Vehicle-mounted positioning optimization method and system based on prior map, terminal and storage medium
By acquiring prior network maps and ground point cloud maps in urban canyon terrain, and using the YOLOv8 model and point cloud extraction model for feature recognition and clustering, and optimizing the navigation satellite model, the problem of inaccurate positioning in urban canyon terrain was solved, achieving robust and accurate positioning results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-24
- Publication Date
- 2026-04-10
AI Technical Summary
Existing technologies for positioning in urban canyon terrain suffer from high resource consumption and inaccurate positioning.
By acquiring prior network maps and ground point cloud maps of the target area, feature recognition is performed using the YOLOv8 model and point cloud extraction model. Clustering is then used to obtain manhole cover and point cloud feature maps. The approximate mean and variance of the matching pairs are calculated, confidence factors are constructed, and the navigation satellite model is optimized to obtain the trajectory route.
It achieves robust and accurate positioning in urban canyon terrain, balancing computational efficiency and recognition accuracy.
Smart Images
Figure CN121829580A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of geographic information analysis, and particularly relates to a vehicle positioning optimization method and system based on prior map, a terminal and a computer readable storage medium. BACKGROUND
[0002] The rapid development of mobile mapping, artificial intelligence and big data analysis and other technologies has greatly promoted the research of smart cities. Efficient and convenient road infrastructure monitoring and updating is an important part of smart cities, and these tasks require a large amount of manpower and material resources.
[0003] In addition, the multipath effect existing in urban canyons, especially in areas with high-rise buildings, often causes signal attenuation and interruption during mobile positioning.
[0004] Therefore, the prior art still needs to be improved and developed. SUMMARY
[0005] The main purpose of the present application is to provide a vehicle positioning optimization method and system based on prior map, a terminal and a computer readable storage medium, which aims to solve the problem of large resource consumption and inaccurate positioning of urban canyon terrain in the prior art.
[0006] To achieve the above purpose, the present application provides a vehicle positioning optimization method based on prior map, which comprises the following steps: Obtain a prior network map of a target area, convert the prior network map into a 2D image map, use a trained YOLOv8 model to perform feature recognition on the 2D image map, and obtain a manhole cover feature map; Obtain a ground point cloud map of the target area, use a trained point cloud extraction model to perform feature recognition on the ground point cloud map, and obtain a point cloud feature map; Cluster the manhole cover feature map and the point cloud feature map of the target area to obtain a plurality of independent clusters, and constrain all the independent clusters to obtain a plurality of matching pairs; Calculate the approximate mean and approximate variance of all the matching pairs, and construct a confidence factor according to the approximate mean and the approximate variance; Optimize the navigation satellite model using the confidence factor, and obtain the trajectory route of the target object in the target area using the optimized navigation satellite model.
[0007] Optionally, the prior map-based vehicle positioning optimization method, wherein the prior network map of the target area is obtained, the prior network map is converted into a 2D image map, a trained YOLOv8 model is used for feature recognition of the 2D image map, and a manhole cover feature map is obtained, specifically comprising: A prior network map of a target area is obtained, all vertices of the prior network map are projected vertically to a horizontal plane by using a network model, and elevation information of each projection point is recorded to obtain an elevation matrix; Orthorectification is performed on the elevation matrix, and the texture color of each vertex on the network model is projected onto the pixel after orthorectification to obtain a 2D image map; A YOLOv8 model is trained using a user-input labeled data set, the 2D image map is input into the trained YOLOv8 model, the trained YOLOv8 model extracts features of the vehicle trajectory in the 2D image map, and a plurality of manhole cover features are obtained; All the manhole cover features are re-projected into the coordinate system of the prior network map to obtain a manhole cover feature map.
[0008] Optionally, the prior map-based vehicle positioning optimization method, wherein the ground point cloud map of the target area is obtained, a trained point cloud extraction model is used for feature recognition of the ground point cloud map to obtain a point cloud feature map, specifically comprising: A user-input manhole cover point cloud data set is obtained, the manhole cover point cloud data set is input into an initial point cloud extraction model for training to obtain a point cloud extraction model; An initial point cloud map of the target area is obtained, and the initial point cloud map is segmented to obtain a ground point cloud map; The ground point cloud map is input into the point cloud extraction model for feature recognition to obtain a point cloud feature map of the target area.
[0009] Optionally, the prior map-based vehicle positioning optimization method, wherein the manhole cover feature map and the point cloud feature map of the target area are clustered to obtain a plurality of independent clusters, and all the independent clusters are constrained to obtain a plurality of matching pairs, and the prior map-based vehicle positioning optimization method further comprises: A plurality of feature sets in the target area are obtained, and the manhole cover feature map and the point cloud feature map corresponding to each feature set are determined; An iterative closest point method is used for feature registration of the manhole cover feature map and the point cloud feature map corresponding to all the feature sets to optimize the transformation matrix between the manhole cover feature map and the point cloud feature map, and a target transformation matrix is obtained: ; wherein, denotes a target change matrix, denotes a feature set, denotes a transformation matrix, denotes a feature set of points, denotes the nearest point in a feature set to .
