Feature point extraction and matching method and device, map making system and medium

Abstract

The application discloses a feature point extraction and matching method and device, a map making system and a medium, and belongs to the technical field of data processing. The method mainly comprises the following steps: converting 3D point cloud data under a 3D bird's-eye coordinate system into a 2D grayscale image by using the pre-established 3D bird's-eye coordinate system; extracting feature points from the 2D grayscale image to obtain feature information of the feature points; and matching the feature points by using the feature information, and taking the matched feature points as associated points. By providing a combination of 3D point cloud and image, the feature points in the driving process of the vehicle are extracted, the feature points are matched with historical frame feature points, and matching pairs with a matching score exceeding a threshold value are selected, so that accurate matching of the feature points of the vehicle is realized, accurate positioning of the vehicle is realized, and the experience effect of the user in the driving process is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a feature point extraction and matching method and device, a map making system, a medium and equipment. BACKGROUND

[0002] The positioning of a vehicle is an important embodiment of the correspondence between an autonomous vehicle and a 3D map. In the driving process of an autonomous vehicle, accurate positioning information can make the autonomous vehicle more accurately perceive the environment. In the process of positioning in a garage, feature extraction and matching are first needed, and the positioning information obtained by using feature point extraction and matching has high accuracy. Therefore, feature point extraction and matching are more popular in the positioning process.

[0003] In the traditional feature point extraction process, a pure image-based or pure point cloud-based extraction and matching method is often used to obtain feature points, and the feature points are used to realize the positioning of an autonomous vehicle.

[0004] However, whether the pure image-based or pure point cloud-based extraction and matching method is used, the nearest neighbor matching method is often used to obtain a matching result in the matching process. However, this method often causes matching errors. For example, after a vehicle drives for a period of time on a road and then returns to the starting point, the starting point and the ending point in the driving process should be able to be matched successfully. However, the ending point is a feature point after driving for a period of time, and the feature point has accumulated data. Therefore, the starting point and the ending point are not regarded as the nearest neighbors of each other in the nearest neighbor matching, so that the starting point and the ending point fail to be matched, causing a matching error, and further causing vehicle positioning errors, which reduces the user experience effect. SUMMARY

[0005] In view of the problem that the existing technology causes matching errors due to the influence of accumulated information, further causes vehicle positioning errors, and reduces the user experience effect, the present application mainly provides a feature point extraction and matching method and device, a map making system, a medium and equipment.

[0006] In a first aspect, the present application provides a feature point extraction and matching method, which includes: converting 3D point cloud data under a 3D bird's eye coordinate system into a 2D grayscale image by using a pre-established 3D bird's eye coordinate system; extracting feature points from the 2D grayscale image to obtain feature information of the feature points; extracting feature points from a bird's eye depth image to obtain feature information of the feature points; and matching the feature points by using the feature information, and taking the matched feature points as associated points.

[0007] In a second aspect, an embodiment of the present application provides a feature point extraction and matching device, which comprises: an image acquisition module, which converts 3D point cloud data under a pre-established 3D bird's eye coordinate system into a 2D grayscale image by using the 3D bird's eye coordinate system; a feature extraction module, which extracts feature points from the 2D grayscale image and acquires feature information of the feature points; and a feature matching module, which matches the feature points by using the feature information and takes the matched feature points as associated points.

[0008] In a third aspect, an embodiment of the present application provides a map making system, which comprises the feature point extraction and matching device, wherein the feature point extraction and matching device comprises: an image acquisition module, which converts 3D point cloud data under a pre-established 3D bird's eye coordinate system into a 2D grayscale image by using the 3D bird's eye coordinate system; a feature extraction module, which extracts feature points from the 2D grayscale image and acquires feature information of the feature points; and a feature matching module, which matches the feature points by using the feature information and takes the matched feature points as associated points.

[0009] In a fourth aspect, an embodiment of the present application provides a computer readable storage medium, which stores computer instructions, and the computer instructions are operated to execute the feature point extraction and matching method in the first aspect.

