A robot vision positioning method, device and system

Abstract

The application relates to a robot vision positioning method, device and system, periodically acquires image information in front of a robot, extracts obstacle image information from the image information, generates an obstacle information image, perceives robot positioning information, uploads the robot positioning information to the cloud, the cloud calls a saved historical obstacle information image corresponding to the positioning position for comparison, and if the image information comparison result is consistent, navigation is continued according to a planned path; if the image information comparison result is inconsistent, a new obstacle appears in front of the path, at this time, the robot pauses and enters a new obstacle judgment mode to judge whether the path needs to be changed. The positioning method can improve the recognition accuracy of obstacles in the robot travel path, and through the setting of a geographical fence of the planned path, the risk of navigation failure caused by the collision between the robot and the newly appeared obstacle can be reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, in particular to a robot walking environment perception method and device, computer equipment and readable storage medium. BACKGROUND

[0002] At present, intelligent robots are more and more widely used in the industrial field, and the path planning technology of intelligent robots is a hot research topic, which has broad application prospect and scientific research value, and has important application in robot automatic navigation positioning, traffic route navigation, power inspection and other fields.

[0003] The navigation methods with relatively mature technology and high accuracy at present mainly include the navigation method based on the human change of environment and the method of matching the environment map model or visual navigation, the former has a narrow application range, and needs to manually lay navigation light strips or magnetic strips to assist positioning, which is not suitable for dynamic and work environment changing scenes; the latter needs the robot to identify the natural features in the working environment for navigation, and the stability and robustness of this navigation method are the focus of current research, how to improve the error correction and deviation correction ability of the robot visual positioning to ensure the accuracy and safety of the robot walking route has become a problem to be solved. SUMMARY

[0004] In order to overcome the defects of the background art, the present application provides a robot visual positioning method, device and system, the scheme is as follows:

[0005] A robot visual positioning method, periodically acquires image information in front of the robot, extracts obstacle image information from the image information, and generates an obstacle information image; at the same time, the robot positioning information is perceived and uploaded to the cloud, and the cloud calls the saved historical obstacle information image corresponding to the positioning place for comparison; the specific comparison method is: for each obstacle appearing in the obstacle information image, the area is divided, and for each obstacle area, the label information containing the large verification mark is formed as the first verification information, and each obstacle area is divided into a plurality of small grids, each small grid contains small identification information as the second verification information;

[0006] The cloud calls the saved label information of the verification obstacle information corresponding to the positioning place to confirm whether the path is correct, first compares the first verification information of each obstacle, and if correct, compares the second verification information of each obstacle, selects the middle position small grid for comparison, and if the small grid label information is identified incorrectly, immediately finds the small grid adjacent to it for label confirmation;

[0007] If the first verification information and the second verification information are consistent, continue to navigate according to the planned path;

[0008] The first verification information comparison result is consistent, and the second verification information comparison result is inconsistent, and then entering the real-time mode;

[0009] The first verification information is inconsistent, and it is considered that the inconsistent information obstacle is a newly added obstacle appearing in front of the path. At this time, the robot pauses the progress and enters the newly added obstacle judgment mode to judge whether the path needs to be changed to enter the real-time mode.

[0010] Further, the obstacle judgment mode is specifically: the region segmentation is performed on the newly added obstacle in the obstacle information image, the region range of the newly added obstacle is identified, the depth image information of the same positioning place is photographed by calling the depth camera, the pixel feature point information of the newly added obstacle in the depth image information is extracted according to the region range of the newly added obstacle, and the pixel feature point information includes the three-dimensional coordinates of each pixel point of the newly added obstacle; all pixel feature point information constitutes a three-dimensional coordinate cluster, whether the newly added obstacle blocks the stored planning path is judged by comparing the two-dimensional coordinates in the three-dimensional coordinate cluster with the historical planning path two-dimensional coordinates saved in the cloud; if it blocks, it is switched to the real-time mode, otherwise it continues to navigate according to the historical path.

