Method and system for creating robot visual navigation map

Abstract

The invention relates to the technical field of navigation map creation, in particular to a robot visual navigation map creation method and system.The robot visual navigation map creation method comprises the steps that an initial identification point is set, a target robot is controlled to conduct global scanning on a to-be-mapped area with the initial identification point as a starting point, and the target robot is provided with a high-definition camera and an odometer; and carrying out image acquisition on the periphery of the initial identification point based on the high-definition camera to obtain an environment image set of the initial identification point, and carrying out feature extraction on the environment image set of the initial identification point to obtain an image feature set of the environment image set. Through the environment image collected by the high-definition camera and the distance data of the odometer, the position of the identification point can be accurately determined, the robot is helped to formulate an optimized path, the navigation precision is improved, a navigation topological graph can be automatically updated to adapt to environment change, and the navigation graph is ensured to be real-time and effective; and the influence of environment change on the navigation accuracy is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of navigation map creation, and in particular to a method and system for creating a robot vision navigation map. Background Art

[0002] A robot is a machine device that can perform tasks automatically. It is usually composed of hardware and software. They can perceive the environment through sensors, process information through algorithms, and perform actions with mechanical systems. Robots have a wide range of applications, including industrial production, medical care, services, exploration and other fields.

[0003] Traditional methods usually rely on a single sensor and may not provide sufficient positioning accuracy, especially in complex or dynamic environments. Traditional methods usually use pre-created static maps and do not have the ability to update in real time, which means that environmental changes may cause maps and navigation instructions to fail, thereby affecting navigation accuracy. In traditional methods, the identification of landmarks and the updating of maps usually require a lot of manual intervention, which increases labor costs and the possibility of errors. Traditional map creation methods often require manual marking and updating of landmarks, which is a cumbersome and time-consuming process, resulting in low efficiency and poor consistency. Summary of the invention

[0004] The purpose of the present invention is to address the problems existing in the background technology and to propose a method and system for creating a robot visual navigation map.

[0005] The technical solution of the present invention is a method for creating a robot visual navigation map, comprising: An initial identification point is set, and a target robot is controlled to perform a full-area scan of the area to be mapped using the initial identification point as a starting point, wherein the target robot is equipped with a high-definition camera and an odometer; Based on the high-definition camera, images are collected around the initial marking point to obtain an environmental image set of the initial marking point, and features are extracted from the environmental image set of the initial marking point to obtain an image feature set of the environmental image set; Identify the identification points based on the image feature set of the environment image set to obtain a next identification point set, control the target robot to move to the next identification point set in sequence, and obtain a distance set between the initial identification point and the next identification point set based on the odometer; Constructing a navigation topology map based on the initial identification point and the next identification point set, and assigning weights to the navigation topology map based on the image feature set and the distance set to obtain a weighted navigation topology map; When the target robot moves to the next identification point, the next identification point is used as the initial identification point, and the above operation is repeated until the traversal of the area to be mapped is completed, so as to update the weighted navigation topology map and obtain the visual navigation map corresponding to the area to be mapped.

[0006] Preferably, the navigation topology graph includes a plurality of nodes and a plurality of connection edges, wherein the nodes correspond to the identification points, and the connection edges correspond to the connected areas between the identification points.

[0007] Preferably, the weighted navigation topology graph includes a plurality of node weights and a plurality of connection edge weights, wherein the node weights correspond to the image feature set, and the connection edge weights correspond to the distance set.

[0008] Preferably, feature extraction is performed on the environment image set of the initial identification point to obtain an image feature set of the environment image set, including: Randomly selecting an environment image from the environment image set, and performing road segmentation on the road in the environment image to obtain a road image corresponding to the environment image; Acquiring the area of ​​the road image by a pixel counting method, and using the area of ​​the road image as a first image feature of the environment image; The second image feature of the environment image is calculated, and the calculation formula of the second image feature is as follows: ; in, The second image feature representing the environment image, i.e., the ratio of the road image area to the environment image area, represents the area of ​​the environment image, Represents the road image area; Acquire a minimum rectangular frame of the road image, and use the length and width of the minimum rectangular frame as the third image feature of the environment image; Combining the first image feature, the second image feature and the third image feature of the environment image to obtain an image feature set corresponding to the environment image; The above operation is repeated until the environment image set is traversed to obtain the image feature set of the environment image set.