[0010] In a fifth aspect, an embodiment of the present application provides a computer device, which comprises: at least one processor; and a memory connected to the at least one processor in communication, wherein the memory stores computer instructions executable by the at least one processor, and the at least one processor operates the computer instructions to execute the feature point extraction and matching method in the first aspect.

[0011] The technical scheme of the embodiment of the present application realizes accurate matching of feature points of a vehicle by providing a 3D point cloud and image combined manner, extracting feature points in a vehicle driving process, matching the feature points with historical frame feature points, and selecting matching pairs with matching scores exceeding a threshold, thereby realizing accurate positioning of the vehicle and improving the experience effect of a user in a driving process. BRIEF DESCRIPTION OF DRAWINGS

[0012] In order to more clearly illustrate the technical schemes in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without any creative labor.

[0013] Figure 1 is a schematic diagram of an optional implementation of a feature point extraction and matching method of the present application;

[0014] Figure 2An optional example of a grid division method in the feature point extraction and matching method of the application is shown.

[0015] Figure 3 Another optional example of a grid division method in the feature point extraction and matching method of the application is shown.

[0016] Figure 4 is a schematic diagram of an optional embodiment of a feature point extraction and matching device of the application.

[0017] Through the above-mentioned drawings, the specific embodiments of the application have been shown, and will be described in more detail hereinafter. These drawings and written descriptions are not intended to limit the scope of the concept of the application in any way, but to illustrate the concept of the application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0018] The preferred embodiments of the application will be described in detail below with reference to the accompanying drawings, so that the advantages and features of the application can be more easily understood by those skilled in the art, and the scope of protection of the application can be more clearly defined.

[0019] It should be noted that in this document, relational terms such as first and second and the like can only be used to distinguish one entity or action from another entity or action, and do not necessarily require or imply that there is any such actual relationship or order between these entities or actions. Moreover, the terms "comprising", "including" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or apparatus that includes a list of elements does not only include those elements, but also includes other elements not explicitly listed, or further includes elements inherent in such a process, method, article or apparatus. Without more limitations, the elements defined by the statement "comprising" do not exclude the presence of additional identical elements in the process, method, article or apparatus that includes the elements defined by the statement "comprising".

[0020] The positioning of the vehicle is an important embodiment of the correspondence degree of the autonomous vehicle and the 3D map. During the driving process of the autonomous vehicle, accurate positioning information can make the perception of the environment of the autonomous vehicle more accurate. In the process of positioning in the garage, feature extraction and matching are first needed. The positioning information obtained by feature point extraction and matching has high accuracy, so feature point extraction and matching is more popular in the positioning process.

[0021] In the traditional feature point extraction process, the extraction and matching method based on pure image or pure point cloud is often used to obtain feature points, and the feature points are used to realize the positioning of the autonomous vehicle.

[0022] However, no matter whether the extraction and matching mode is based on pure image or pure point cloud, the nearest neighbor matching mode is usually adopted to obtain the matching result in the matching process, but this mode often causes matching errors; for example, when the vehicle drives on the road for a period of time and then returns to the original position, the starting point and the ending point in the driving process of the vehicle should be able to be matched successfully, but since the ending point is a feature point after driving for a period of time, the feature point has accumulated data, and therefore the starting point and the ending point are not regarded as the nearest neighbors of each other when the nearest neighbor matching is performed, so that the starting point and the ending point fail to be matched, causing a matching error, and further causing vehicle positioning error, reducing the user experience effect.

[0023] In view of the problem in the prior art that the matching error is caused due to the influence of accumulated information, further causing vehicle positioning error, reducing the user experience effect, the present application mainly provides a feature point extraction and matching method, device, map making system, medium and equipment. The feature point extraction and matching method comprises: converting 3D point cloud data collected by a laser radar into a 2D gray image by using a pre-established 3D bird's eye coordinate system; extracting feature points from the 2D gray image to obtain feature information of the feature points; and matching the feature points by using the feature information, and taking the matched feature points as associated points.

[0024] By providing a 3D point cloud and image combined mode, the feature points in the driving process of the vehicle are extracted, and the feature points are matched with historical frame feature points, and the matching pairs with matching scores exceeding a threshold value are selected, so as to realize accurate matching of the feature points of the vehicle, and further realize accurate positioning of the vehicle, so that the user experience effect in the driving process is improved.

