A vehicle fusion positioning method, apparatus, device, medium and product

Abstract

The application discloses a vehicle fusion positioning method, device, equipment, medium and product, and relates to the technical field of intelligent transportation. The method comprises the following steps: acquiring environment perception data, map priori data and multi-source positioning sensor data of a target vehicle; performing calculation on the map priori data and the multi-source positioning sensor data to obtain motion state parameters of the target vehicle; performing confidence quantification evaluation according to the environment perception data, the multi-source positioning sensor data and the motion state parameters to obtain scene state parameters; performing stage-by-stage modulation on original covariance of each sensor according to the scene state parameters, the multi-source positioning sensor data and the map priori data to obtain modulation covariance of each sensor; and obtaining a fusion positioning result of the target vehicle by using a multi-sensor fusion filtering method according to the modulation covariance of each sensor, the multi-source positioning sensor data and the motion state parameters. The application improves the robustness of vehicle positioning in a complex scene, the map credibility and the full-scene adaptability.

Description

Technical Field

[0001] This application relates to the field of intelligent transportation technology, and in particular to a vehicle fusion positioning method, device, equipment, medium and product. Background Technology

[0002] High-reliability vehicle positioning typically relies on visual inertial odometry, laser inertial odometry, Global Navigation Satellite System (GNSS) / Real-Time Kinematic (RTK) positioning, high-precision maps, and filtering fusion algorithms.

[0003] Visual-Inertial Navigation System-Fusion (VINS-Fusion) methods are highly sensitive to lighting, texture, and dynamic obstacles. In rain, snow, fog, nighttime conditions, low-texture roads, or dynamic, dense traffic scenarios, they are prone to problems such as feature tracking failure, increased reprojection errors, and cumulative drift. Fast Direct LiDAR-Inertial Odometry (FAST-LIO2) technologies are susceptible to geometric degradation in scenarios involving long straight roads, tunnels, repetitive structures, or insufficient point cloud geometry. Furthermore, most GNSS, Inertial Measurement Unit (IMU), and LiDAR fusion positioning schemes employ pre-defined noise models or fixed-weight models, which are insufficient for representing the degradation levels of different sensors in various scenarios. Summary of the Invention

[0004] The purpose of this application is to provide a vehicle fusion positioning method, device, equipment, medium and product that can improve the robustness of vehicle positioning, map reliability and adaptability to all scenarios in complex scenarios.

[0005] To achieve the above objectives, this application provides the following solution.

[0006] In a first aspect, this application provides a vehicle fusion localization method, comprising: acquiring environmental perception data, map prior data, and multi-source localization sensing data of a target vehicle; the multi-source localization sensing data being collected by multiple sensors of different types; calculating the map prior data and the multi-source localization sensing data respectively to obtain motion state parameters of the target vehicle; the motion state parameters including visual pose, lidar pose, and global correction constraints; performing confidence quantification evaluation based on the environmental perception data, the multi-source localization sensing data, and the motion state parameters to obtain scene state parameters; the scene state parameters including observation confidence index, weather confidence index, and obstacle confidence index; modulating the original covariance of each sensor in stages based on the scene state parameters, the multi-source localization sensing data, and the map prior data to obtain the modulation covariance of each sensor; and using a multi-sensor fusion filtering method based on the modulation covariance of each sensor, the multi-source localization sensing data, and the motion state parameters to obtain the fusion localization result of the target vehicle.

[0007] Secondly, this application provides a vehicle fusion positioning device, comprising: a data acquisition module, used to acquire environmental perception data, map prior data, and multi-source positioning sensing data of a target vehicle; the multi-source positioning sensing data is collected by multiple sensors of different types; a motion state calculation module, used to calculate the map prior data and the multi-source positioning sensing data respectively to obtain motion state parameters of the target vehicle; the motion state parameters include visual pose, lidar pose, and global correction constraints; and a scene state evaluation module, used to evaluate the scene state based on the environmental perception data, the multi-source positioning sensing data, and the multi-source positioning sensing data. The motion state parameters are evaluated using confidence metrics to obtain scene state parameters. These scene state parameters include observation confidence indices, weather confidence indices, and obstacle confidence indices. A covariance modulation module is used to modulate the original covariance of each sensor in stages based on the scene state parameters, the multi-source positioning sensor data, and the map prior data to obtain the modulation covariance of each sensor. A vehicle fusion positioning module is used to obtain the fusion positioning result of the target vehicle by using a multi-sensor fusion filtering method based on the modulation covariance of each sensor, the multi-source positioning sensor data, and the motion state parameters.