[0010] Optionally, the prior map-based vehicle positioning optimization method, wherein the clustering processing of the manhole cover feature map and the point cloud feature map of the target area obtains a plurality of independent clusters, and the constraint of all the independent clusters obtains a plurality of matching pairs, specifically includes: the clustering processing of the manhole cover feature map and the point cloud feature map of the target area obtains a plurality of independent clusters: ; ; wherein, denotes an independent cluster of a feature set , denotes an independent cluster of a feature set , , and respectively denote the 1st, 2nd and cluster in , , and respectively denote the 1st, 2nd and cluster in ; determines the centroid of each cluster, and calculates the Euclidean distance between all the centroids in the independent cluster and all the centroids in other independent clusters, to obtain a plurality of matching pairs: ; wherein, denotes the Euclidean distance of the matching pair, and respectively denote the point cloud feature of a feature set and a feature set , and respectively denote the cluster of and , and respectively denote the and the first centroid, represents a preset threshold.
[0011] Optionally, the vehicle positioning optimization method based on a priori map, wherein the calculating the approximate mean and the approximate variance of all the matching pairs, constructing the confidence factor according to the approximate mean and the approximate variance, specifically comprises: calculating the approximate mean and the approximate variance of all the matching pairs according to all the Euclidean distances; constructing the confidence factor according to the similarity between each matching pair, the approximate mean and the approximate variance: ; ; wherein, represents the confidence factor, represents the covariance matrix, represents the first matching pair, and respectively represent the first point cloud feature of the feature set and the first point cloud feature of the feature set represents the exponential function, represents the distance between the first matching pair, represents the approximate mean, represents the approximate variance.
[0012] Optionally, the vehicle positioning optimization method based on a priori map, wherein the utilizing the confidence factor to optimize the navigation satellite model, and utilizing the optimized navigation satellite model to obtain the trajectory route of the target object in the target area, specifically comprises: obtaining an initial target function of the navigation satellite model, and adding the confidence factor to the initial target function to obtain a final target function: ; wherein, represents the final target function, represents the pseudo-range factor, represents the Doppler frequency factor, represents the inertial measurement unit pre-fusion factor, , and respectively represent the covariance matrices of , and ; The navigation satellite model is optimized by using the final target function to obtain a target navigation satellite model, and a trajectory route of a target object in the target area is obtained through the target navigation satellite model.
[0013] In addition, to achieve the above object, the application further provides a vehicle positioning optimization system based on a prior map, wherein the vehicle positioning optimization system based on the prior map comprises: A first feature extraction module is configured to obtain a prior network map of a target area, convert the prior network map into a 2D image map, perform feature recognition on the 2D image map by using a trained YOLOv8 model, and obtain a manhole cover feature map. A second feature extraction module is configured to obtain a ground point cloud map of the target area, perform feature recognition on the ground point cloud map by using a trained point cloud extraction model, and obtain a point cloud feature map. A feature fusion module is configured to perform clustering processing on the manhole cover feature map and the point cloud feature map of the target area to obtain a plurality of independent clusters, and perform constraint on all the independent clusters to obtain a plurality of matching pairs. A confidence adding module is configured to calculate an approximate mean and an approximate variance of all the matching pairs, and construct a confidence factor according to the approximate mean and the approximate variance. A model optimization module is configured to optimize a navigation satellite model by using the confidence factor, and obtain a trajectory route of a target object in the target area by using the optimized navigation satellite model.
[0014] In addition, to achieve the above object, the application further provides a terminal, wherein the terminal comprises a memory, a processor, and a vehicle positioning optimization program based on a prior map stored on the memory and executable on the processor, and the vehicle positioning optimization program based on the prior map implements the steps of the vehicle positioning optimization method based on the prior map when executed by the processor.
[0015] In addition, to achieve the above object, the application further provides a computer readable storage medium, wherein the computer readable storage medium stores a vehicle positioning optimization program based on a prior map, and the vehicle positioning optimization program based on the prior map implements the steps of the vehicle positioning optimization method based on the prior map when executed by a processor.