[0009] Preferably, segmenting the roads in the environment image to obtain a road image corresponding to the environment image includes: The minimum and maximum grayscale values ​​of the environment image are obtained, and a first segmentation threshold is calculated based on the minimum and maximum grayscale values. The calculation formula of the first segmentation threshold is as follows: ; in, represents the first segmentation threshold, Represents the minimum gray value, Indicates the maximum gray value; Segmenting the environment image into a first road image and a first background environment image based on the first segmentation threshold; The average grayscale value of the first road image and the average grayscale value of the first background environment image are obtained, and a second segmentation threshold is calculated based on the average grayscale value of the first road image and the average grayscale value of the first background environment image. The second segmentation threshold calculation formula is as follows: ; in, represents the second segmentation threshold, represents the average gray value of the first road image, represents the average gray value of the first background environment image; The first segmentation threshold is compared with the second segmentation threshold. If the first segmentation threshold is equal to the second segmentation threshold, the environment image is segmented into a second road image based on the second segmentation threshold, and the second road image is the final road image. Otherwise, the second segmentation threshold is assigned to the first segmentation threshold, and the above operation is repeated.

[0010] Preferably, identifying the identification points based on the image feature set of the environment image set to obtain the next identification point set includes: Performing dimensionality reduction on the image feature set of the environment image to obtain a reduced-dimensional image feature set; Inputting the reduced-dimensionality image feature set into a trained road classification model, and outputting the road type corresponding to the reduced-dimensionality image feature set, that is, the road type corresponding to the environment image, through the trained road classification model, wherein the road type includes a closed road and a connected road; Determine whether there is a marking point in the environment image based on the road type corresponding to the environment image; if there is a marking point in the environment image, perform target detection on the environment image through a trained target detection model to obtain the next marking point, wherein the target detection is used to detect the intersection of the roads in the environment image; The above operation is repeated until the environment image set is traversed to obtain the next identification point set.

[0011] Preferably, reducing the dimension of the image feature set of the environment image set to obtain the reduced-dimensional image feature set includes: The image feature set is standardized to obtain an input matrix ; Calculate the input matrix , and decompose the covariance matrix to obtain an eigenvalue set corresponding to the covariance matrix, and sort the eigenvalue set in descending order to obtain a descending eigenvalue set; The descending eigenvalue sets are formed into a diagonal matrix, an optimization function is constructed based on the diagonal matrix, the optimization function is solved to obtain a preset number of minimum eigenvalues, corresponding eigenvectors are obtained based on the preset number of minimum eigenvalues, and the eigenvectors are combined to obtain a projection matrix, and the optimization function is as follows: ; in, represents the number of eigenvalues ​​in the descending eigenvalue set, represents the number of column vectors of a diagonal matrix, represents the first eigenvalues, Represents the data of the original space retained in the sub-dimensional space; The input matrix is ​​converted based on the projection matrix Perform dimensionality reduction to obtain a reduced-dimensional image feature set.

[0012] Preferably, the road classification model adopts an improved support vector machine model, and the road classification model improves the support vector machine model by using a gray wolf optimization algorithm, wherein the improvement of the support vector machine model by using the gray wolf optimization algorithm includes: The initialization parameters of the gray wolf optimization algorithm are set to initialize the position of the gray wolf population. The gray wolf population position initialization formula is as follows: ; in, and Indicates The iteration and The position of the gray wolf individual at the iteration, Represents the parameters of the preset control nonlinear mapping; Calculate the fitness value of the gray wolf individual and calculate the elite individual position. The elite individual position calculation formula is as follows: ; in, Indicates The first iteration The reverse solution of the gray wolf individual, Indicates The first iteration The elite solution for the gray wolf individual, Representation interval The random number inside Indicates The maximum position of the gray wolf individual at the iteration, Indicates The minimum position of the individual gray wolf at the iteration; Marking a preset number of the gray wolf individuals with the best fitness values ​​as leader wolves, and updating the positions of the remaining gray wolf individuals in the gray wolf population; Calculate the updated individual fitness values ​​of the gray wolves and sort them, obtain the elite gray wolves, and update the positions of the elite gray wolves; Repeat the above steps until the optimal solution is found or the maximum number of iterations of the gray wolf optimization algorithm is reached, so as to obtain the optimal penalty factor and kernel function parameters of the support vector machine model, and improve the support vector machine model based on the optimal penalty factor and kernel function parameters.

