method, which has the following beneficial effects: 1. The intelligent settlement robot usually uses image recognition technology to recognize the label code on the commodity, obtains commodity information, and performs subsequent amount calculation according to the commodity information, therefore, a label code recognition model is constructed, the feature deviation value is obtained by comprehensively analyzing the feature point deviation number and the feature point deviation degree through the label code recognition model, so as to not only pay attention to the deviation number of the feature point, but also consider the degree of deviation, these two aspects can balance the influence of these two factors, avoid the single factor leading to the comparison result, and further identify the feature deviation degree on the basis of identifying the non-coincidence feature point, optimize the identification matching process, and improve the identification efficiency; 2. When the local coincidence signal appears, the commodity information after label code image recognition is compared with the settlement features set in the robot, the same feature is marked as coincidence, and the different feature is marked as missing, the feature missing degree is calculated, the feature missing degree and the feature comparison value when the local coincidence signal is generated in a plurality of historical settlement periods are collected, the historical missing and comparison sets are integrated according to time respectively, and the suspected correlation group is composed of elements at the same time, and the suspected correlation value is output, so as to reflect the closeness between the commodity information after label code image recognition and the settlement features set in the intelligent settlement robot, which is helpful to adjust the commodity feature recognition operation under the condition of local coincidence according to the closeness obtained, so as to improve the accuracy and efficiency of settlement; 3. The local coincidence signal of the label code image is classified and processed, for the single-point coincidence signal, the non-coincidence feature point coordinates in the label code image and the label recognition model are extracted, the vector is constructed and the vector angle is calculated, the adjustment angle is obtained through mean value processing, so as to adjust the feature point, for the multi-point coincidence signal, the barycentric coordinates of the matching point set and the non-matching point set are obtained respectively, the related vector is determined, the straight line perpendicular to the vector and passing through the barycentric coordinates of the matching point set is taken as the rotation axis, the rotation direction is determined by using the inverse tangent function, the preliminary adjustment angle is obtained through mean value calculation, the feature point of the label code image is precisely adjusted by rotating around the rotation origin and combining with the coordinate transformation, and it is determined whether to iteratively adjust or replace the feature point detection method logic in the label recognition model according to the feature point detection result. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 A flow chart of the intelligent settlement robot method of the application; Figure 2 A structure diagram of the intelligent settlement robot system of the application. DETAILED DESCRIPTION