[0025] In the following, the technical solutions of the present application and how the technical solutions of the present application solve the above technical problems will be described in detail with specific embodiments. The specific embodiments described below can be combined with each other to form new embodiments. For the same or similar ideas or processes described in one embodiment, they can not be described again in other embodiments. In the following, the embodiments of the present application will be described with reference to the accompanying drawings.

[0026] Figure 1 An optional embodiment of the feature point extraction and matching method of the present application is shown.

[0027] In Figure 1 In the optional embodiment shown, the feature point extraction and matching method mainly comprises the step S101 of converting 3D point cloud data under the 3D bird's eye coordinate system into a 2D gray image by using a pre-established 3D bird's eye coordinate system.

[0028] In the optional embodiment, the 3D point cloud data collected by the radar loaded on the autonomous vehicle is projected into the bird's eye coordinate system with the autonomous vehicle as the center of the bird's eye view, and the 3D point cloud data of the bird's eye coordinate system is converted to obtain a 2D grayscale image, which provides a condition for subsequent extraction of feature points from the 2D grayscale image.

[0029] In an optional embodiment of the present application, converting the 3D point cloud data collected by the laser radar into a 2D grayscale image further comprises: pre-establishing a 3D bird's eye coordinate system with the bird's eye view as the center; establishing a 3D grid of a preset range in the 3D bird's eye coordinate system; converting the 3D point cloud data in the 3D grid into a grayscale value to obtain a 2D grayscale image.

[0030] In the optional embodiment, a 3D bird's eye coordinate system is pre-established with the bird's eye view as the center, and a grid is established in the coordinate system, so that each 3D point cloud data scanned by the radar is in the grid. When the present scheme is applied on the ground, the preset range can be set relatively large because there are enough feature information on the ground; when the present scheme is applied underground, the preset range can be set relatively small because there are few feature information underground; which provides a basis for the accuracy of subsequent feature point extraction.

[0031] In an optional embodiment of the present application, the 3D grid of a preset range is established in the 3D bird's eye coordinate system, further comprising: dividing the 3D bird's eye coordinate system into 3D grids of equal size.

[0032] Figure 2 An optional example of the grid division method in the feature point extraction and matching method of the present application is shown.

[0033] In Figure 2 In the example shown, the grid is divided into equal-sized squares, Figure 2 The hollow circle in the grid is the position of the radar, i.e. the center point of the 3D bird's eye coordinate system, Figure 2 The black solid circle in the grid is the 3D point cloud data scanned by the radar. Preferably, when applied on the ground, the size of the grid can be set to 15*15CM, and when applied underground, the size of the grid can be set to 10*10CM.

[0034] It should be noted that the size of the preset range in the present application can also be other data, including but not limited to 5*5CM, 25*25CM; the examples of the present application only provide a preferred scheme, and the specific size of the preset range is not restricted, as long as the preset range set can provide the matching accuracy of the subsequent feature points.

[0035] In one optional embodiment of this application, a 3D mesh with a preset range is established in the 3D bird's-eye view coordinate system. The 3D bird's-eye view coordinate system is further divided into meshes according to the distance between the 3D bird's-eye view coordinate system and the axis, so that the 3D mesh with a larger distance from the axis has a larger radial width.

[0036] In this optional embodiment, since the 3D point cloud data collected by radar scanning exhibits a pattern where the closer the 3D point cloud data is to the radar, the denser the 3D point cloud data becomes, and the farther the 3D point cloud data is to the radar, the sparser the 3D point cloud data becomes. Therefore, in order to improve the accuracy of the 2D grayscale image obtained after converting the 3D point cloud data in the grid into grayscale values, the grid can be divided into square grids of different sizes. The closer the 3D point cloud data is to the radar, the smaller the grid is, and the farther the 3D point cloud data is to the radar, the larger the grid is. In two adjacent grids, the grid closer to the radar is no larger than the grid farther away from the radar.