[0011] Further, the real-time mode no longer receives the historical data information of the place environment, and guides the robot to drive according to the real-time environment image information.

[0012] Further, the stored planning path is a geographical fence including positioning information and combining a robot contour, wherein the coordinates of the geographical fence are determined by the following method: setting the positioning point coordinates Am=(xi,yi) on the planning path, and the coordinates of the geographical fence on both sides of the positioning point corresponding to the path are Lm=(lxi,lyi) and Rm=(rxi,ryi), wherein the connecting line of Lm and Rm is perpendicular to the connecting line between Am and Am-1, and the intersection point of the two connecting lines is the midpoint of the connecting line of Lm and Rm; in this way, the angle values of the angles α and β can be calculated, and then γ=β-α and θ=90°-β-α are obtained; the formula for calculating the coordinates of the geographical fence is as follows: Figure 3

[0013] (1);

[0014] (2);

[0015] Wherein, γ is the included angle between the connecting line of Am and Rm and the x-axis;

[0016] The above formula can be used to calculate the geographical fence coordinates Rm(rx i, ry i ) on the right side of the current positioning point Am. ​

[0017] The geographic fence coordinate Lm(lx i, ly i ) on the left side of the current positioning point Am can be calculated by the following formula:

[0018] (3);

[0019] (4);

[0020] Wherein, θ is the included angle between the line connecting Am-1 point and Lm point and the y-axis; S is the width of the robot planar projection contour, L is the length of the longest side of the robot planar projection contour or the diameter of the smallest circle that can accommodate the robot planar projection contour; △t is a margin set according to experience, generally, the value of △t can be between (0, (L-S) / 2).

[0021] A robot vision positioning device, comprising an image acquisition module, periodically acquiring image information in front of the robot; a positioning module, acquiring robot positioning information; and uploading the robot positioning information to the cloud, a cloud processing module, calling the saved historical obstacle information image, and sending it to the robot; an image processing module, extracting obstacle image information from the image information, generating an obstacle information image, and comparing it with the historical obstacle information image transmitted by the cloud.

[0022] Further, it further comprises a depth camera, which collects a new obstacle image, and according to the range after the region division, obtains the pixel feature point information of the new obstacle in the depth image information, and transmits it to the image processing module to judge whether the new obstacle blocks the stored planning path.

[0023] A robot vision positioning system, comprising: a memory for storing a computer program; a processor for executing the computer program to implement the aforementioned robot vision positioning method.

[0024] By adopting the environment perception method, device and system, the identification accuracy of obstacles in the robot travel path can be improved, and by setting the geographic fence of the planning path, the risk of navigation failure caused by collision between the robot and newly appeared obstacles can be reduced. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 is a method flowchart according to an embodiment of the present application;

[0026] Figure 2 is a grid division schematic diagram according to an embodiment of the present application;

[0027] Figure 3 is a geographic fence calculation schematic diagram according to an embodiment of the present application; DETAILED DESCRIPTION

[0028] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail with reference to the drawings. The description introduces specific embodiments consistent with the principles of the present application by way of examples but not by way of limitation, and the description of these embodiments is sufficient to enable those skilled in the art to practice the present application, other embodiments can be used and the structure of each element can be changed and / or replaced without departing from the scope and spirit of the present application. Therefore, the following detailed description should not be understood in a limiting sense. Specific embodiment 1

[0030] A robot vision positioning method, as shown in Figure 1 Periodically acquires image information in front of the robot, extracts obstacle image information from the image information, and generates an obstacle information image; at the same time, the robot positioning information is perceived and uploaded to the cloud, and the cloud calls the historical obstacle information image corresponding to the positioning location which has been saved for comparison; the specific comparison method is: for each obstacle appearing in the obstacle information image, the region is divided, and for each obstacle region, the label information containing the large verification mark is formed as the first verification information, and each obstacle region is divided into a plurality of small grids, each small grid contains small identification information as the second verification information;