[0017] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0018] Example 1 like Figure 1 As shown in the embodiment of the present invention, an intelligent checkout robot method is provided. When the intelligent checkout robot settles accounts for goods, it includes a product information acquisition stage, a promotional rule matching stage, and a cost calculation stage. The product information acquisition stage is particularly important because the intelligent checkout robot typically uses image recognition technology to identify the label code on the product, obtain product information, and then performs subsequent settlement operations such as the promotional rule matching stage and the cost calculation stage based on the product information. Specifically, it includes the following steps: Step 1: Based on the label code recognition area of ​​the intelligent checkout robot, construct a label code recognition model; It should be noted that the label code recognition area is set by those skilled in the art based on the shape of the product label code, including a rectangular recognition area or a square recognition area; For example, taking a rectangular recognition region as an example, the process of constructing a label code recognition model is as follows: A label code recognition model is constructed using the width of the rectangular recognition area as the X-axis and the length of the rectangular recognition area as the Y-axis. Understandably, the purpose of building a label recognition model is: Function 1: It can concentrate the model's recognition resources on a specific area (label code recognition area), which not only effectively eliminates interference from external factors, such as background or patterns on other products outside the label code recognition area, but also avoids blindly searching for the label code in the entire image, thereby quickly locating the target label code and significantly improving recognition efficiency; Function 2: Since the shape and size of the recognition area are fixed, the model can extract the feature points of the label code more stably. No matter where the label code is in the image, as long as it is within the recognition area, the model can extract and compare the feature points of the product label code in sequence according to the feature point detection method, thereby improving the recognition accuracy of the product label code. Step 2: Convert the product label code into a label code image and input it into the label recognition model. Then, use the feature point detection method to compare the image feature points and obtain the image comparison results. The image comparison results include locally overlapping signals or completely offset signals; In some embodiments, the process of converting a product label code into a label code image is as follows: A1. Preprocess the collected product label codes, including grayscale preprocessing. A2. The grayscale preprocessed product label code is processed using the threshold segmentation method to obtain the label code image; It needs to be explained that the purpose of converting product label codes into label code images is as follows: First, converting color images into grayscale images reduces the image data from three-dimensional images (RGB three channels) to one-dimensional images. This is because the amount of computation required to process one-dimensional images is less than that required to process three-dimensional images, which helps intelligent checkout robots to quickly complete the label code recognition operation and improves the efficiency of product checkout. Function 2: By using the threshold segmentation method, the grayscale image is divided into foreground and background according to the threshold, so that the features and background in the processed label image have a clearer contrast. Compared with the feature point detection method, the clear contrast helps to improve the recognition accuracy of feature points in the label image. The label code image is input into the label recognition model, and quantitative analysis is performed on the number and degree of feature point deviation. The overlap between the feature points in the label code image and the feature points in the label recognition model is compared. Specifically, extract the overlapping feature points between the label code image and the label recognition model, and mark them as overlapping feature points; Extract non-overlapping feature points from the tag code image and the tag recognition model, count the number of non-overlapping feature points, and calculate the ratio of the number of feature points to the total number of feature points in the tag recognition model to obtain the feature deviation. Specifically, arbitrarily select a non-overlapping feature point; Obtain the recognition coordinates of non-overlapping feature points within the label recognition model, as well as the reference coordinates within the label recognition model. Input these coordinates into the coordinate distance calculation formula, and calculate the ratio between the coordinates and the total perimeter of the label recognition model. Output the single-point deviation. The characteristic deviation degree is obtained by averaging all the single-point deviations. The number and degree of feature bias are input into the geometric mean model, and the feature comparison value is output. It should be noted that the geometric mean model is a mathematical method used to calculate the average level of multiple values. The principle is to multiply multiple numbers and then take the square root of the corresponding power. The reason for combining the number and degree of feature deviations into the geometric mean model is that, overall, both the number and degree of feature deviations only apply to a single object (feature point deviation). They are obtained through a comprehensive analysis from two different dimensions of feature point deviation. More specifically, the number of feature deviations reflects the number of non-overlapping feature points in the label image and the label recognition model, while the degree of deviation reflects the degree of deviation for each non-overlapping feature point. Therefore, based on the principle of the geometric mean model, it considers not only the number of feature point deviations but also the degree of deviation. These two aspects balance the influence of these two factors in judging the deviation or overlap between the label image and the label recognition model, avoiding a single factor dominating the comparison result. This helps the robot more accurately determine whether the product label image matches the model, thereby improving the accuracy and reliability of recognition. To further explain, the meaning of the feature comparison value is: it comprehensively reflects the overall deviation between the feature points in the label code image and the label recognition model. By using the geometric mean model, these two indicators that describe the deviation of feature points from different dimensions are combined, balancing the impact of the number and degree of feature deviation on the result. This avoids making a one-sided judgment based on only one factor, so that when the checkout robot recognizes the product label code image, it will not make a misjudgment due to a large number of feature point deviations but a small degree of deviation, or a large degree of deviation but a small number of deviations. This improves the recognition accuracy of the product label code. Furthermore, by identifying non-overlapping feature points, it further identifies the degree of feature deviation, optimizes the recognition matching process, and improves recognition efficiency. The feature alignment values ​​are compared with the feature alignment threshold, as follows: If the feature comparison value is greater than the feature comparison threshold, it indicates that the proportion of non-overlapping feature points between the tag code image and the tag recognition model is high, and the degree of deviation is high, resulting in the generation of local overlapping signals. If the feature comparison value is less than or equal to the feature comparison threshold, it means that the proportion of non-overlapping feature points between the tag code image and the tag recognition model is low, and the degree of deviation is low, resulting in a completely overlapping signal. The specific implementation scheme of this invention is as follows: Intelligent checkout robots typically use image recognition technology to identify the label codes on goods, obtain product information, and perform subsequent amount calculations based on the product information. Therefore, a label code recognition model is constructed. The label code recognition model comprehensively analyzes the number and degree of feature point deviations to obtain feature deviation values. This approach considers not only the number of feature point deviations but also the degree of deviation. These two aspects balance the influence of these two factors in determining the degree of deviation or overlap between the label code image and the label recognition model, avoiding a single factor dominating the comparison results. Furthermore, by identifying non-overlapping feature points, the degree of feature deviation is further identified, optimizing the recognition and matching process and improving recognition efficiency.