[0016] In the present application, the prior network map of the target area is obtained, the prior network map is converted into a 2D image map, the trained YOLOv8 model is used for feature recognition of the 2D image map, and a manhole cover feature map is obtained; the ground point cloud map of the target area is obtained, the trained point cloud extraction model is used for feature recognition of the ground point cloud map, and a point cloud feature map is obtained; the manhole cover feature map and the point cloud feature map of the target area are clustered to obtain a plurality of independent clusters, and all the independent clusters are constrained to obtain a plurality of matching pairs; the approximate mean and the approximate variance of all the matching pairs are calculated, and the confidence factor is constructed according to the approximate mean and the approximate variance; the navigation satellite model is optimized by using the confidence factor, and the trajectory of the target object in the target area is obtained by using the optimized navigation satellite model. The present application can effectively balance the calculation efficiency and the recognition accuracy, and realizes robust and accurate positioning. BRIEF DESCRIPTION OF DRAWINGS
[0017] Figure 1 is a flowchart of a preferred embodiment of the vehicle positioning optimization method based on the prior map of the present application; Figure 2 is a processing flowchart of a preferred embodiment of the vehicle positioning optimization method based on the prior map of the present application; Figure 3 is a manhole cover feature detection flowchart of a preferred embodiment of the vehicle positioning optimization method based on the prior map of the present application; Figure 4 is a point cloud detection flowchart of a preferred embodiment of the vehicle positioning optimization method based on the prior map of the present application; Figure 5 is a feature matching flowchart of a preferred embodiment of the vehicle positioning optimization method based on the prior map of the present application; Figure 6 is a factor graph optimization schematic diagram of a preferred embodiment of the vehicle positioning optimization method based on the prior map of the present application; Figure 7 is a schematic diagram of an experimental device of a preferred embodiment of the vehicle positioning optimization method based on the prior map of the present application; Figure 8 is a schematic diagram of the number of satellites of a preferred embodiment of the vehicle positioning optimization method based on the prior map of the present application; Figure 9 is a trajectory accuracy comparison diagram of a preferred embodiment of the vehicle positioning optimization method based on the prior map of the present application; Figure 10 is a comparison diagram of the offset trajectory of a preferred embodiment of the vehicle positioning optimization method based on the prior map of the present application; Figure 11is a comparison chart of the results of ablation experiments of a preferred embodiment of the vehicle positioning optimization method based on a priori map of the present application; Figure 12 is a structure diagram of a preferred embodiment of the vehicle positioning optimization system based on a priori map of the present application; Figure 13 is a structure diagram of a preferred embodiment of the terminal of the present application. DETAILED DESCRIPTION
[0018] In order to make the purpose, technical scheme and advantages of the present application clearer and more explicit, the present application will be further described in detail below with reference to the drawings and by taking examples. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.
[0019] The vehicle positioning optimization method based on a priori map according to the preferred embodiment of the present application, as shown in Figure 1 includes the following steps: Step S10, obtaining a prior network map of a target area, converting the prior network map into a 2D image map, using a trained YOLOv8 model to perform feature recognition on the 2D image map, and obtaining a manhole cover feature map.
[0020] As shown in Figure 2 , the present application realizes the positioning optimization process through two kinds of feature extraction, including the extraction of point cloud features and the extraction of manhole cover features.
[0021] Specifically, a prior network map of a target area is obtained, all vertices of the prior network map are projected vertically to a horizontal plane using a network model, and the elevation information of each projection point is recorded to obtain an elevation matrix; The elevation matrix is subjected to orthorectification processing, and the texture color of each vertex on the network model is projected onto the pixel after orthorectification processing to obtain a 2D image map; The YOLOv8 model is trained using a user-input labeled data set, the 2D image map is input into the trained YOLOv8 model, the trained YOLOv8 model performs feature extraction on the vehicle trajectory in the 2D image map, and a plurality of manhole cover features are obtained; All the manhole cover features are re-projected into the coordinate system of the prior network map to obtain a manhole cover feature map.
[0022] As shown in Figure 3 , the detection of road manhole cover features is based on the obtained 3D mesh map (such as Figure 3 (a) in the figure, i.e., the prior network map); in the embodiments disclosed in the present application, the original data in the prior network map is first converted (such as Figure 3(b) Orthophoto (i.e.) Figure 2 In the context of DOM (Document Object Model) and digital surface imagery (i.e.) Figure 2 In the DSM (Digital Surface Model), then, as Figure 3 As shown in (c), feature extraction is performed on the two image maps. For DSM, all vertices of the prior network map are vertically projected onto the horizontal reference plane, and the Z coordinate value (elevation information) of each projection point is recorded to generate an elevation matrix of a regular grid. For DOM, orthorectification is performed from a vertically downward perspective based on the elevation information provided by DSM (to eliminate distortion caused by perspective and terrain undulations), and the texture color on the prior network map is "rendered" onto the corresponding corrected pixels to generate an image map with a uniform scale.
[0023] Furthermore, through transfer learning, a YOLOv8 model is trained using a self-annotated dataset (i.e., Figure 2 The 2D feature detection network in the image is a real-time object detection model that supports image classification, object detection, and instance segmentation tasks. Using the trained model, manhole cover features (i.e., ...) are extracted from two types of image data. Figure 2 The system identifies feature 2) and filters it using vehicle trajectory data. Then, using elevation information from the DSM data, it reprojects the detected manhole cover features back to their original coordinates in the same coordinate system as the prior network map, ensuring spatial accuracy and correlation, thus obtaining a manhole cover feature map (e.g., ...). Figure 3 (d) in the middle.