[0008] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described vehicle fusion positioning method.

[0009] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described vehicle fusion positioning method.

[0010] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the above-described vehicle fusion positioning method.

[0011] Based on the specific embodiments provided in this application, the following technical effects are disclosed.

[0012] This application provides a vehicle fusion positioning method, device, equipment, medium, and product. By acquiring environmental perception data, map prior data, and multi-source positioning sensor data of the target vehicle, the input data covers three core types of information: vehicle motion perception, surrounding environment feature recognition, and global map benchmark constraints. This ensures sufficient information dimensions and constraints for the subsequent fusion positioning process. The map prior data and multi-source positioning sensor data are processed separately. High-frequency relative motion information is output through local pose calculations using visual and lidar methods. Simultaneously, global correction constraints generated from the map prior data suppress the cumulative drift error of single-sensor calculations, providing a layered, multi-dimensional motion state benchmark for the subsequent fusion process. Confidence quantification assessments are conducted to obtain scene state parameters, enabling real-time quantification of the reliability of various sensor observations in the current driving scenario and accurate acquisition of... This method captures the impact of scene factors such as weather changes and dense obstacle distribution on sensor observation accuracy, providing a scenario-based quantitative basis for subsequent dynamic adjustment of covariance. Based on scene state parameters, multi-source positioning sensor data, and prior map data, the original covariance of each sensor is modulated in stages. This dynamically corrects the observation reliability of each sensor from two dimensions: scene adaptation and map constraints, effectively avoiding the inherent defect of fixed covariance being unable to adapt to complex and changing driving scenarios. The multi-sensor fusion filtering method is used to obtain the fused positioning result of the target vehicle, fully leveraging the complementary redundancy advantages of multiple sensors. By rationally allocating the fusion weights of each sensor observation based on the dynamically adjusted covariance, stable and high-precision positioning output is maintained even in complex driving scenarios such as severe weather and dense obstacles, effectively improving the overall robustness and positioning accuracy of vehicle positioning. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is an application environment diagram of a vehicle fusion positioning method according to an embodiment of this application.

[0015] Figure 2This is a flowchart illustrating a vehicle fusion positioning method provided in an embodiment of this application.

[0016] Figure 3 This is a schematic diagram of a three-layer positioning system architecture provided in an embodiment of this application.

[0017] Figure 4 This is a schematic diagram of the scene adaptive redundancy mechanism judgment process provided in an embodiment of this application.

[0018] Figure 5 This is a schematic diagram of the functional modules of a vehicle fusion positioning device provided in an embodiment of this application.

[0019] Figure 6 This is a schematic diagram of the overall structure of a vehicle fusion positioning device provided in another embodiment of this application.

[0020] Figure 7 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0021] Taking autonomous vehicle positioning technology as an example, VINS-Fusion has disclosed an optimized multi-sensor state estimation framework that can be used for self-localization in scenarios such as unmanned vehicles, drones, and augmented reality (AR) / virtual reality (VR); FAST-LIO2 has disclosed a fast and robust laser inertial odometry method that achieves high-frequency laser inertial mapping and positioning through direct registration of the original point cloud and incremental k-dimensional tree (ikd-Tree) map structure; in addition, there are also publicly available solutions such as GNSS / IMU / LiDAR fusion positioning, visual and high-precision map matching positioning, and multi-sensor particle filter positioning.

[0022] Existing VINS-Fusion-type methods can achieve high-frequency local pose estimation using cameras and IMUs, but these methods usually focus more on visual inertial state estimation itself and lack scene recognition, sensor failure judgment and map reverse constraint mechanisms for the full-scene operation of autonomous logistics vehicles.