[0013] Preferably, the target detection model adopts Faster R-CNN.

[0014] The technical solution of the present invention is a system for creating a robot visual navigation map, which is applicable to the method for creating a robot visual navigation map, comprising: An initial setting unit, the initial setting unit is used to set an initial identification point, and control the target robot to perform a full-area scan of the area to be mapped with the initial identification point as a starting point, wherein the target robot is equipped with a high-definition camera and an odometer; A feature extraction unit, the feature extraction unit is used to collect images around the initial identification point based on the high-definition camera to obtain an environmental image set of the initial identification point, and perform feature extraction on the environmental image set of the initial identification point to obtain an image feature set of the environmental image set; A marking recognition unit, the marking recognition unit is used to recognize the marking points based on the image feature set of the environment image set to obtain a next marking point set, control the target robot to move to the next marking point set in sequence, and obtain a distance set between the initial marking point and the next marking point set based on the odometer; A topology mapping unit, the topology mapping unit is used to construct a navigation topology map based on the initial identification point and the next identification point set, and to assign weights to the navigation topology map based on the image feature set and the distance set to obtain a weighted navigation topology map; A map construction unit, wherein when the target robot moves to the next identification point, the map construction unit uses the next identification point as the initial identification point, and repeats the above operation until the traversal of the area to be mapped is completed, so as to update the weighted navigation topology map, thereby obtaining a visual navigation map corresponding to the area to be mapped.

[0015] Compared with the prior art, the above technical solution of the present invention has the following beneficial technical effects: The present invention can accurately determine the position of the identification point through the environmental image collected by the high-definition camera and the distance data of the odometer, help the robot to formulate an optimized path, improve navigation accuracy, and can automatically update the navigation topology map to adapt to environmental changes, ensure that the navigation map is real-time and effective, and reduce the impact of environmental changes on navigation accuracy. In addition, through image feature extraction and identification point recognition, important locations can be automatically discovered and marked, reducing manual intervention. In addition, the map creation process is simplified by automatically identifying and updating identification points, reducing manual intervention and improving efficiency and consistency. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic flow chart of the steps of an overall method in an embodiment of the present invention; Figure 2 A schematic diagram of the system architecture of an overall system in an embodiment of the present invention.

[0017] Figure numerals: 1. Initial setting unit; 2. Feature extraction unit; 3. Identification unit; 4. Topological mapping unit; 5. Map construction unit. DETAILED DESCRIPTION

[0018] Embodiment 1, as Figure 1 As shown, a method for creating a robot visual navigation map proposed by the present invention includes: S1. Set an initial identification point, and use the initial identification point as the starting point to control the target robot to perform a full-area scan of the area to be mapped. The target robot is equipped with a high-definition camera and an odometer. S2, collecting images around the initial marking point based on a high-definition camera to obtain an environmental image set of the initial marking point, and extracting features from the environmental image set of the initial marking point to obtain an image feature set of the environmental image set; S3, identifying the identification points based on the image feature set of the environment image set to obtain the next identification point set, controlling the target robot to move to the next identification point set in sequence, and obtaining the distance set between the initial identification point and the next identification point set based on the odometer; S4, constructing a navigation topology map based on the initial identification point and the next identification point set, and assigning weights to the navigation topology map based on the image feature set and the distance set to obtain a weighted navigation topology map; S5. When the target robot moves to the next identification point, the next identification point is used as the initial identification point, and the above operation is repeated until the traversal of the area to be mapped is completed, so as to update the weighted navigation topology map and obtain the visual navigation map corresponding to the area to be mapped.

[0019] In the present invention, odometer refers to a sensor that records the robot's moving distance and angle changes, and is used to estimate the robot's position and path in space; landmarks refer to specific points on the map used for positioning and navigation, which are usually prominent and easily recognizable features; navigation topology map is a graph structure used to represent the relationship between different landmarks; visual navigation map refers to a map generated by the navigation topology map, which is used for the robot's navigation and positioning in the environment.

[0020] Embodiment 2, a method for creating a robot visual navigation map proposed in the present invention, compared with embodiment 1, this embodiment also includes: the navigation topology map includes multiple nodes and multiple connection edges, wherein the nodes correspond to the identification points, and the connection edges correspond to the connected areas between the identification points.