[0024] Step S20: Obtain the ground point cloud map of the target area, and use the trained point cloud extraction model to perform feature recognition on the ground point cloud map to obtain a point cloud feature map.
[0025] Among them, such as Figure 4 As shown, point cloud data of manhole covers is collected by a radar sensor (LiDAR sensor) to obtain a 3D map (i.e., the initial point cloud map), and further processing is performed to obtain a point cloud feature map.
[0026] Specifically, the point cloud dataset of manhole covers input by the user is obtained, and the point cloud dataset of manhole covers is input into the initial point cloud extraction model for training to obtain the point cloud extraction model; Obtain an initial point cloud map of the target area, and segment the initial point cloud map to obtain a ground point cloud map; The ground point cloud map is input into the point cloud extraction model for feature recognition to obtain the point cloud feature map of the target area.
[0027] Among them, such as Figure 4(a) of (b) in the specification, the 3D point cloud map is obtained by extracting the manhole cover point cloud dataset of the target region, and then the ground is segmented (such as Figure 4 (b) in the specification) to distinguish the ground point cloud and the non-ground point cloud, so as to reduce the data complexity by isolating the ground points from other feature points, thereby facilitating subsequent analysis and simplifying the subsequent point cloud feature recognition process.
[0028] Further, the point cloud extraction model is trained by the manually labeled manhole cover point cloud data, so as to obtain a model capable of identifying the accurate position of the manhole cover, and the feature detection (such as Figure 4 (c) in the specification) is performed on the ground point cloud map, so that the point cloud feature map (such as Figure 4 (d) in the specification) of the target region is obtained. The labeled dataset not only supports the initial training, but also serves as a benchmark for evaluating the performance of the model, and then the traditional point cloud filtering method is applied to remove outliers, thereby improving the accuracy of the point cloud feature extraction.
[0029] Further, a plurality of feature sets in the target region are obtained, and a manhole cover feature map and a point cloud feature map corresponding to each feature set are determined. The iterative closest point method is used to perform feature registration on the manhole cover feature map and the point cloud feature map corresponding to all the feature sets, so as to optimize the transformation matrix between the manhole cover feature map and the point cloud feature map, and obtain a target transformation matrix: ; wherein, represents the target transformation matrix, represents the feature set, represents the transformation matrix, represents the feature set in the specification, represents the closest point to in the feature set .
[0030] Wherein, two manhole cover feature maps are detected from two different data sources after the feature detection network; and in order to integrate the factor graph set subsequently, it is crucial to perform accurate feature matching, therefore, in the embodiments disclosed in the specification, two transformation matrices of the manhole cover feature maps are introduced, which are used for the fusion of the two features.
[0031] Step S30: performing clustering processing on the manhole cover feature map and the point cloud feature map of the target region to obtain a plurality of independent clusters, and performing constraint on all the independent clusters to obtain a plurality of matching pairs.
[0032] In one of the embodiments disclosed in this invention, an RCCM (Root Cause Assessment and Countermeasure Development, registration-based center clustering matching) method is introduced. This method enables the aggregation of manhole cover feature maps and point cloud feature maps (e.g., Figure 5 (As shown).
[0033] Specifically, clustering is performed on the manhole cover feature map and the point cloud feature map of the target area to obtain multiple independent clusters: ; ; in, Represents the feature set Independent clustering, Represents the feature set Independent clustering, , and They represent The first, second and third in One cluster, , and They represent The first, second and third in One cluster; Determine the centroid of each cluster, and calculate the Euclidean distance between all centroids in each independent cluster and all centroids in the other independent clusters, to obtain multiple matching pairs: ; in, Indicates the first The Euclidean distance between a matching pair and Representing feature sets respectively and feature set The Point cloud features, and They represent and The One cluster, and They represent and The The center of mass, This indicates a preset threshold.
[0034] First, two feature sets are defined for the target area. Each feature set corresponds to a manhole cover feature map and a point cloud feature map. Then, the manhole cover feature map and the point cloud feature map in each feature set are registered using a registration technique. This process can be achieved using the transformation matrix generated by the above process.
[0035] However, directly using the iterative nearest-point method for registration has limitations (matching points are often concentrated in corner areas). Therefore, in the embodiments disclosed in this invention, the two feature sets are then clustered to generate independent manhole cover clusters and their corresponding labels, dividing the point cloud into independent clusters, each defined as a single manhole cover. Further, the centroid of each cluster is calculated to form a geometric descriptor for matching; then, matching pairs are determined based on the Euclidean distance between the centroids under a preset threshold constraint.
[0036] Step S40: Calculate the approximate mean and approximate variance of all the matching pairs, and construct a confidence factor based on the approximate mean and approximate variance.