[0031] The cloud calls the saved label information of the verification obstacle information corresponding to the positioning location to confirm whether the path is correct, first compares the first verification information of each obstacle, and if correct, compares the second verification information of each obstacle, selects the middle position small grid for comparison, and if the small grid label information is identified incorrectly, immediately finds the small grid adjacent thereto for label confirmation;

[0032] The first verification information and the second verification information are consistent, and the navigation is continued according to the planned path;

[0033] The first verification information comparison result is consistent, and the second verification information comparison result is inconsistent, and the real-time mode is entered;

[0034] The first verification information is inconsistent, and the obstacle with inconsistent information is considered to be a new obstacle appearing in front of the path, at this time, the robot pauses and enters the new obstacle judgment mode to judge whether it needs to change the path to enter the real-time mode.

[0035] Since the robot walks according to the map on which information has been recorded, the cloud map information can be called by comparing the image information collected, to guide the robot to walk the optimized path stored, and to ensure the accuracy of the use of the existing map information, so that the image information in the image collection and the mark information generate input image pixel characteristics points that are easy to compare with the image of the map information already stored, reducing the operation calculation, for example, whether there is a difference between the image information of the surrounding buildings, fixed objects and other reference objects collected by the positioning information combined with the above image information comparison and the cloud information, when there is no difference in the surrounding environment, the optimal route stored in the cloud can be directly used for walking guidance, reducing the reconstruction of the map each time. When comparing, the obstacle information image is extracted first, and each obstacle image is divided into grids and feature points and mark information are extracted for comparison, thereby saving the comparison of the entire image, thereby improving the comparison speed, for example, the large verification mark of the extracted building image can be whether the size and contour parameters of the building relative to the robot position are the same when the robot walks to the same position; and the small grid mark information can be whether the color parameters reflected in the building image are the same at the same position.

[0036] Further, the obstacle judgment mode specifically includes the following steps: region segmentation is performed on the new obstacle in the obstacle information image, the region range of the new obstacle is identified, a depth image information of the same positioning location is captured by calling a depth camera, pixel feature point information of the new obstacle in the depth image information is extracted according to the region range of the new obstacle, and the pixel feature point information includes three-dimensional coordinates of each pixel point of the new obstacle; all pixel feature point information constitutes a three-dimensional coordinate cluster, whether the new obstacle blocks the stored planning path is judged by comparing two-dimensional coordinates in the three-dimensional coordinate cluster with two-dimensional coordinates of the historical planning path saved in the cloud; if the new obstacle blocks the stored planning path, the real-time mode is switched to, otherwise the robot continues to navigate according to the historical path; in the real-time mode, the robot immediately scans the surrounding environment in real time, and plans and drives the route according to the actual surrounding environment.

[0037] A robot visual positioning device includes an image acquisition module, which periodically acquires image information in front of the robot; a positioning module, which acquires robot positioning information; and uploads the robot positioning information to the cloud; a cloud processing module, which calls the saved historical obstacle information image and sends it to the robot; an image processing module, which extracts obstacle image information from the image information, generates an obstacle information image, and compares it with the historical obstacle information image transmitted by the cloud.

[0038] Further, a depth camera is further included, which collects a new obstacle image, and according to the range after the region division, obtains pixel feature point information of the new obstacle in the depth image information, and transmits to the image processing module to judge whether the new obstacle blocks the stored planning path.