[0021] In an optional embodiment, the weighted navigation topology graph includes a plurality of node weights and a plurality of connection edge weights, wherein the node weights correspond to the image feature set, and the connection edge weights correspond to the distance set.

[0022] In an optional embodiment, feature extraction is performed on the environment image set of the initial identification point to obtain an image feature set of the environment image set, including: Randomly selecting an environment image from the environment image set, and performing road segmentation on the road in the environment image to obtain a road image corresponding to the environment image; The area of ​​the road image is obtained by a pixel counting method, and the area of ​​the road image is used as the first image feature of the environment image; The second image feature of the environment image is calculated. The calculation formula of the second image feature is as follows: ; in, The second image feature representing the environment image, i.e., the ratio of the road image area to the environment image area, represents the area of ​​the environment image, Represents the road image area; Obtaining a minimum rectangular frame of the road image, and using the length and width of the minimum rectangular frame as the third image feature of the environment image; Combining the first image feature, the second image feature, and the third image feature of the environment image to obtain an image feature set corresponding to the environment image; The above operation is repeated until the environment image set is traversed to obtain the image feature set of the environment image set.

[0023] It should be noted that road segmentation refers to an image processing technology that aims to extract roads from environmental images; the pixel counting method is a method used to calculate the area of ​​a specific area in an image (such as a road area) by estimating its area by counting the number of pixels in the area in the image; the minimum rectangular frame refers to enclosing the road area in the road image in a minimum rectangular frame. The length and width of this rectangular frame are the circumscribed rectangle of the road area, which can effectively describe the size and shape of the road area.

[0024] In an optional embodiment, segmenting a road in an environment image to obtain a road image corresponding to the environment image includes: Get the minimum and maximum grayscale values ​​of the environment image, and calculate the first segmentation threshold based on the minimum and maximum grayscale values. The calculation formula of the first segmentation threshold is as follows: ; in, represents the first segmentation threshold, Represents the minimum gray value, Indicates the maximum gray value; Segmenting the environment image into a first road image and a first background environment image based on a first segmentation threshold; The average grayscale value of the first road image and the average grayscale value of the first background environment image are obtained, and the second segmentation threshold is calculated based on the average grayscale value of the first road image and the average grayscale value of the first background environment image. The second segmentation threshold calculation formula is as follows: ; in, represents the second segmentation threshold, represents the average gray value of the first road image, represents the average gray value of the first background environment image; The first segmentation threshold is compared with the second segmentation threshold. If the first segmentation threshold is equal to the second segmentation threshold, the environment image is segmented into a second road image based on the second segmentation threshold, and the second road image is the final road image. Otherwise, the second segmentation threshold is assigned to the first segmentation threshold, and the above operation is repeated.

[0025] In an optional embodiment, identifying the identification points based on the image feature set of the environment image set to obtain the next identification point set includes: Performing dimensionality reduction on the image feature set of the environment image to obtain a dimensionality-reduced image feature set; Inputting the reduced-dimensionality image feature set into a trained road classification model, and outputting the road type corresponding to the reduced-dimensionality image feature set through the trained road classification model, that is, the road type corresponding to the environment image, wherein the road type includes a closed road and a connected road; Based on the road type corresponding to the environment image, it is determined whether there is a landmark point in the environment image. If there is a landmark point in the environment image, target detection is performed on the environment image through the trained target detection model to obtain the next landmark point, wherein target detection is used to detect the intersection of roads in the environment image; Repeat the above operation until the environment image set is traversed to obtain the next identification point set.

[0026] In an optional embodiment, reducing the dimension of the image feature set of the environment image set to obtain the reduced-dimensional image feature set includes: Normalize the image feature set to obtain the input matrix ; Calculate the input matrix The covariance matrix of is decomposed to obtain the eigenvalue set corresponding to the covariance matrix, and the eigenvalue set is sorted in descending order to obtain a descending eigenvalue set; The descending eigenvalue set is formed into a diagonal matrix, and an optimization function is constructed based on the diagonal matrix. The optimization function is solved to obtain a preset number of minimum eigenvalues, and the corresponding eigenvectors are obtained based on the preset number of minimum eigenvalues. The eigenvectors are combined to obtain a projection matrix. The optimization function is as follows: ; in, represents the number of eigenvalues ​​in the descending eigenvalue set, represents the number of column vectors of a diagonal matrix, represents the first eigenvalues, Represents the data of the original space retained in the sub-dimensional space; Based on the projection matrix, the input matrix Perform dimensionality reduction to obtain a reduced-dimensional image feature set.