[0037] Based on the aforementioned feature matching results (i.e., matching pairs), in the embodiments disclosed in this invention, a factor graph optimization method integrating GNSS (Global Navigation Satellite System), INS (Inertial Navigation System), and deep learning-based manhole cover features is proposed.
[0038] Specifically, based on all the said Euclidean distances, the approximate mean and approximate variance of all the said matching pairs are calculated; A confidence factor is constructed based on the similarity between each matching pair, the approximate mean, and the approximate variance: ; ; in, Indicates the confidence factor. Represents the covariance matrix. Indicates the first One matching pair, and Representing feature sets respectively and feature set The Point cloud features, denotes an exponential function, denotes the distance between the th matching pair, denotes the approximate mean, denotes the approximate variance.
[0039] In the matching process, the proposed matching method may fail, and the matching factor determined only by the nearest neighbor principle may introduce errors. Therefore, in the embodiments of the present disclosure, spatial consistency is introduced as a confidence factor for verification. By incorporating spatial context information into the decision-making process, the matching result is verified and the overall system reliability is improved.
[0040] In step S50, the navigation satellite model is optimized using the confidence factor, and the trajectory of the target object in the target area is obtained using the optimized navigation satellite model.
[0041] In the embodiments of the present disclosure, a plurality of factors (such as Figure 6 are shown, wherein, , , , , and respectively represent different matching pairs), and the matching with a difference closer to the mean is given a higher confidence, thereby effectively excluding abnormal matching that deviates significantly from the expected value, and significantly improving the accuracy of feature extraction.
[0042] Specifically, an initial objective function of the navigation satellite model is obtained, and the confidence factor is added to the initial objective function to obtain a final objective function: ; wherein, denotes the final objective function, denotes the pseudorange factor, denotes the Doppler frequency factor, denotes the inertial measurement unit pre-fusion factor, , and respectively represent the covariance matrices of , and ; The navigation satellite model is optimized using the final objective function to obtain a target navigation satellite model, and the trajectory of the target object in the target area is obtained through the target navigation satellite model.
[0043] Further, in the embodiments of the present disclosure, as shown in Figure 7The data acquisition equipment shown is used to perform the above method. The number of satellites in the experimental section is as follows: Figure 8 As shown, the final results of the trajectory accuracy comparison under different methods were obtained (e.g.) Figure 9 and Figure 10 As shown in the figure, the first method represents the joint FGO (Factor Graph Optimization) method of GNSS and INS, the second method represents the joint FGO method of GNSS, INS and SLAM (Simultaneous Localization and Mapping), the third method represents the joint PICP (Point-to-Point Iterative Closest Point) method of GNSS, INS and DLMC, and the fourth method represents the joint FGO method of GNSS, INS and DLMC (i.e., the method disclosed in this invention).
[0044] Finally, as Figure 11 As shown, the factor graph optimization method for integrating GNSS, INS, and deep learning-based manhole cover features disclosed in this invention can effectively balance computational efficiency and feature recognition accuracy, achieving robust and accurate positioning.
[0045] Furthermore, such as Figure 12 As shown, based on the above-described vehicle positioning optimization method based on prior maps, the present invention also provides a vehicle positioning optimization system based on prior maps, wherein the vehicle positioning optimization system based on prior maps includes: The first feature extraction module 51 is used to obtain a prior network map of the target area, convert the prior network map into a 2D image map, and use the trained YOLOv8 model to perform feature recognition on the 2D image map to obtain a manhole cover feature map. The second feature extraction module 52 is used to obtain a ground point cloud map of the target area, and to perform feature recognition on the ground point cloud map using a trained point cloud extraction model to obtain a point cloud feature map. The feature fusion module 53 is used to perform clustering processing on the manhole cover feature map and the point cloud feature map of the target area to obtain multiple independent clusters, and to constrain all the independent clusters to obtain multiple matching pairs. The confidence addition module 54 is used to calculate the approximate mean and approximate variance of all the matching pairs, and construct a confidence factor based on the approximate mean and approximate variance; The model optimization module 55 is configured to optimize the navigation satellite model based on the confidence factor, and obtain a trajectory of the target object in the target area based on the optimized navigation satellite model.
[0046] Further, as shown in Figure 13 Based on the above-mentioned vehicle positioning optimization method and system based on a priori map, the application further provides a terminal, which comprises a processor 10, a memory 20 and a display 30. Figure 13 Only some components of the terminal are shown, but it should be understood that all the shown components are not required, and more or less components can be alternatively implemented.
[0047] The memory 20 can be an internal storage unit of the terminal, such as a hard disk or a memory of the terminal, in some embodiments. The memory 20 can also be an external storage device of the terminal, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc., in other embodiments. Further, the memory 20 can include both an internal storage unit and an external storage device of the terminal. The memory 20 is configured to store application software and various data installed on the terminal, such as program codes of the terminal, etc. The memory 20 can also be configured to temporarily store data that has been output or is to be output. In an embodiment, the memory 20 stores a vehicle positioning optimization program based on a priori map 40, which can be executed by the processor 10, so as to implement the vehicle positioning optimization method based on a priori map in the application.