[0039] A robot vision positioning system, comprising: a memory for storing a computer program; a processor for implementing the aforementioned robot vision positioning method when executing the computer program. Specific embodiment 2

[0041] Referring to Figure 2 Periodically acquire image information in front of the robot, extract obstacle image information from the image information, and generate obstacle information image; at the same time, perceive the robot positioning information, and upload the robot positioning information to the cloud, and the cloud calls the saved historical obstacle information image corresponding to the positioning place for comparison; the specific comparison method is: for each obstacle appearing in the obstacle information image, region division is performed on it, and for each obstacle region, label information containing a large verification mark is formed as primary verification information, and each obstacle region is divided into a plurality of small grids, each small grid contains small identification information as secondary verification information;

[0042] The cloud calls the saved verification obstacle information label information corresponding to the positioning place to confirm whether the path is correct, first compares the primary verification information of each obstacle, and if correct, compares the secondary verification information of each obstacle, selects the middle position small grid for comparison, and if the small grid label information is identified incorrectly, immediately finds the small grid adjacent thereto for label confirmation;

[0043] The primary verification information and the secondary verification information are consistent, and the navigation is continued according to the planning path;

[0044] The primary verification information comparison result is consistent, and the secondary verification information comparison result is inconsistent, and the real-time mode is entered;

[0045] The primary verification information is inconsistent, and the obstacle with inconsistent information is considered to be a new obstacle appearing in front of the path, at this time, the robot pauses to travel, enters the new obstacle judgment mode, and judges whether it is necessary to change the path to enter the real-time mode.

[0046] In a preferred embodiment, the fixed obstacle image is collected at a certain walking position of the robot, the fixed obstacle image size contour parameters are selected as the large verification mark, and the local image of the obstacle image is selected for grid division. The selection of the local image is the local image of the middle rectangular region of the entire obstacle image, and the pixel data of each small grid related region of the image is uploaded to the cloud. The pixel characteristic points in the related region are selected, which can be points representing the pixel color and gray scale characteristics of the local image, and are marked with small mark information. The grid of the local image can be divided into 3X3 or 5X5 form, which facilitates the selection of the middle small grid for comparison.

[0047] Specifically, when the large verification mark is correct, the small mark information of the middle small grid is verified, and if it is consistent, it is considered that the image comparison result is consistent, and the robot continues to move according to the planned path.

[0048] When the large verification mark is correct, if the small mark information of the middle small grid does not match, the middle small grid o and the adjacent small grid are immediately checked. Preferably, as shown in Figure 2 The small grid a, small grid b, small grid c, and small grid around the small grid o are selected for verification. If all of them are consistent, it is considered that the small mark information is correct, otherwise, it is judged that the small mark is inconsistent, then the real-time mode is entered, the historical data information of the location environment is no longer received, and the robot is guided to move according to the real-time environment image information.

[0049] When the large verification mark is incorrect, it is considered that the inconsistent information in the current collected image is a new obstacle appearing in front of the path. At this time, the robot pauses and enters the new obstacle judgment mode to determine whether it needs to change the path and enter the real-time mode.

[0050] Preferably, the shape of the selected small grid is set to a square or a rectangle. At this time, the center point of the square or rectangular image can be calculated by a simple mathematical formula, thereby reducing the computational burden.

[0051] In specific practice, since the periodic acquisition of image information in front of the robot is a planar image, when a new obstacle is identified in the image, it is necessary to further determine whether the obstacle is located on the planning path stored by the robot. If so, switch to real-time mode, otherwise continue to navigate according to the historical path. Specifically, the obstacle determination mode includes the following steps: region segmentation of the new obstacle in the obstacle information image, identifying the region range of the new obstacle, calling the depth camera to shoot the depth image information of the same positioning place, extracting the pixel feature point information of the new obstacle in the depth image information according to the region range of the new obstacle, the pixel feature point information includes the three-dimensional coordinates of each pixel point of the new obstacle; all pixel feature point information constitutes a three-dimensional coordinate cluster, and whether the new obstacle blocks the stored planning path is determined by comparing the two-dimensional coordinates in the three-dimensional coordinate cluster with the historical planning path two-dimensional coordinates saved in the cloud; if it blocks, switch to real-time mode, otherwise continue to navigate according to the historical path. In real-time mode, the robot immediately scans the surrounding environment in real time, and plans and drives the route according to the actual surrounding environment.