[0037] In one optional embodiment of this application, establishing a 3D mesh within a preset range in a 3D bird's-eye view coordinate system further includes: drawing a circle with the axis of the 3D bird's-eye view coordinate system as the center to obtain concentric circles with unequal radii; and sending rays from the center of the circles according to a preset angle difference to obtain a 3D mesh, wherein the 3D mesh is a closed area enclosed by the arcs of the concentric circles and the rays.

[0038] Figure 3 This paper illustrates another optional example of the grid partitioning method in a feature point extraction and matching method of this application.

[0039] exist Figure 3 In the example shown, a 3D bird's-eye view coordinate system is established with the bird's-eye view as the center and the farthest distance scanned by the LiDAR as the radius. Figure 3 The hollow circle in the diagram represents the location of the lidar, i.e., the center of the 3D bird's-eye view coordinate system. The largest circle represents the farthest distance scanned by the lidar. Rays are emitted from the center of the circle in the 3D bird's-eye view coordinate system at fixed angle intervals, and circles are drawn at fixed length intervals to obtain... Figure 3 The concentric circles are arranged such that the radius difference between any two adjacent concentric circles is equal. Preferably, when applied to the ground, the fixed angle can be set to 45° and the radius difference can be set to 10 cm, in which case the radii of the concentric circles, from smallest to largest, are 10 cm, 20 cm, 30 cm, 40 cm, and 50 cm; when applied underground, the fixed angle can be set to 30° and the radius difference can be set to 5 cm, in which case the radii of the concentric circles, from smallest to largest, are 5 cm, 10 cm, 15 cm, 20 cm, and 25 cm.

[0040] Since the 3D point cloud data scanned by the laser radar is more dense when the distance from the laser radar is closer, and is more sparse when the distance from the laser radar is farther, in order to improve the accuracy of the 2D gray image obtained after the 3D point cloud data in the grid is converted into a gray value, the grid can be divided into grids of different sizes, that is, the fixed radius difference is changed to a variable radius difference, so that the closer the distance from the laser radar, the smaller the grid, and the farther the distance from the laser radar, the larger the grid, and the closer the distance from the laser radar in the two adjacent grids, the grid is not greater than the distance from the laser radar.

[0041] Therefore, the radius difference can be determined as an arithmetic progression value, for example, when the arithmetic interval is 5 cm, the radius difference of the concentric circles from inside to outside is 5 cm, 10 cm, 15 cm, 20 cm, 20 cm; At this time, the radii of the concentric circles from small to large are 5 cm, 15 cm, 30 cm, 50 cm, 75 cm. Or determine the radius difference as an irregular value that increases successively, as long as the previous radius difference is not greater than the next radius difference; For example, the radius difference of the concentric circles from inside to outside is set to 5 cm, 8 cm, 11 cm, 20 cm, 25 cm, and at this time, the radii of the concentric circles from small to large are 5 cm, 13 cm, 24 cm, 44 cm, 69 cm.

[0042] It should be noted that, Figure 3 The fixed angle and the size of the radius difference described in the above embodiment can also be other data; the examples of the present application only provide a preferred scheme, and the specific size of the fixed angle and the radius difference is not restricted, as long as the fixed angle and the size of the radius difference can improve the matching accuracy of the subsequent feature points.

[0043] In an optional embodiment of the present application, the 3D point cloud data in the 3D grid is converted into a gray value to obtain a 2D gray image, further comprising: calculating the average height and average variance of the 3D point cloud data in the 3D grid to obtain the height information and variance information corresponding to the 3D grid; obtaining the gray value of the 3D grid according to the height information and the variance information; and obtaining a 2D gray image from the gray value.

[0044] In the specific embodiment, after the grid is divided according to the above manner, the average height of the 3D point cloud data in the grid and the average variance of the height of the 3D point cloud data in the grid are calculated in the same grid; the average height of the 3D point cloud data in each grid and the average variance of the height of the 3D point cloud data in the grid are calculated according to the above calculation manner; the height information and the variance information corresponding to each grid obtained are the depth information of each grid, and the depth information of each grid is respectively mapped into the gray value of 0-255 to obtain the gray value corresponding to each grid, and the gray value corresponding to each grid is converted to obtain a 2D gray image; wherein the grid and the pixel in the 2D gray image are in a one-to-one correspondence. The 2D gray image obtained in the S102 step provides a basis for subsequent extraction of feature points from the image.