[0023] Existing FAST-LIO2-type laser inertial odometry technology can complete point cloud registration and local mapping at high frequencies, exhibiting good real-time performance and accuracy. However, in scenarios such as long straight roads, tunnels, underground parking garages, open areas, repetitive structural environments, or insufficient point cloud geometric features, laser registration is prone to geometric degradation. Although FAST-LIO2 can improve robustness through tight coupling between the laser and the IMU, its focus is on the laser inertial odometry itself, and it has not systematically addressed the issue of coordinated weight adjustment among vision, GNSS, high-definition (HD) maps, dynamic event layers, and multi-level redundancy.

[0024] Existing GNSS / IMU / LiDAR fusion positioning schemes typically employ Kalman filtering, extended Kalman filtering, particle filtering, or uniformity filtering to fuse GNSS, IMU, laser, or wheel speedometer information to improve vehicle positioning accuracy. However, most of these methods use pre-defined noise models or fixed-weight models, which are insufficient to represent the degradation levels of different sensors in different scenarios. For example, in urban canyons, under overpasses, tunnel entrances, underground parking garages, and areas with tree cover, GNSS suffers from multipath propagation, occlusion, or signal loss; if fixed weights are still used for fusion, the fused state can easily be skewed by abnormal GNSS observations.

[0025] Existing high-precision map-based positioning methods typically use the map as a static prior, matching it with elements such as lane lines, curbs, traffic signs, and road boundaries in the HD map through camera, laser, or GNSS observations to achieve lane-level or centimeter-level positioning. Publicly available high-precision map visual positioning schemes have proposed fusing visual features, bird's-eye view (BEV) conversion, image segmentation results, and GNSS positioning results to mitigate drift problems in weak GNSS areas. However, these schemes usually assume the map prior is reliable and lack credibility assessments for factors such as map timeliness, construction changes, temporary road closures, lane line wear, and traffic sign changes. If the map is inconsistent with the actual road environment, erroneous map priors may conversely cause positioning errors.

[0026] Existing multi-sensor fusion patents also include schemes for vehicle localization using Global Positioning System (GPS), IMU, camera, odometer, and vector map. For example, some patents disclose generating particle swarms on vector maps based on GPS positioning deviations and calculating particle weights by combining camera lane detection results, thereby achieving lane-level localization in autonomous driving scenarios. While such schemes can reduce costs to some extent, they primarily rely on lane-level geometric constraints and particle filtering, and have not yet developed multi-source information entropy modulation, scene state machine switching, multi-layer positioning architecture, and a closed-loop update mechanism for certified maps applicable to complex, full-scene scenarios.

[0027] In summary, the existing technology has at least the following shortcomings.

[0028] First, existing technologies mostly use fixed or semi-fixed fusion weights, which cannot switch fusion strategies in real time according to scenarios such as urban canyons, tunnels, underground parking garages, rain, snow, fog, open highways, and complex intersections.

[0029] Second, existing technologies do not adequately express the degradation state of sensors. They typically adjust noise based on a single metric, such as GNSS accuracy factor, number of visual features, or point cloud matching residuals, and lack a unified method for modeling information entropy uncertainty.

[0030] Third, existing HD maps are mostly used as passive matching objects, lacking map credibility calculation and map reverse constraint mechanisms, and cannot actively reduce map constraint weights when maps expire, roads are under construction, temporary controls are implemented, or traffic signs change.

[0031] Fourth, existing technologies typically fuse visual, laser, GNSS, and IMU observations as observations at the same level, lacking a hierarchical collaborative mechanism between high-frequency local positioning, mid-frequency structural positioning, and low-frequency global correction.

[0032] Fifth, existing technologies lack multi-level redundancy under extreme conditions, such as when GNSS failure, visual failure, and laser degradation occur simultaneously, the system lacks a clear fallback positioning path.

[0033] Sixth, existing technologies do not adequately consider the security, traceability, and authentication update mechanisms of map data, making it difficult to meet the requirements of reliable map versions, compliant data, and auditable updates for the long-term operation of autonomous logistics vehicles.