[0048] The processor 10 can be a central processing unit (CPU), a microprocessor or other data processing chip, in some embodiments, which is configured to run program codes or process data stored in the memory 20, such as to execute the vehicle positioning optimization method based on a priori map, etc.
[0049] The display 30 can be an LED display, a liquid crystal display, a touch liquid crystal display, an OLED (Organic Light-Emitting Diode) touch, etc., in some embodiments. The display 30 is configured to display information of the terminal and display a visualized user interface. The components of the terminal communicate with each other through a system bus.
[0050] In an embodiment, the following steps are implemented when the processor 10 executes the vehicle positioning optimization program based on a priori map 40 in the memory 20: obtaining a prior network map of a target area, converting the prior network map into a 2D image map, performing feature recognition on the 2D image map by using a trained YOLOv8 model, and obtaining a manhole cover feature map; obtaining a ground point cloud map of the target area, performing feature recognition on the ground point cloud map by using a trained point cloud extraction model, and obtaining a point cloud feature map; performing clustering processing on the manhole cover feature map and the point cloud feature map of the target area to obtain a plurality of independent clusters, and performing constraint on all the independent clusters to obtain a plurality of matching pairs; calculating an approximate mean and an approximate variance of all the matching pairs, and constructing a confidence factor according to the approximate mean and the approximate variance; optimizing a navigation satellite model by using the confidence factor, and obtaining a trajectory route of a target object in the target area by using the optimized navigation satellite model.
[0051] The obtaining of the prior network map of the target area, the conversion of the prior network map into the 2D image map, the performance of the feature recognition on the 2D image map by using the trained YOLOv8 model, and the obtaining of the manhole cover feature map specifically include: obtaining a prior network map of a target area, converting the prior network map into a 2D image map, performing feature recognition on the 2D image map by using a trained YOLOv8 model, and obtaining a manhole cover feature map; performing orthorectification processing on the height matrix, and projecting the texture color of each vertex on the network model onto the pixel after the orthorectification processing to obtain a 2D image map; training a YOLOv8 model by using a user-input labeled data set, inputting the 2D image map into the trained YOLOv8 model, performing feature extraction on the vehicle-mounted trajectory in the 2D image map by using the trained YOLOv8 model, and obtaining a plurality of manhole cover features; reprojecting all the manhole cover features into the coordinate system of the prior network map to obtain a manhole cover feature map.
[0052] The obtaining of the ground point cloud map of the target area, the performance of the feature recognition on the ground point cloud map by using the trained point cloud extraction model, and the obtaining of the point cloud feature map specifically include: obtaining a user-input manhole cover point cloud data set, inputting the manhole cover point cloud data set into an initial point cloud extraction model for training, and obtaining a point cloud extraction model; obtaining an initial point cloud map of the target area, performing segmentation processing on the initial point cloud map, and obtaining a ground point cloud map; Input the ground point cloud map into the point cloud extraction model for feature recognition to obtain a point cloud feature map of the target region.
[0053] The clustering processing of the manhole cover feature map and the point cloud feature map of the target region obtains a plurality of independent clusters, and the constraint of all the independent clusters obtains a plurality of matching pairs, and the previous further includes: Obtain a plurality of feature sets in the target region, and determine the manhole cover feature map and the point cloud feature map corresponding to each feature set; Perform feature registration on the manhole cover feature map and the point cloud feature map corresponding to all the feature sets by using an iterative closest point method to optimize the transformation matrix between the manhole cover feature map and the point cloud feature map, and obtain a target transformation matrix: ; Wherein, represents a target transformation matrix, represents a feature set, represents a transformation matrix, represents a point in the feature set , represents the closest point to in the feature set .
[0054] The clustering processing of the manhole cover feature map and the point cloud feature map of the target region obtains a plurality of independent clusters, and the constraint of all the independent clusters obtains a plurality of matching pairs, and the previous further includes: The clustering processing of the manhole cover feature map and the point cloud feature map of the target region obtains a plurality of independent clusters: ; ; Wherein, represents an independent cluster of the feature set , represents an independent cluster of the feature set , , and respectively represent the 1st, 2nd and cluster in , , and respectively represent the 1st, 2nd and cluster in ; Determine the centroid of each cluster, and calculate the Euclidean distance between all centroids in each independent cluster and all centroids in the other independent clusters, to obtain multiple matching pairs: ; in, Indicates the first The Euclidean distance between a matching pair and Representing feature sets respectively and feature set The Point cloud features, and They represent and The One cluster, and They represent and The The center of mass, This indicates a preset threshold.