[0019] In Example 2 like Figure 1 As shown in the figure, the intelligent settlement robot method provided in this embodiment of the invention further includes the following steps: Step 3: Based on the local overlapping signal, extract the product information after the label code image recognition, and compare it with the settlement features to obtain the feature missing degree. Analyze the correlation between the feature missing degree and the feature comparison value to obtain the overlap association value, and compare it with the overlap association threshold to obtain the association evaluation result. The information assessment results include closely related signals or distantly related signals; To help understand this, product information includes, but is not limited to, features such as product type or product price; It should be noted that the settlement features are features set by those skilled in the art within the intelligent settlement robot, and these settlement features include, but are not limited to, features such as product type or product price. In some embodiments, the product information obtained after the tag code image is recognized is compared with the settlement features. If the recognized product information contains the same settlement features as the settlement features, it is marked as a coincident feature. If the identified product information contains settlement features that are different from the settlement features, it is marked as a missing feature. The number of missing features is counted and the ratio is calculated with the total number of settlement features in the complete product information to obtain the feature missingness. For example, feature missing value and feature comparison value are extracted when the intelligent settlement robot generates locally overlapping signals in multiple historical settlement cycles; The feature missing values ​​when generating locally overlapping signals in multiple historical settlement periods are integrated according to the time series to obtain the historical missing set; The feature comparison values ​​generated when local overlapping signals were generated in multiple historical settlement periods were integrated according to the time series to obtain the historical comparison set; Specifically, elements that generated locally overlapping signals at the same time were extracted from the historical missing set and the historical comparison set, respectively, and these were used as a group of suspected associations. It should be noted that the elements generated at the same time when the local overlapping signal is generated refer to the elements in the historical missing set and the elements in the historical comparison set being obtained when the same local overlapping signal is generated. The suspected association value is obtained by calculating the ratio of elements within the suspected association group; Substituting the suspected correlation values ​​into the coefficient of variation model yields the suspected correlation coefficient. It is understandable that the suspected correlation coefficient represents the degree of closeness between the product information after the tag code image recognition and the settlement features set in the intelligent checkout robot, in terms of product information and settlement features. This helps to adjust the product feature recognition operation in cases of local overlap based on the obtained closeness, thereby improving the accuracy and efficiency of settlement. In terms of the intelligent checkout robot, the suspected correlation coefficient is an important indicator for evaluating the performance of the intelligent checkout robot in the process of image recognition and product settlement. It also reflects whether image recognition technologies such as feature point detection are suitable for intelligent checkout robots to recognize product tag codes during the product settlement process. The formula for calculating the coefficient of variation is as follows: The CV coefficient of the suspected correlation was calculated. in, It was obtained by averaging all suspected correlation values. It was obtained by calculating the standard deviation of all suspected correlation values; It should be noted that the reason for obtaining the suspected correlation coefficient through the coefficient of variation model is that the essence of the coefficient of variation model is to reflect the fluctuation of data at different times. The suspected correlation coefficient obtained through the coefficient of variation model can reflect the stability of the correlation between feature missing value and feature comparison value in different historical periods. The suspected correlation coefficient is compared with a preset correlation coefficient threshold, as follows: If the suspected correlation coefficient is greater than or equal to the preset correlation coefficient threshold, it indicates that the correlation between the degree of missing settlement features and the degree of missing feature points in multiple historical settlement cycles is not very strong, generating a signal of distant correlation. If the suspected correlation coefficient is less than the preset correlation coefficient threshold, it indicates that the degree of missing settlement features and the degree of missing feature points are closely related in multiple historical settlement cycles, generating a closely related signal. The specific implementation scheme of this invention is as follows: When a partial overlap signal occurs, the product information after the tag code image recognition is extracted and compared with the settlement features set in the robot. Identical features are marked as overlap, and differences are marked as missing. The feature missing degree is calculated accordingly. The feature missing degree and feature comparison value when partial overlap signals are generated in multiple historical settlement cycles are collected and integrated into historical missing and comparison sets according to time. Elements at the same time are extracted to form a suspected association group, and the suspected association value is output. This reflects the closeness between the product information after tag code image recognition and the settlement features set in the intelligent settlement robot. It helps to adjust the product feature recognition operation in the case of partial overlap based on the obtained closeness, thereby improving the accuracy and efficiency of settlement.