[0034] In view of the aforementioned deficiencies in the prior art, the core technical problem to be solved in this application is as follows.

[0035] How can autonomous logistics vehicles achieve reliable fusion of multi-source sensor information, continuous positioning output, and reliable generation and dynamic updating of certified maps when facing complex scenarios such as urban canyons, tunnels, underground parking garages, severe weather, dynamic traffic, and map changes?

[0036] Specifically, this application aims to solve the following technical problems.

[0037] 1. Address the issue of existing fusion positioning methods lacking scene awareness capabilities.

[0038] Existing multi-sensor fusion methods typically directly filter and fuse data from vision, laser, GNSS, and IMU, lacking consideration of the current road environment and sensor applicability. One embodiment of this application constructs a scene reliability state machine to identify GNSS availability, visual quality, laser structure stability, map matching reliability, weather conditions, and dynamic obstacle density in real time, thereby automatically switching fusion strategies according to different scenarios.

[0039] 2. To address the lack of a unified mathematical basis for existing sensor weight adjustment.

[0040] In existing technologies, covariance adjustment typically relies on empirical thresholds or single health indicators, making it difficult to uniformly express the uncertainties of visual, laser, GNSS, and map measurements. One embodiment of this application introduces an information entropy covariance modulation mechanism, which transforms visual feature quality, laser geometric degradation, GNSS solution status, and map matching residuals into a unified information entropy index. Based on this index, the covariance matrix is ​​dynamically expanded or contracted to achieve a theoretically grounded adaptive weight allocation.

[0041] 3. Solve the problem that existing HD maps passively participate in positioning and cannot determine their own reliability.

[0042] Existing HD map positioning systems often assume the map prior is correct, lacking assessment of map timeliness and consistency with the real environment. One embodiment of this application proposes a map credibility reverse constraint mechanism. This mechanism calculates map credibility by using the matching residual between sensor observations and map priors. Map constraints are strengthened when the map matches the real environment, and automatically reduced when the map does not match real-time road conditions, thus preventing erroneous maps from misleading positioning results.

[0043] 4. To address the problems of unclear hierarchical structure and difficulty in balancing continuity and global consistency in existing positioning systems.

[0044] Existing fusion positioning systems often directly input multi-sensor observations into the same filter, making it difficult to simultaneously satisfy high-frequency continuous output, mid-frequency structural constraints, and low-frequency global correction. One embodiment of this application proposes a multi-layer positioning architecture that combines an IMU, wheel speed, and a Visual-Inertial Navigation System (VINS) into a local high-frequency positioning layer; FAST-LIO2 and point cloud matching into a mid-frequency structural constraint layer; and RTK, HD maps, Standard Definition (SD) maps, geomagnetic maps, and dynamic event layers into a low-frequency global correction layer, thereby achieving a balance between real-time performance, robustness, and global consistency.

[0045] 5. Solve the problems of positioning failure and drift accumulation in complex degradation scenarios.

[0046] To address issues such as GNSS occlusion, weak visual texture, laser degradation, and map inconsistency, one embodiment of this application utilizes scene state parameters, information entropy covariance modulation, map credibility inverse constraints, and multi-level redundancy mechanisms to achieve automatic switching and fallback positioning under different sensor failure conditions, ensuring that logistics vehicles can continuously output usable positioning results in tunnels, underground garages, severe weather, and complex urban roads.

[0047] 6. Address the issues of lack of authentication, traceability, and dynamic updates in the map generation process.

[0048] Existing map update schemes often focus on map construction accuracy, but they do not adequately consider data sources, version authentication, update review, and dynamic event overlay. One embodiment of this application uses an authenticated map generation mechanism to anonymize, encrypt, version-hash, and manage incremental updates of prior map data, enabling the map to be traceable, auditable, verifiable, and iteratively updated.