[0055] However, the step of calculating the approximate mean and approximate variance of all the matched pairs, and constructing a confidence factor based on the approximate mean and approximate variance, specifically includes: Based on all the Euclidean distances, calculate the approximate mean and approximate variance of all the matching pairs; A confidence factor is constructed based on the similarity between each matching pair, the approximate mean, and the approximate variance: ; ; in, Indicates the confidence factor. Represents the covariance matrix. Indicates the first One matching pair, and Representing feature sets respectively and feature set The Point cloud features, Represents an exponential function. Indicates the first The distance between matching pairs This represents the approximate mean. This represents the approximate variance.
[0056] The step of optimizing the navigation satellite model using the confidence factor and obtaining the trajectory of the target object in the target area using the optimized navigation satellite model specifically includes: An initial objective function of a navigation satellite model is acquired, and a confidence factor is added to the initial objective function to obtain a final objective function: ; Wherein, represents the final objective function, represents a pseudorange factor, represents a Doppler frequency factor, represents an inertial measurement unit pre-fusion factor, , and respectively represent , and covariance matrices of ; The navigation satellite model is optimized using the final objective function to obtain a target navigation satellite model, and a target object trajectory in the target area is obtained through the target navigation satellite model.
[0057] The application also provides a computer readable storage medium, wherein the computer readable storage medium stores a vehicle positioning optimization program based on a prior map, and the vehicle positioning optimization program based on the prior map, when executed by a processor, implements the steps of the vehicle positioning optimization method based on the prior map.
[0058] In summary, the application provides a vehicle positioning optimization method based on a prior map and related equipment, the method comprising: acquiring a prior network map of a target area, converting the prior network map into a 2D image map, using a trained YOLOv8 model to perform feature recognition on the 2D image map to obtain a manhole cover feature map; acquiring a ground point cloud map of the target area, using a trained point cloud extraction model to perform feature recognition on the ground point cloud map to obtain a point cloud feature map; performing clustering processing on the manhole cover feature map and the point cloud feature map of the target area to obtain a plurality of independent clusters, and performing constraint on all the independent clusters to obtain a plurality of matching pairs; calculating the approximate mean and approximate variance of all the matching pairs, constructing a confidence factor according to the approximate mean and the approximate variance; using the confidence factor to optimize a navigation satellite model, and using the optimized navigation satellite model to obtain a target object trajectory in the target area.
[0059] The application can effectively balance the calculation efficiency and the recognition accuracy, and realize robust and accurate positioning.
[0060] It should be noted that, in the present document, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element preceded by "comprises a" does not, without more constraints, foreclose the existence of additional identical elements in the process, method, article, or apparatus that comprises the element.
[0061] Of course, those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing relevant hardware (such as a processor, a controller, etc.) through a computer program. The program can be stored in a computer readable computer readable storage medium, and the program can include the processes of the above-mentioned method embodiments when executed. The computer readable storage medium can be a memory, a magnetic disc, an optical disc, etc.
[0062] It should be understood that the application is not limited to the above examples, and those skilled in the art can make improvements or changes according to the above description, and all these improvements and changes shall fall within the protection scope of the appended claims of the present application.
Claims
1. A priori map-based vehicle positioning optimization method, characterized in that, The prior map-based vehicle positioning optimization method comprises: obtaining a prior network map of a target area, converting the prior network map into a 2D image map, performing feature recognition on the 2D image map by using a trained YOLOv8 model to obtain a manhole cover feature map; obtaining a ground point cloud map of the target area, performing feature recognition on the ground point cloud map by using a trained point cloud extraction model to obtain a point cloud feature map; performing clustering processing on the manhole cover feature map and the point cloud feature map of the target area to obtain a plurality of independent clusters, and performing constraint on all the independent clusters to obtain a plurality of matching pairs; calculating approximate mean and approximate variance of all the matching pairs, and constructing a confidence factor according to the approximate mean and the approximate variance; optimizing a navigation satellite model by using the confidence factor, and obtaining a trajectory route of a target object in the target area by using the optimized navigation satellite model.
2. The priori map-based vehicle positioning optimization method according to claim 1, wherein, The prior network map of the target area is obtained, all vertices of the prior network map are projected vertically to a horizontal plane by using a network model, and elevation information of each projection point is recorded to obtain an elevation matrix. Orthorectification is performed on the elevation matrix, and the texture color of each vertex on the network model is projected onto the pixel after orthorectification processing to obtain a 2D image map. The YOLOv8 model is trained by using a user-input label data set, the 2D image map is input into the trained YOLOv8 model, the vehicle trajectory in the 2D image map is feature-extracted by the trained YOLOv8 model, and a plurality of manhole cover features are obtained. All the manhole cover features are re-projected into the coordinate system of the prior network map to obtain a manhole cover feature map. The ground point cloud map of the target area is obtained, and the feature recognition is performed on the ground point cloud map by using the trained point cloud extraction model to obtain a point cloud feature map.