[0045] In Figure 1 In the optional implementation shown, the feature point extraction and matching method further includes a step S102 of extracting a feature point from the 2D gray image to obtain feature information of the feature point.

[0046] In the optional implementation, the feature point of the vehicle trajectory is extracted based on the 2D gray image to obtain feature point information of the feature point; wherein the feature point is extracted based on the image, which saves the computation amount of the system, occupies small memory, and improves the experience effect of the user.

[0047] In an optional embodiment of the present application, the feature point is extracted from the 2D gray image to obtain feature information of the feature point, which further includes: inputting the 2D gray image into a preset deep learning model to obtain the feature information.

[0048] In the optional embodiment, the 2D gray image is input into the preset deep learning model, and the preset deep learning model outputs the feature information of each feature point according to the environment where the feature point is located, wherein the feature information is description information of the environment where the feature point is located; the process of obtaining the feature information in the present scheme provides a condition for subsequent matching of the feature point.

[0049] In Figure 1 In the optional implementation shown, the feature point extraction and matching method further includes a step S103 of matching the feature points using the feature information, and taking the matched feature points as the associated points.

[0050] In the optional embodiment, the feature points are matched by using the acquired feature information, and the matched feature points are taken as the associated points. For example, after the autonomous vehicle travels along a road for a distance and then returns to the original position, if the traditional feature point matching mode is used, the two feature points at the same position cannot be matched because the feature points returned have more accumulated data than the feature points traveled. According to the matching mode based on the environmental description information of the present application, it is not necessary to consider whether the feature points are the feature points traveled or the feature points returned, but it is only necessary to consider whether the environments where the two feature points are located are similar or identical. The feature points located in similar or identical environments are determined as the associated points. The success rate of matching is improved, and the effect of matching affected by other factors is avoided.

[0051] In an optional embodiment of the present application, the feature points are matched by using the feature information, which further includes: inputting the feature information into a preset graph neural network model; and the graph neural network model matches the feature information with a similarity greater than a preset threshold.

[0052] In an optional embodiment of the present application, the preset threshold of the similarity is set in advance, and after the feature information is input into the preset graph neural network model, the graph neural network model analyzes the feature information two by two, respectively calculates the similarity of the two feature information, when the similarity is greater than the preset threshold, it is determined that the two feature points are matched successfully, and the graph neural network model outputs the matching result. The preset threshold is self-defined, as long as the accuracy of matching is not reduced, and preferably, the preset threshold can be set to 99%.

[0053] In an optional embodiment of the present application, after the matched feature points are taken as the associated points, it further includes: comparing the pose information of the associated points, and when the difference of the pose information is greater than a preset difference threshold, the associated points are corrected.

[0054] In the optional embodiment, after the matching is successful, the pose information of the associated points is compared, the difference between the pose information is calculated, and when the difference is greater than a preset difference threshold, it indicates that the positioning of the associated points is wrong, and therefore the pose information of the associated points is corrected to make the pose information coincide, thereby improving the positioning accuracy. Preferably, the preset difference threshold can be set to 20 meters.

[0055] Figure 4 An optional embodiment of a feature point extraction and matching device of the present application is shown.

[0056] In Figure 4In the optional embodiment shown, the feature point extraction and matching device mainly comprises: an image acquisition module 401, which converts 3D point cloud data under a 3D bird's eye coordinate system into a 2D grayscale image by using a pre-established 3D bird's eye coordinate system; a feature extraction module 402, which extracts feature points from the 2D grayscale image and obtains feature information of the feature points; and a feature matching module 403, which matches the feature points by using the feature information and takes the matched feature points as associated points.

[0057] In an optional embodiment of the present application, converting 3D point cloud data under a 3D bird's eye coordinate into a 2D grayscale image further comprises: pre-establishing a 3D bird's eye coordinate system with a bird's eye perspective as the center; establishing a 3D grid of a preset range in the 3D bird's eye coordinate system; converting 3D point cloud data in the 3D grid into a grayscale value to obtain a 2D grayscale image.