[0049] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0050] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0051] The vehicle fusion positioning method provided in this application embodiment can be applied to, for example... Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be set up independently, integrated into server 104, or placed in the cloud or on another server. Terminal 102 can send environmental perception data, map prior data, and multi-source positioning sensor data of the target vehicle to server 104. After receiving the environmental perception data, map prior data, and multi-source positioning sensor data of the target vehicle, server 104 performs data processing and confidence quantification evaluation, and modulates the original covariance of each sensor in stages. Finally, it uses a multi-sensor fusion filtering method to obtain the fused positioning result of the target vehicle. Server 104 can feed back the obtained fused positioning result of the target vehicle to terminal 102. Furthermore, in some embodiments, the vehicle fusion localization method can also be implemented independently by the server 104 or the terminal 102. For example, the terminal 102 can directly perform data processing and confidence quantification evaluation on the environmental perception data, map prior data, and multi-source localization sensor data of the target vehicle, and modulate the original covariance of each sensor in stages. Finally, a multi-sensor fusion filtering method is used to obtain the fusion localization result of the target vehicle. Alternatively, the server 104 can obtain the environmental perception data, map prior data, and multi-source localization sensor data of the target vehicle from the data storage system, perform data processing and confidence quantification evaluation, and modulate the original covariance of each sensor in stages. Finally, a multi-sensor fusion filtering method is used to obtain the fusion localization result of the target vehicle.

[0052] The terminal 102 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 104 can be implemented using a standalone server or a server cluster composed of multiple servers, or it can be a cloud server.

[0053] In one exemplary embodiment, such as Figure 2 As shown, a vehicle fusion positioning method is provided. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps 201 to 205.

[0054] Step 201: Acquire environmental perception data, map prior data, and multi-source positioning sensor data of the target vehicle. The multi-source positioning sensor data is collected by multiple sensors of different types.

[0055] Step 202: Solve the map prior data and the multi-source positioning sensor data respectively to obtain the motion state parameters of the target vehicle. The motion state parameters include visual pose, lidar pose, and global correction constraints.

[0056] Step 203: Based on the environmental perception data, the multi-source positioning sensor data, and the motion state parameters, perform a confidence quantification assessment to obtain scene state parameters. The scene state parameters include observation confidence index, weather confidence index, and obstacle confidence index.

[0057] Step 204: Based on the scene state parameters, the multi-source positioning sensor data, and the map prior data, the original covariance of each sensor is modulated in stages to obtain the modulation covariance of each sensor.

[0058] Step 205: Based on the modulation covariance of each sensor, the multi-source positioning sensor data, and the motion state parameters, a multi-sensor fusion filtering method is used to obtain the fusion positioning result of the target vehicle.

[0059] In one example, the multi-source positioning sensor data includes inertial observation data, odometer observation data, visual observation data, lidar observation data, and satellite observation data. The visual pose includes local visual position coordinates, local visual attitude angles, and velocity components. The lidar pose includes local lidar position coordinates and local lidar attitude angles. The global correction constraints include global coordinate constraints, road topology constraints, and lane geometry constraints.

[0060] Step 202 above can be replaced by steps 2021 to 2026.

[0061] Step 2021: The inertial observation data, the odometer observation data, and the visual observation data are fused and calculated to obtain the visual pose.

[0062] Step 2022: Perform point cloud front-end processing on the lidar observation data to obtain an initial pose estimate.

[0063] Step 2023: The initial pose estimation is matched with point cloud data using a point cloud registration method to obtain the matched pose.

[0064] Step 2024: Perform local map matching correction on the matched pose to obtain the lidar pose.

[0065] Step 2025: Construct the global coordinate constraints based on the satellite observation data.

[0066] Step 2026: Construct the road topology constraints and lane geometry constraints based on the map prior data.

[0067] In one example, as follows Figure 3 The three-layer positioning system shown acquires the motion state parameters of the target vehicle to improve the positioning continuity and robustness in different scenarios.

[0068] The first layer is a local high-frequency positioning layer, which consists of an IMU, wheel speed sensor, and binocular VINS-Fusion. It is mainly used for high-frequency motion prediction in a short period of time and outputs the following 100Hz level vehicle visual pose estimation.

[0069] .