[0027] It should be noted that the projection matrix refers to a matrix composed of selected eigenvectors, which is used to map high-dimensional data to a low-dimensional space to achieve data dimensionality reduction.

[0028] In an optional embodiment, the road classification model adopts an improved support vector machine model, and the road classification model improves the support vector machine model by using the gray wolf optimization algorithm, wherein the improvement of the support vector machine model by using the gray wolf optimization algorithm includes: Set the initialization parameters of the gray wolf optimization algorithm to initialize the position of the gray wolf population. The gray wolf population position initialization formula is as follows: ; in, and Indicates The iteration and The position of the gray wolf individual at the iteration, Represents the parameters of the preset control nonlinear mapping; Calculate the fitness value of the gray wolf individual and calculate the position of the elite individual. The calculation formula for the position of the elite individual is as follows: ; in, Indicates The iteration The reverse solution of the gray wolf individual, Indicates The iteration The elite solution for the gray wolf individual, Representation interval The random number inside, Indicates The maximum position of the gray wolf individual at the iteration, Indicates The minimum position of the individual gray wolf at the iteration; Mark a preset number of gray wolf individuals with the best fitness values ​​as leader wolves, and update the positions of the remaining gray wolf individuals in the gray wolf population; Calculate the updated individual fitness values ​​of the gray wolves and sort them, obtain the elite gray wolves, and update the positions of the elite gray wolves; Repeat the above steps until the optimal solution is found or the maximum number of iterations of the gray wolf optimization algorithm is reached, so as to obtain the optimal penalty factor and kernel function parameters of the support vector machine model, and improve the support vector machine model based on the optimal penalty factor and kernel function parameters.

[0029] It should be noted that support vector machine refers to a supervised learning model used for classification and regression analysis. SVM classifies data by finding an optimal hyperplane (or multiple hyperplanes) in the feature space, with the aim of maximizing the boundary (or interval) between classes. In the binary classification problem, the support vector machine separates the data points into two different categories, so that the points in each category are as far away from the dividing line as possible; the gray wolf optimization algorithm refers to a heuristic optimization algorithm that simulates the hunting behavior of gray wolves to find solutions to optimization problems. The gray wolf optimization algorithm updates the search position by simulating the social behavior of gray wolves and the prey capture process (including encirclement, tracking, attacking, etc.), and gradually finds the optimal solution to the objective function. It is a swarm intelligence algorithm that is widely used to solve optimization problems; the fitness values ​​of gray wolves include but are not limited to classification accuracy, classification error rate, cross-validation results, precision, recall rate and F1 score.

[0030] In an optional embodiment, the target detection model adopts Faster R-CNN.

[0031] It should be noted that Faster R-CNN is an advanced target detection model and the third generation model in the R-CNN series. Based on the previous R-CNN and Fast R-CNN, Faster R-CNN accelerates the target detection process by introducing RPN.

[0032] Embodiment three, as Figure 2 As shown, a system for creating a robot visual navigation map, which is applicable to the above-mentioned method for creating a robot visual navigation map, includes: Initial setting unit 1, the initial setting unit 1 is used to set an initial identification point, and control the target robot to perform a full-area scan of the area to be mapped with the initial identification point as the starting point, wherein the target robot is equipped with a high-definition camera and an odometer; Feature extraction unit 2, the feature extraction unit 2 is used to collect images around the initial identification point based on a high-definition camera to obtain an environmental image set of the initial identification point, and perform feature extraction on the environmental image set of the initial identification point to obtain an image feature set of the environmental image set; The identification unit 3 is used to identify the identification points based on the image feature set of the environment image set to obtain the next identification point set, control the target robot to move to the next identification point set in sequence, and obtain the distance set between the initial identification point and the next identification point set based on the odometer; A topology mapping unit 4 is used to construct a navigation topology map based on an initial identification point and a next identification point set, and to assign weights to the navigation topology map based on an image feature set and a distance set to obtain a weighted navigation topology map; Map construction unit 5, map construction unit 5 is used to take the next identification point as the initial identification point when the target robot moves to the next identification point, and repeat the above operation until the traversal of the area to be mapped is completed, so as to update the weighted navigation topology map, thereby obtaining a visual navigation map corresponding to the area to be mapped.