[0020] Example 3 like Figure 1 As shown in the figure, the intelligent settlement robot method provided in this embodiment of the invention further includes the following steps: Step 4: Based on closely related signals, classify locally overlapping signals, obtain and adjust the included angle, and adjust the feature points on the label code image; Among them, local overlapping signals include single-point overlapping signals or multi-point overlapping signals; It should be noted that a single-point coincidence signal refers to a situation where, when comparing the feature points of the tag code image with those of the tag recognition model, there is only one coincidence feature point between the tag code image and the tag recognition model. Multi-point overlapping signal refers to the situation where, when comparing the feature points of the tag code image with those of the tag recognition model, there are two or more overlapping feature points between the tag code image and the tag recognition model. For example, for a single-point coincidence signal, the coordinates corresponding to the single-point coincidence feature points in the tag code image are extracted ( , ); Extract the coordinates of multiple non-overlapping feature points in the label code image respectively. , ), and the coordinates corresponding to multiple non-overlapping feature points in the label recognition model ( , ); Where j represents the total number of coordinates corresponding to non-overlapping feature points in the label code image or label recognition model, ( , ) represents the coordinates of the non-overlapping feature points in the j-th tag image. , ) represents the coordinates of the non-overlapping feature points in the j-th label recognition model; B1. Arbitrarily select the coordinates corresponding to multiple non-overlapping feature points in a label code image, and the coordinates corresponding to multiple non-overlapping feature points in the label recognition model, and obtain the vector. The process is as follows: ; ; B2, based on the obtained vector, input it into the vector angle formula: The rotation angle is calculated. ; B3, rotate all the included angles Perform mean calculation and compare with Perform the product and output the adjusted angle. ; Using the origin within the tag code recognition model as the rotation origin, the included angle is adjusted accordingly. Perform rotational adjustment; For example, for multi-point overlapping signals, the coordinates corresponding to multiple overlapping feature points in the tag code image are extracted ( , ), and integrate them to obtain a set of matching points; Extract the coordinates of multiple non-overlapping feature points in the label code image respectively. , ), and integrate them to obtain a set of non-matching points; S1, summing all coordinates within the matching point set and calculating the ratio with the total number of coordinates in the matching point set, yields the centroid coordinates of the matching point set. , ); S2 sums up all coordinates within the non-matching point set and calculates the ratio of this sum to the total number of coordinates within the non-matching point set to obtain the centroid coordinates of the non-matching point set. , ); S3, based on the centroid coordinates of the matching point set ( , The centroid coordinates of the non-matching point set () , ), output ; S4, Select the centroid of the matching point set that is perpendicular to the vector. The straight line is the axis of rotation and passes through the arctangent function. ; S5, if If it is positive, then rotate counterclockwise. If the value is negative, then rotate clockwise; S6, arbitrarily select the coordinates of an element from the set of non-matching points, and compare them with ( , Combination And using the formula for the angle between vectors, we can obtain: ; S7, the included angles corresponding to the coordinates of all elements within the non-matching point set. The mean value is calculated to obtain the adjusted angle. ; S8 uses the origin of the tag code recognition model as the origin of rotation and rotates around the rotation axis. At the same time, it combines two-dimensional rotating rectangle coordinate transformation. For example, choose one arbitrarily ( , Rotate around the origin of rotation The coordinates after the angle are ( , )for ; S9. After rotation adjustment, the matching is re-performed using the feature point detection method to check whether the number of non-overlapping feature points has decreased compared to the number of feature deviations. If the number decreases, iterative adjustments are made until the feature points in the tag code image completely overlap with the corresponding feature points in the tag recognition model. If there is no reduction, then the feature point detection method in the label recognition model needs to be replaced with a logical recognition method. The specific implementation scheme of this invention is as follows: Local overlapping signals in the label code image are classified and processed. For single-point overlapping signals, the coordinates of non-overlapping feature points in the label code image and label recognition model are extracted. Vectors are constructed and the angle between the vectors is calculated. After mean-averaging, the adjustment angle is obtained to adjust the feature points. For multi-point overlapping signals, the centroid coordinates of the matching point set and the non-matching point set are obtained separately, and the relevant vectors are determined. The rotation axis is taken as the straight line perpendicular to the vector and the centroid of the matching point set. The rotation direction is determined using the arctangent function. The preliminary adjustment angle is obtained through mean-averaging. The rotation is performed around the origin and combined with coordinate transformation. Based on the feature point detection results, it is determined whether to iteratively adjust or replace the feature point detection logic within the label recognition model, thereby achieving precise adjustment of the feature points in the label code image. In a specific implementation process, the corrected product identification data and settlement results can also be uploaded to the blockchain in real time for evidence storage to ensure the immutability of transaction data.