[0070] in, The visual pose of the target vehicle. Let X be the position coordinates of the target vehicle in the X-axis direction of the local visual coordinate system. Let Y be the position coordinates of the target vehicle in the Y-axis direction of the local visual coordinate system. Let Z be the position coordinates of the target vehicle in the Z-axis direction of the local visual coordinate system. Let the roll angle of the target vehicle be the local visual coordinate system. Let be the pitch angle of the target vehicle in the local visual coordinate system. Let be the heading angle of the target vehicle in the local visual coordinate system. Let X be the velocity component of the target vehicle in the X-axis direction. Let Y be the velocity component of the target vehicle in the Y-axis direction. The velocity component of the target vehicle in the Z-axis direction.

[0071] In this example, the visual pose of the target vehicle is obtained by fusing observation data from an IMU, wheel velocity meter, and binocular VINS-Fusion after time synchronization. The IMU provides angular velocity and linear acceleration; the wheel velocity meter provides longitudinal velocity constraints; and the binocular VINS-Fusion provides visual-inertial pose estimation. A unified local state estimation result is formed through pre-integration calculation, kinematic constraints, and sliding window optimization, thus yielding the visual pose of the target vehicle. The first layer focuses on ensuring short-term continuity, maintaining smooth vehicle trajectory even during GNSS interruptions or laser obstruction.

[0072] The second layer is the mid-frequency structured constraint localization layer. It calculates the laser pose using the FAST-LIO2 laser odometry method, point cloud registration method, and local map matching method, outputting a structured localization result (LiDAR pose) from 10Hz to 20Hz. The LiDAR pose is obtained through voxel grid filtering, outlier removal, and registration using Normal Distribution Transform (NDT) or Iterative Closest Point (ICP).

[0073] .

[0074] in, The lidar pose of the target vehicle. Let X be the position coordinates of the target vehicle in the X-axis direction of the local coordinate system of the lidar. Let Y be the position coordinates of the target vehicle in the local coordinate system of the lidar along the Y-axis. Let Z be the position coordinates of the target vehicle in the Z-axis direction of the local coordinate system of the lidar. The roll angle of the target vehicle in the local coordinate system of the lidar. Let be the pitch angle of the target vehicle in the local coordinate system of the lidar. The heading angle of the target vehicle in the local coordinate system of the lidar.

[0075] The LiDAR pose of the target vehicle is calculated from the original LiDAR point cloud data after VoxelGrid downsampling, outlier removal, feature extraction, NDT or ICP point cloud registration, and local map matching. FAST-LIO2 first outputs the initial pose estimate, then uses NDT or ICP to obtain the matched pose, and finally combines local map constraints to obtain the LiDAR pose. In this example, the degree of geometric degradation of the point cloud is also determined based on the eigenvalues ​​of the Hessian matrix.

[0076] The third layer is the global low-frequency correction positioning layer, composed of RTK GPS, HD map, SD map, geomagnetic map, and dynamic event layer. It provides global coordinate constraints, road topology constraints, lane geometry constraints, and performs long-term drift correction. Specifically, RTK GPS primarily provides global coordinate constraints; the SD map primarily provides road topology constraints, including road connectivity, road network structure, and administrative region information; the HD map primarily provides lane-level geometric constraints and performs long-term drift correction; the geomagnetic map is used for positional correction in GNSS failure scenarios; and the dynamic event layer is used to correct for temporary changes in areas such as construction, road closures, and traffic control.

[0077] In one example, such as Figure 3 As shown, the above three-layer positioning results are fused as unified input, and the positioning reliability of each layer is expressed by the covariance matrix, thus achieving the unification of high-frequency continuity, mid-frequency structural stability and low-frequency global consistency.

[0078] In one example, the multi-source positioning sensing data includes visual observation data and lidar observation data. Step 203 above can be replaced by steps 2031 to 2033.

[0079] Step 2031: Perform scene confidence assessment on the motion state parameters to obtain the observation confidence index.

[0080] Step 2032: Quantify and evaluate the confidence level of the environmental perception data to obtain the weather confidence index.

[0081] Step 2033: Based on the lidar observation data and the visual observation data, obstacle detection is performed to obtain the obstacle distribution density, and the obstacle distribution density is quantitatively evaluated to obtain the obstacle confidence index.