3. The priori map-based vehicle positioning optimization method according to claim 1, wherein, The manhole cover point cloud data set input by the user is obtained, the manhole cover point cloud data set is input into the initial point cloud extraction model for training to obtain a point cloud extraction model. The initial point cloud map of the target area is obtained, and the initial point cloud map is segmented to obtain a ground point cloud map. The ground point cloud map is input into the point cloud extraction model for feature recognition to obtain the point cloud feature map of the target area. The plurality of feature sets in the target area are obtained, and the manhole cover feature map and the point cloud feature map corresponding to each feature set are determined.
4. The priori map-based vehicle positioning optimization method of claim 1, wherein, The feature registration is performed on the manhole cover feature map and the point cloud feature map corresponding to all the feature sets by using an iterative closest point method to optimize a transformation matrix between the manhole cover feature map and the point cloud feature map, and a target transformation matrix is obtained. ； wherein, denotes a target change matrix, denotes a feature set, denotes a transformation matrix, denotes a feature set of points, denotes the closest point in the feature set to .
5. The priori map-based vehicle positioning optimization method of claim 1, wherein, The clustering processing of the manhole cover feature map and the point cloud feature map of the target region obtains a plurality of independent clusters, and the constraint of all the independent clusters obtains a plurality of matching pairs, specifically including: The clustering processing of the manhole cover feature map and the point cloud feature map of the target region obtains a plurality of independent clusters: ； ； wherein denotes an independent cluster of the feature set , denotes an independent cluster of the feature set , , and denote the 1st, 2nd and the 3rd cluster in , , and denote the 1st, 2nd and the 3rd cluster in ; The centroid of each cluster is determined, and the Euclidean distance between all the centroids in the independent cluster and all the centroids in other independent clusters is calculated to obtain a plurality of matching pairs: ； wherein, denotes the Euclidean distance of the th matching pair, and denote the th point cloud feature of the feature set and the feature set , respectively, and denote the th cluster of the th point cloud feature of the feature set and the feature set , respectively, and denote the th centroid of the th cluster of the feature set denotes a preset threshold value.
6. The priori map-based vehicle positioning optimization method according to claim 5, wherein, The approximate mean and the approximate variance of all the matching pairs are calculated, and a confidence factor is constructed according to the approximate mean and the approximate variance, specifically including: According to all the Euclidean distances, the approximate mean and the approximate variance of all the matching pairs are calculated; According to the similarity between each matching pair, the approximate mean and the approximate variance, a confidence factor is constructed: ； ； wherein, denotes a confidence factor, denotes a covariance matrix, denotes the th matching pair, and denote the th point cloud feature of the feature set and the feature set , respectively, denotes an exponential function, denotes the distance between the th matching pair, denotes an approximated mean, denotes an approximated variance.
7. The priori map-based vehicle positioning optimization method according to claim 6, wherein, The navigation satellite model is optimized by using the confidence factor, and the trajectory route of the target object in the target region is obtained by using the optimized navigation satellite model, specifically including: An initial target function of a navigation satellite model is obtained, and the confidence factor is added to the initial target function to obtain a final target function: ； wherein, represents the final target function, represents a pseudorange factor, represents a Doppler frequency factor, represents an inertial measurement unit pre-fusion factor, , and respectively represent , and covariance matrices of The navigation satellite model is optimized by using the final target function to obtain a target navigation satellite model, and the trajectory route of the target object in the target region is obtained by using the target navigation satellite model.
8. A priori map based vehicle positioning optimization system, characterized by, The vehicle positioning optimization system based on a priori map is applied to the vehicle positioning optimization method based on a priori map as claimed in any one of claims 1-7, and the vehicle positioning optimization system based on a priori map comprises: A first feature extraction module is configured to obtain a priori network map of a target region, convert the priori network map into a 2D image map, perform feature recognition on the 2D image map by using a trained YOLOv8 model, and obtain a manhole cover feature map; A second feature extraction module is configured to obtain a ground point cloud map of the target region, perform feature recognition on the ground point cloud map by using a trained point cloud extraction model, and obtain a point cloud feature map; A feature fusion module is configured to perform clustering processing on the manhole cover feature map and the point cloud feature map of the target region to obtain a plurality of independent clusters, and perform constraint on all the independent clusters to obtain a plurality of matching pairs; A confidence adding module is configured to calculate the approximate mean and the approximate variance of all the matching pairs, and construct a confidence factor according to the approximate mean and the approximate variance; A model optimization module is configured to optimize a navigation satellite model by using the confidence factor, and obtain the trajectory route of the target object in the target region by using the optimized navigation satellite model.
9. A terminal, characterized by comprising: The terminal comprises a memory, a processor and a priori map based on-board positioning optimization program stored on the memory and executable on the processor, the priori map based on-board positioning optimization program, when executed by the processor, implements the steps of the priori map based on-board positioning optimization method according to any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a priori map based on-board positioning optimization program, the priori map based on-board positioning optimization program, when executed by the processor, implements the steps of the priori map based on-board positioning optimization method according to any one of claims 1-7.