[0052] A robot visual positioning device, comprising an image acquisition module, periodically acquiring image information in front of the robot; a positioning module, acquiring robot positioning information; and uploading the robot positioning information to the cloud, a cloud processing module, calling the saved historical obstacle information image, and sending it to the robot; an image processing module, extracting obstacle image information from the image information, generating an obstacle information image, and comparing it with the historical obstacle information image transmitted by the cloud.

[0053] Further, it further comprises a depth camera, which collects the image of the new obstacle, and according to the range after region division, obtains the pixel feature point information of the new obstacle in the depth image information, and transmits it to the image processing module to determine whether the new obstacle blocks the stored planning path.

[0054] A robot visual positioning system, comprising: a memory for storing a computer program; a processor for executing the computer program to implement the steps of the aforementioned robot visual positioning method. Specific embodiment 3

[0056] Reference Figures 2-3Periodically acquire image information in front of the robot, extract obstacle image information from the image information, and generate obstacle information images; at the same time, perceive the robot positioning information and upload the robot positioning information to the cloud, and the cloud calls the historical obstacle information images corresponding to the positioning location saved to perform comparison; the specific comparison method is that for each obstacle appearing in the obstacle information image, the region is divided, and for each obstacle region, label information containing a large verification mark is formed as primary verification information, and each obstacle region is divided into a plurality of small grids, each small grid contains small identification information as secondary verification information;

[0057] The cloud calls the saved label information of the verification obstacle information corresponding to the positioning location to confirm whether the path is correct, first compares the primary verification information of each obstacle, and if correct, compares the secondary verification information of each obstacle, selects the middle position small grid for comparison, and if the small grid label information is identified incorrectly, immediately finds the small grid adjacent thereto for label confirmation;

[0058] The primary verification information and the secondary verification information are consistent, and the navigation is continued according to the planned path;

[0059] The primary verification information comparison result is consistent, and the secondary verification information comparison result is inconsistent, and the real-time mode is entered;

[0060] The primary verification information is inconsistent, and the information inconsistent obstacle is considered to be a new obstacle appearing in front of the path, at this time, the robot pauses the travel and enters the new obstacle judgment mode to judge whether it is necessary to change the path to enter the real-time mode.

[0061] As a further improvement of the above technical solution, the obstacle judgment mode is specifically: region segmentation is performed on the new obstacle in the obstacle information image, the region range of the new obstacle is identified, a depth image information of the same positioning location is shot by calling a depth camera, pixel feature point information of the new obstacle in the depth image information is extracted according to the region range of the new obstacle, the pixel feature point information includes three-dimensional coordinates of each pixel point of the new obstacle; all pixel feature point information constitutes a three-dimensional coordinate cluster, whether the new obstacle blocks the stored planning path is judged by comparing two-dimensional coordinates in the three-dimensional coordinate cluster with the historical planning path two-dimensional coordinates saved in the cloud; if it blocks, it is switched to the real-time mode, otherwise the navigation is continued according to the historical path.

[0062] In practice, it is found that due to the certain volume of the robot itself, the distance factor between the positioning points and the obstacles is less considered in the existing path planning, and when a new obstacle appears, if the distance between the new obstacle and the positioning point is not enough for the robot to pass, but the position of the obstacle itself does not cover the navigation positioning point according to the path data, according to the navigation operation of the prior art, the robot will continue to advance according to the original path, which may cause the robot to collide with the newly appeared obstacle and cause navigation failure.

[0063] Next, on the basis of Embodiment 2 of the present application, the above technical solution is further optimized. The technical solution of combining the navigation route with the geographic fence is used to improve the above defects.