[0033] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto, and various changes can be made within the knowledge scope of technicians in the relevant technical field without departing from the purpose of the present invention.

Claims

1. A method for creating a robot visual navigation map, characterized in that: include: An initial identification point is set, and a target robot is controlled to perform a full-area scan of the area to be mapped using the initial identification point as a starting point, wherein the target robot is equipped with a high-definition camera and an odometer; Based on the high-definition camera, images are collected around the initial marking point to obtain an environmental image set of the initial marking point, and features are extracted from the environmental image set of the initial marking point to obtain an image feature set of the environmental image set; Identify the identification points based on the image feature set of the environment image set to obtain a next identification point set, control the target robot to move to the next identification point set in sequence, and obtain a distance set between the initial identification point and the next identification point set based on the odometer; Constructing a navigation topology map based on the initial identification point and the next identification point set, and assigning weights to the navigation topology map based on the image feature set and the distance set to obtain a weighted navigation topology map; When the target robot moves to the next identification point, the next identification point is used as the initial identification point, and the above operation is repeated until the traversal of the area to be mapped is completed, so as to update the weighted navigation topology map and obtain the visual navigation map corresponding to the area to be mapped.

2. The method for creating a robot visual navigation map according to claim 1, characterized in that: The navigation topology graph includes a plurality of nodes and a plurality of connection edges, wherein the nodes correspond to the identification points, and the connection edges correspond to the connected areas between the identification points.

3. The method for creating a robot visual navigation map according to claim 1, characterized in that: The weighted navigation topology graph includes a plurality of node weights and a plurality of connection edge weights, wherein the node weights correspond to the image feature set, and the connection edge weights correspond to the distance set.

4. The method for creating a robot visual navigation map according to claim 1, characterized in that: Extracting features from the environment image set of the initial identification point to obtain an image feature set of the environment image set includes: Randomly selecting an environment image from the environment image set, and performing road segmentation on the road in the environment image to obtain a road image corresponding to the environment image; Acquiring the area of ​​the road image by a pixel counting method, and using the area of ​​the road image as a first image feature of the environment image; The second image feature of the environment image is calculated, and the calculation formula of the second image feature is as follows: ； in, The second image feature representing the environment image, i.e., the ratio of the road image area to the environment image area, represents the area of ​​the environment image, Represents the road image area; Acquire a minimum rectangular frame of the road image, and use the length and width of the minimum rectangular frame as the third image feature of the environment image; Combining the first image feature, the second image feature and the third image feature of the environment image to obtain an image feature set corresponding to the environment image; The above operation is repeated until the environment image set is traversed to obtain the image feature set of the environment image set.

5. The method for creating a robot visual navigation map according to claim 4, characterized in that: Segmenting the road in the environment image to obtain a road image corresponding to the environment image includes: The minimum and maximum grayscale values ​​of the environment image are obtained, and a first segmentation threshold is calculated based on the minimum and maximum grayscale values. The calculation formula of the first segmentation threshold is as follows: ； in, represents the first segmentation threshold, Represents the minimum gray value, Indicates the maximum gray value; Segmenting the environment image into a first road image and a first background environment image based on the first segmentation threshold; The average grayscale value of the first road image and the average grayscale value of the first background environment image are obtained, and a second segmentation threshold is calculated based on the average grayscale value of the first road image and the average grayscale value of the first background environment image. The second segmentation threshold calculation formula is as follows: ； in, represents the second segmentation threshold, represents the average gray value of the first road image, represents the average gray value of the first background environment image; The first segmentation threshold is compared with the second segmentation threshold. If the first segmentation threshold is equal to the second segmentation threshold, the environment image is segmented into a second road image based on the second segmentation threshold, and the second road image is the final road image. Otherwise, the second segmentation threshold is assigned to the first segmentation threshold, and the above operation is repeated.

6. The method for creating a robot visual navigation map according to claim 4, characterized in that: Identifying the identification points based on the image feature set of the environment image set to obtain the next identification point set includes: Performing dimensionality reduction on the image feature set of the environment image to obtain a reduced-dimensional image feature set; Inputting the reduced-dimensionality image feature set into a trained road classification model, and outputting the road type corresponding to the reduced-dimensionality image feature set, that is, the road type corresponding to the environment image, through the trained road classification model, wherein the road type includes a closed road and a connected road; Determine whether there is a marking point in the environment image based on the road type corresponding to the environment image; if there is a marking point in the environment image, perform target detection on the environment image through a trained target detection model to obtain the next marking point, wherein the target detection is used to detect the intersection of the roads in the environment image; The above operation is repeated until the environment image set is traversed to obtain the next identification point set.