[0021] Example 4 like Figure 2 As shown in the figure, the intelligent settlement robot method provided in this embodiment of the invention further includes the following modules: Recognition Model Construction Module: Constructs a label code recognition model using the rectangular recognition area of ​​the intelligent checkout robot; Image feature recognition module: Converts product label codes into label code images and inputs them into the label recognition model. It compares the number and degree of feature point deviations to determine the feature point missingness and whether the feature points of the label code image and the label recognition model locally overlap. Local overlap includes single-point overlap or multi-point overlap. Missing Feature Analysis Module: If there is local overlap, the module obtains the degree of missing settlement features after tag code image recognition and determines whether there is a correlation between missing settlement features and missing feature points. Classification adjustment module: If there is a correlation, the adjustment angle is obtained for single-point overlap and multi-point overlap respectively, and the rotation adjustment is performed according to the adjustment angle.

[0022] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for an intelligent settlement robot, characterized in that: include: A label code recognition model is constructed using the rectangular recognition area of ​​the intelligent checkout robot; The product label code is converted into a label code image and input into the label recognition model. The feature point missing degree is compared by the number and degree of feature point deviation, and it is determined whether the feature points of the label code image and the label recognition model locally overlap. Local overlap includes single-point overlap or multi-point overlap; If there is local overlap, the degree of missing settlement features after the label code image recognition is obtained, and it is determined whether there is a correlation between the missing settlement features and the missing feature points. If a correlation exists, the adjustment angle is obtained for single-point overlap and multi-point overlap respectively, and the rotation adjustment is performed based on the adjustment angle.

2. The intelligent settlement robot method according to claim 1, characterized in that: The construction process of the tag code recognition model is as follows: A label code recognition model is constructed using the width of the rectangular recognition area as the X-axis and the length of the rectangular recognition area as the Y-axis.

3. The intelligent settlement robot method according to claim 1, characterized in that: The comparison is based on the number of feature point deviations, and the process is as follows: Feature points that overlap with the tag code image and the tag recognition model are considered as overlapping feature points; conversely, feature points that do not overlap are considered as non-overlapping feature points. The proportion of non-overlapping feature points to the total number of feature points in the label recognition model is used to obtain the feature deviation count.