[0064] Further, the stored planning path is a geographic fence including positioning information and combining the robot contour, wherein the coordinates of the geographic fence are determined by the following method, please refer to Figure 3 , set the coordinates of the positioning point Am=(xi, yi) on the planning path, and the coordinates of the geographic fence on both sides of the positioning point corresponding to the path are Lm=(lxi, lyi) and Rm=(rxi, ryi), wherein the line connecting Lm and Rm is perpendicular to the line connecting Am and Am-1, and the intersection of the two lines is the midpoint of the line connecting Lm and Rm; in this way, the angles Figure 3 of the angles α and β can be calculated, and then γ=β-α and θ=90°-β-α are obtained; the formula for calculating the coordinates of the geographic fence is as follows:

[0065] (1);

[0066] (2);

[0067] Wherein, γ is the angle between the line connecting Am and Rm and the x-axis;

[0068] Through the above formula, the geographic fence coordinates Rm(rx i, ry i ) on the right side of the current positioning point Am can be calculated;

[0069] The geographic fence coordinates Lm(lx i, ly i ) on the left side of the current positioning point Am can be calculated by the following formula:

[0070] (3);

[0071] (4);

[0072] Wherein, θ is the angle between the line connecting Am-1 point and Lm point and the y-axis; S is the width of the robot planar projection contour, L is the length of the longest side of the robot planar projection contour or the diameter of the smallest circle that can accommodate the robot planar projection contour; △t is a margin set according to experience, generally, the value of △t is between (0, (L-S) / 2).

[0073] The current navigation positioning point and the previous navigation positioning point are taken as the advancing direction of the robot, and then the position coordinates of the geographic fence corresponding to the current positioning point are determined at the positions perpendicular to the advancing direction on both sides of the current positioning point, and the coordinate points of the adjacent two geographic fences on the same side are connected as the geographic fence of the robot navigation path, once a new obstacle is identified, the depth image of the current position is obtained through the depth camera, and whether the obstacle falls into the geographic fence is judged through the position of the pixel point in the obstacle image, which can more accurately identify the spatial position of the obstacle and prevent misjudgment, and further improve the accuracy of robot positioning and navigation.

[0074] A robot vision positioning device, comprising an image acquisition module, periodically acquiring image information in front of the robot; a positioning module, acquiring robot positioning information; and uploading the robot positioning information to the cloud, a cloud processing module, calling the saved historical obstacle information image and sending it to the robot; an image processing module, extracting obstacle image information from the image information, generating an obstacle information image, and comparing it with the historical obstacle information image transmitted by the cloud.

[0075] Further, it further comprises a depth camera, which collects a new obstacle image, and according to the range after the region division, obtains the pixel feature point information of the new obstacle in the depth image information, and transmits it to the image processing module to judge whether the new obstacle blocks the stored planning path.

[0076] A robot vision positioning system, comprising: a memory for storing a computer program; a processor for executing the computer program to implement the steps of the aforementioned robot vision positioning method.

[0077] Those skilled in the art can realize that, in combination with the method steps and units described in the embodiments disclosed in the present document, the present document can be realized in electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the steps and components of the embodiments have been described in the above description in a general manner. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present document. The method or steps described in combination with the embodiments disclosed in the present document can be implemented in hardware, software executed by a processor, or a combination of both. The software program can be stored in a storage medium such as a Random Access Memory (RAM), a memory, a Read-Only Memory (ROM), an Electrically Programmable Read-Only Memory (EPROM), an Electrically Erasable Programmable Read-Only Memory (EEPROM), a register, a hard disk, a removable disk, or the like.

[0078] Furthermore, other implementations of the present document will be apparent from the disclosure of the specification and drawings herein. The embodiments and / or aspects of the embodiments can be used alone or in any combination with one another. The specification and examples given should be considered exemplary only, with the actual scope and spirit of the present document being indicated by the appended claims.