7. The method for creating a robot visual navigation map according to claim 6, characterized in that: Performing dimensionality reduction on the image feature set of the environment image set to obtain a reduced-dimensional image feature set, including: The image feature set is standardized to obtain an input matrix ; Calculate the input matrix , and decompose the covariance matrix to obtain an eigenvalue set corresponding to the covariance matrix, and sort the eigenvalue set in descending order to obtain a descending eigenvalue set; The descending eigenvalue sets are formed into a diagonal matrix, an optimization function is constructed based on the diagonal matrix, the optimization function is solved to obtain a preset number of minimum eigenvalues, corresponding eigenvectors are obtained based on the preset number of minimum eigenvalues, and the eigenvectors are combined to obtain a projection matrix, and the optimization function is as follows: ； in, represents the number of eigenvalues ​​in the descending eigenvalue set, Represents a diagonal matrix The number of column vectors, represents the first eigenvalues, Represents the data of the original space retained in the sub-dimensional space; The input matrix is ​​converted based on the projection matrix Perform dimensionality reduction to obtain a reduced-dimensional image feature set.

8. The method for creating a robot visual navigation map according to claim 6, characterized in that: The road classification model adopts an improved support vector machine model, and the road classification model improves the support vector machine model by using the gray wolf optimization algorithm, wherein the support vector machine model is improved by using the gray wolf optimization algorithm, including: The initialization parameters of the gray wolf optimization algorithm are set to initialize the position of the gray wolf population. The gray wolf population position initialization formula is as follows: ； in, and Indicates The iteration and The position of the gray wolf individual at the iteration, Represents the parameters of the preset control nonlinear mapping; Calculate the fitness value of the gray wolf individual and calculate the elite individual position. The elite individual position calculation formula is as follows: ； in, Indicates The iteration The reverse solution of the gray wolf individual, Indicates The iteration The elite solution for the gray wolf individual, Representation interval The random number inside, Indicates The maximum position of the gray wolf individual at the iteration, Indicates The minimum position of the individual gray wolf at the iteration; Marking a preset number of the gray wolf individuals with the best fitness values ​​as leader wolves, and updating the positions of the remaining gray wolf individuals in the gray wolf population; Calculate the updated individual fitness values ​​of the gray wolves and sort them, obtain the elite gray wolves, and update the positions of the elite gray wolves; Repeat the above steps until the optimal solution is found or the maximum number of iterations of the gray wolf optimization algorithm is reached, so as to obtain the optimal penalty factor and kernel function parameters of the support vector machine model, and improve the support vector machine model based on the optimal penalty factor and kernel function parameters.

9. The method for creating a robot visual navigation map according to claim 6, characterized in that: The target detection model adopts Faster R-CNN.

10. A system for creating a robot visual navigation map, which is applicable to a method for creating a robot visual navigation map according to any one of claims 1 to 9, characterized in that: include: An initial setting unit (1), the initial setting unit (1) is used to set an initial identification point, and control a target robot to perform a full-area scan of a region to be mapped with the initial identification point as a starting point, wherein the target robot is equipped with a high-definition camera and an odometer; A feature extraction unit (2), the feature extraction unit (2) being used to collect images around the initial identification point based on the high-definition camera to obtain an environmental image set of the initial identification point, and to extract features from the environmental image set of the initial identification point to obtain an image feature set of the environmental image set; A marking recognition unit (3), the marking recognition unit (3) being used to recognize the marking points based on the image feature set of the environment image set to obtain a next marking point set, control the target robot to move to the next marking point set in sequence, and obtain a distance set between the initial marking point and the next marking point set based on the odometer; A topology mapping unit (4), the topology mapping unit (4) being used to construct a navigation topology map based on the initial identification point and the next identification point set, and to assign weights to the navigation topology map based on the image feature set and the distance set, so as to obtain a weighted navigation topology map; A map construction unit (5), wherein the map construction unit (5) is used to use the next identification point as the initial identification point when the target robot moves to the next identification point, and repeat the above operation until the traversal of the area to be mapped is completed, so as to update the weighted navigation topology map, thereby obtaining a visual navigation map corresponding to the area to be mapped.

Images