[0058] In an optional embodiment of the present application, converting 3D point cloud data in the 3D grid into a grayscale value to obtain a 2D grayscale image further comprises: calculating the average height and average variance of the 3D point cloud data in the 3D grid to obtain height information and variance information corresponding to the 3D grid; obtaining the grayscale value of the 3D grid according to the height information and the variance information; and obtaining the 2D grayscale image by using the grayscale value.

[0059] In an optional embodiment of the present application, extracting feature points from a 2D grayscale image and obtaining feature information of the feature points further comprises: inputting the 2D grayscale image into a pre-established deep learning model to obtain the feature information.

[0060] In an optional embodiment of the present application, matching the feature points by using the feature information further comprises: inputting the feature information into a pre-established graph neural network model, and matching the feature information with a similarity greater than a preset threshold by using the graph neural network model.

[0061] In an optional embodiment of the present application, after taking the matched feature points as associated points, the device further comprises: comparing pose information of the associated points, and correcting the associated points when a difference between the pose information is greater than a preset difference threshold.

[0062] In an optional embodiment of the present application, each functional module in the feature point extraction and matching device can be directly in hardware, in a software module executed by a processor, or in a combination of both.

[0063] The software module can reside in a RAM memory, a flash memory, a ROM memory, an EPROM memory, an EEPROM memory, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor, such that the processor can read information from, and write information to, the storage medium.

[0064] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof, etc. The general-purpose processor can be a microprocessor, but in the alternative, the processor can be any conventional processor, controller, microcontroller, or state machine. The processor can also be a combination of a computer and a DSP, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration. In alternative embodiments, the storage medium can be integral to the processor. The processor and the storage medium can reside in an ASIC. The ASIC can reside in a user terminal. In alternative embodiments, the processor and the storage medium can reside as discrete components in a user terminal.

[0065] The feature point extraction and matching device provided in the present application can be used to execute the feature point extraction and matching method described in any of the above embodiments, and has similar implementation principles and technical effects, which will not be described here again.

[0066] In another optional embodiment of the present application, a map making system is provided, which comprises a feature point extraction and matching device, wherein the feature point extraction and matching device comprises: an image acquisition module, which converts 3D point cloud data under a 3D bird's eye coordinate system into a 2D grayscale image by using the pre-established 3D bird's eye coordinate system; a feature extraction module, which extracts feature points from the 2D grayscale image and acquires feature information of the feature points; and a feature matching module, which matches the feature points by using the feature information, and takes the matched feature points as associated points.

[0067] The map making system provided in the present application can be used to execute the feature point extraction and matching method described in any of the above embodiments, and has similar implementation principles and technical effects, which will not be described here again.

[0068] In another optional embodiment of the present application, a computer readable storage medium is provided, which stores computer instructions, and the computer instructions are operated to execute the feature point extraction and matching method described in the above embodiments.

[0069] In another optional implementation of the present application, a computer device includes at least one processor, and a memory connected to the at least one processor in communication; wherein the memory stores computer instructions executable by the at least one processor, and the at least one processor operates the computer instructions to perform the feature point extraction and matching method described in the above embodiments.

[0070] In several embodiments provided in the present application, it should be understood that the disclosed apparatus and method can be implemented in other manners. For example, the described apparatus embodiments are merely schematic. The division of the units is merely a logical function division. There can be another division manner for the actual implementation, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections can be indirect couplings or communication connections through some interfaces, devices or units, and can be in electrical, mechanical or other forms.

[0071] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e., can be located in one place, or can be distributed on multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the present embodiment.

[0072] The above description is merely some embodiments of the present application, and is not intended to limit the patent scope of the present application. Any equivalent structure transformation made by using the content of the present application specification and drawings, or directly or indirectly applied to other related technical fields, is also included in the patent protection scope of the present application.