Claims

1. A robot vision positioning method, characterized in that, periodically acquiring image information in front of the robot, extracting obstacle image information from the image information, and generating obstacle information images; sensing robot positioning information and uploading the robot positioning information to the cloud, and the cloud calling historical obstacle information images corresponding to the positioning location to perform comparison; the specific comparison method is that: for each obstacle appearing in the obstacle information image, the region is divided, and for each obstacle region, the label information containing the large verification mark is formed as the first verification information, and each obstacle region is divided into a plurality of small grids, each small grid contains small identification information as the second verification information; the cloud calls the saved label information of the verification obstacle information corresponding to the positioning location to confirm whether the path is correct, first compares the first verification information of each obstacle, and if correct, compares the second verification information of each obstacle, selects the middle position small grid for comparison, and if the small grid label information recognition is wrong, immediately finds the small grid adjacent to it for label confirmation; if the first verification information and the second verification information are consistent, the robot continues to navigate according to the planned path; if the first verification information is consistent and the second verification information is inconsistent, the real-time mode is entered; if the first verification information is inconsistent, it is considered that the obstacle with inconsistent information is a new obstacle appearing in front of the path, at this time, the robot pauses and enters the new obstacle judgment mode to determine whether to change the path to enter the real-time mode; the new obstacle judgment mode is specifically: the region of the new obstacle in the obstacle information image is segmented, the region range of the new obstacle is identified, the depth camera is called to shoot the depth image information of the same positioning location, the pixel feature point information of the new obstacle in the depth image information is extracted according to the region range of the new obstacle, the pixel feature point information includes the three-dimensional coordinates of each pixel point of the new obstacle; all pixel feature point information forms a three-dimensional coordinate cluster, whether the new obstacle blocks the stored planning path is judged by comparing the two-dimensional coordinates in the three-dimensional coordinate cluster with the historical planning path two-dimensional coordinates saved in the cloud; if it blocks, it switches to the real-time mode, otherwise it continues to navigate according to the historical path; wherein, the large verification mark selects fixed obstacle image size contour parameters, and the small verification mark is: selecting a local image of the obstacle image for grid division, the local image is a local image of the middle rectangular region of the entire obstacle image, and the pixel data of each small grid related region of the image is uploaded to the cloud, the pixel characteristic points in the related region are selected as the small verification mark.

2. The method of robot vision positioning according to claim 1, wherein, The real-time mode no longer receives historical data information of the location environment, and guides the robot to drive according to the real-time environment image information. 3.The robot vision positioning method of claim 2, characterized in that, The stored planning path is a geofence including positioning information and a robot contour, wherein coordinates of the geofence are determined by setting coordinates of a positioning point Am=(xi, yi) on the planning path, and coordinates of the geofence on both sides of the path corresponding to the positioning point are Lm=(lxi, lyi) and Rm=(rxi, ryi), wherein a line connecting the two points Lm and Rm is perpendicular to a line connecting the two points Am and Am-1, and a point of intersection of the two lines is a midpoint of the line connecting the two points Lm and Rm; wherein α is an included angle between a line connecting the point Am and the point Lm and a line connecting the point Am and the point Am-1; β is an included angle between a line connecting the point Am and the point Am-1 and an x-axis; γ is an included angle between a line connecting the point Am and the point Rm and the x-axis; γ = β - α; θ = 90° - β - α, and a formula for calculating the coordinates of the geofence is as follows: tan γ = (ryi - yi - 1) / (rxi - xi - 1) (1); a geofence coordinate Rm(rxi, ryi) on a right side of the current positioning point Am is calculated by the above formula; a geofence coordinate Lm(lxi, lyi) on a left side of the current positioning point Am is calculated by the following formula: tan θ = (lxi - xi - 1) / (lyi - yi - 1) (3); wherein θ is an included angle between a line connecting the point Am and the point Lm and a y-axis; S is a width of a robot planar projection contour, L is a length of a longest side of the robot planar projection contour or a diameter of a smallest circle capable of accommodating the robot planar projection contour; and Δt is a margin set according to experience, and a value of Δt is (0, (L-S) / 2).

4. A robotic vision positioning system, comprising: a memory for storing a computer program; a processor for implementing the robot visual positioning method of any one of claims 1-3 when executing the computer program.

Images