4. The intelligent settlement robot method according to claim 3, characterized in that: The comparison is based on the degree of deviation of feature points, and the process is as follows: Obtain the coordinates of any selected non-coincident feature point and the coordinate distance between them and the corresponding reference coordinates within the label recognition model. Calculate the proportion of the feature point's total perimeter within the label recognition model and then perform average output to obtain the feature deviation degree.

5. The intelligent settlement robot method according to claim 3, characterized in that: The process for determining whether the feature points of the tag code image and the tag recognition model completely overlap or only partially overlap is as follows: The number and degree of feature bias are input into the geometric mean model, and the feature comparison value is output. If the feature comparison value is greater than the feature comparison threshold, a local overlap signal is generated; If the feature comparison value is less than or equal to the feature comparison threshold, a completely overlapping signal is generated.

6. The intelligent settlement robot method according to claim 1, characterized in that: The process for obtaining the degree of missing settlement features after tag code image recognition is as follows: The settlement features after recognizing the tag code image are compared with the set settlement features to extract the missing settlement features. The proportion of the number of missing settlement features to the number of set settlement features is counted, and the degree of missing settlement features is output.

7. The intelligent settlement robot method according to claim 6, characterized in that: The correlation between the degree of missing settlement features and the degree of missing feature points is analyzed as follows: The degree of missing settlement features and the degree of missing feature points were extracted for the intelligent settlement robot in multiple historical settlement cycles. Based on the degree of missing settlement features and the degree of missing feature points in multiple historical settlement cycles, a historical missing set and a historical comparison set are constructed respectively. The elements in the historical missing set and the historical comparison set are combined accordingly to obtain multiple suspected related groups. Input any set of elements from a suspected association group into the coefficient of variation model to obtain the suspected association coefficient. If the suspected correlation coefficient is less than the preset correlation coefficient threshold, a closely correlated signal is generated.

8. The intelligent settlement robot method according to claim 1, characterized in that: For cases where a single point coincides, the included angle is obtained and adjusted as follows: If there is only one overlapping feature point between the tag code image and the tag recognition model, then obtain the single-point deviation degree corresponding to any non-overlapping feature point, input it into the vector angle formula, perform mean calculation, and output the adjusted angle. Using the overlapping feature points of the tag code image and the tag recognition model as the origin of rotation, the rotation is adjusted according to the included angle.

9. The intelligent settlement robot method according to claim 1, characterized in that: For cases where multiple points overlap, the adjustment angle is obtained, and the process is as follows: If there are two or more overlapping feature points between the tag code image and the tag code recognition model, the non-overlapping feature points and overlapping feature points are integrated respectively to obtain the matching point set and the non-matching point set. The centroid coordinates of the matching point set and the non-matching point set are obtained respectively, input into the vector angle formula, and mean value calculation is performed to obtain the adjustment angle. The origin in the tag code recognition model is used as the rotation origin, and the adjustment angle is used to rotate around the rotation axis. After the rotation adjustment, the feature point detection method is used to rematch. If the number of overlapping feature points decreases, iterative adjustment is performed until the feature points in the tag code image completely overlap with the corresponding feature points in the tag recognition model.

10. An intelligent checkout robot system, applied to the intelligent checkout robot method according to any one of claims 1-9, characterized in that: Includes the following modules: Recognition Model Construction Module: Constructs a label code recognition model using the rectangular recognition area of ​​the intelligent checkout robot; Image feature recognition module: Converts product label codes into label code images and inputs them into the label recognition model. It compares the number and degree of feature point deviations to determine the feature point missingness and whether the feature points of the label code image and the label recognition model locally overlap. Local overlap includes single-point overlap or multi-point overlap. Missing Feature Analysis Module: If there is local overlap, the module obtains the degree of missing settlement features after tag code image recognition and determines whether there is a correlation between missing settlement features and missing feature points. Classification and rotation adjustment module: If there is a correlation, the adjustment angle is obtained for single-point overlap and multi-point overlap respectively, and the rotation adjustment is performed according to the adjustment angle.

Images