Claims

1. A method for extracting and matching feature points, characterized in that, include: Using a pre-established 3D bird's-eye view coordinate system, the 3D point cloud data in the 3D bird's-eye view coordinate system is converted into a 2D grayscale image. The conversion of the 3D point cloud data in the 3D bird's-eye view coordinate system into a 2D grayscale image includes: A 3D bird's-eye view coordinate system is pre-established with the autonomous vehicle as the center of the bird's-eye view. A 3D mesh with a preset range is established in the 3D bird's-eye view coordinate system; Converting the 3D point cloud data in the 3D mesh into grayscale values ​​to obtain a 2D grayscale image, wherein the step of converting the 3D point cloud data in the 3D mesh into grayscale values ​​to obtain a 2D grayscale image includes: Calculate the average height and average variance of the 3D point cloud data within the 3D mesh, and obtain the height information and variance information corresponding to the 3D mesh; Based on the height information and the variance information, the grayscale value of the 3D mesh is obtained; The 2D grayscale image is obtained using the grayscale values; Feature points are extracted from the 2D grayscale image to obtain the feature information of the feature points; The feature points are matched using the feature information, and the matched feature points are used as association points.

2. The feature point extraction and matching method according to claim 1, characterized in that, The step of establishing a 3D mesh within a preset range in the 3D bird's-eye view coordinate system further includes: The 3D bird's-eye view coordinate system is divided into 3D grids of equal size.

3. The feature point extraction and matching method according to claim 1, characterized in that, The step of establishing a 3D mesh within a preset range in the 3D bird's-eye view coordinate system also includes: The 3D bird's-eye view coordinate system is divided into grids based on the distance between the grid and the axis of the 3D bird's-eye view coordinate system, such that the 3D grid with a greater distance from the axis has a larger radial width.

4. The feature point extraction and matching method according to claim 1, characterized in that, The step of establishing a 3D mesh within a preset range in the 3D bird's-eye view coordinate system also includes: Draw a circle with the axis of the 3D bird's-eye view coordinate system as the center to obtain concentric circles with different radii; Based on a preset angle difference, a ray is sent from the center of the circle to obtain the 3D mesh, wherein the 3D mesh is a closed area enclosed by the arc of the concentric circle and the ray.

5. The feature point extraction and matching method according to claim 1, characterized in that, The step of extracting feature points from the 2D grayscale image and obtaining feature information of the feature points further includes: The 2D grayscale image is input into a preset deep learning model to obtain the feature information.

6. The feature point extraction and matching method according to claim 1, characterized in that, The matching of feature points using the feature information further includes: The feature information is input into a preset graph neural network model; The graph neural network model matches feature information with a similarity greater than a preset threshold.

7. The feature point extraction and matching method according to claim 1, characterized in that, After using the matched feature points as association points, the method further includes: The pose information of the associated points is compared, and when the difference in pose information is greater than a preset difference threshold, the associated points are corrected.

8. A device for extracting and matching feature points, characterized in that, include: The image acquisition module utilizes a pre-established 3D bird's-eye view coordinate system to convert 3D point cloud data in the 3D bird's-eye view coordinate system into a 2D grayscale image. This conversion includes: pre-establishing the 3D bird's-eye view coordinate system centered on the autonomous vehicle's bird's-eye view; establishing a 3D grid with a preset range within the 3D bird's-eye view coordinate system; converting the 3D point cloud data in the 3D grid into grayscale values ​​to obtain a 2D grayscale image. This conversion further includes: calculating the average height and average variance of the 3D point cloud data within the 3D grid to obtain height and variance information corresponding to the 3D grid; obtaining the grayscale value of the 3D grid based on the height and variance information; and using the grayscale value to obtain the 2D grayscale image. The feature extraction module extracts feature points from the 2D grayscale image and obtains the feature information of the feature points; The feature matching module uses the feature information to match the feature points and uses the matched feature points as association points.

9. A map creation system, characterized in that, The map creation system includes the feature point extraction and matching device as described in claim 8.

10. A computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are operated to perform the feature point extraction and matching method according to any one of claims 1-7.

11. A computer device, characterized in that, include: At least one processor; as well as A memory that is communicatively connected to the at least one processor; The memory stores computer instructions executable by the at least one processor, which operates the computer instructions to perform the feature point extraction and matching method as described in any one of claims 1-7.

Images