[0082] In one example, to avoid using fixed weights in traditional multi-sensor fusion methods under different environments, a scene confidence state machine is introduced to identify the vehicle's environment in real time and dynamically adjust sensor weights and fusion strategies accordingly. This example constructs the following scene state parameter vector.

[0083] .

[0084] in, This is a vector of scene state parameters. As a GNSS availability indicator, Confidence level for visual texture and lighting quality. This is a confidence index for the structural stability of laser point clouds. Match confidence metrics to the map. This is a weather confidence index, indicating the impact of environmental factors such as rain, snow, fog, and low light. It is an obstacle confidence index, indicating the density of dynamic obstacles. , , and Together, they constitute the observation confidence index.

[0085] The aforementioned scene state parameters are calculated from multi-source positioning sensor data, environmental perception data, and motion state parameters. Among them, It is calculated from RTK status, number of satellites, position dilution of precision (PDOP) and horizontal dilution of precision (HDOP); It is calculated from the number of effective feature points, image brightness, reprojection error, and optical flow consistency. It is calculated from point cloud density, matching residual, Hessian matrix eigenvalues, and the number of degradation directions; Calculated from map matching residuals and matching success rate; It is calculated from camera image quality, rain and fog detection results, and environmental sensor data; It is calculated from the number of dynamic targets, the proportion of occupancy, and the target motion state.

[0086] In one example, the current scene type is determined based on the aforementioned motion state parameters. Scene types include high-speed open roads, urban canyons, complex intersections, tunnels, underground parking garages, severe weather, and construction areas. The method for determining the scene type is as follows.

[0087] .

[0088] in, for Corresponding scene types, Preset scene type judgment rules.

[0089] Different fusion strategies are implemented for different scenario types. For example, in open highways, the weight of RTK GPS and laser is increased; in urban canyons, the weight of GNSS is decreased, while the weight of visual semantics and HD map matching is increased; in tunnels or underground parking garages, the weight of GNSS is turned off, while laser, IMU, wheel speed, and geomagnetic fusion are enabled; in rain, snow, and fog environments, the weight of visual data is decreased, while the weight of laser structural features and road topology constraints is increased. By calculating scenario state parameters, this application does not simply overlay multi-sensor data, but achieves adaptive fusion positioning based on environmental understanding.

[0090] In one example, step 204 above can be replaced by steps 2041 to 2047.

[0091] Step 2041: Calculate the health indicators of each sensor based on the scene state parameters.

[0092] Step 2042: For any sensor, construct the information entropy of the sensor based on the sensor's health indicators.

[0093] Step 2043: Based on the information entropy, the original covariance of the sensor is initially modulated using the following formula to obtain the initial modulation covariance.

[0094] .

[0095] .

[0096] in, Let be the covariance inflation coefficient of the i-th sensor. Let be the information entropy of the i-th sensor. Let be the information entropy sensitivity coefficient of the i-th sensor. Let be the initial modulation covariance of the i-th sensor. Let be the original covariance of the i-th sensor.

[0097] When a sensor is in a degenerate state, its information entropy increases, and the corresponding covariance automatically expands, thereby reducing the weight of that sensor in the fusion localization. The sensor fusion weights are defined as follows, thus achieving an adaptive allocation of uncertainty based on information entropy.

[0098] .

[0099] in, The fusion weights are for the i-th sensor. Let be the initial modulation covariance of the j-th sensor. Operators for solving the trace of a matrix.

[0100] Step 2044: Calculate the map matching residual based on the multi-source positioning sensor data and the map prior data.

[0101] Step 2045: Calculate the map confidence index based on the map matching residuals.

[0102] Step 2046: Based on the map confidence index, the initial modulation covariance is modulated a second time using the following formula to obtain the second modulation covariance.

[0103] .

[0104] in, Let be the quadratic modulation covariance of the i-th sensor. The confidence index for the map is... This is the constant of the stability term.

[0105] Step 2047: Obtain the modulation covariance based on the second-order modulation covariance.

[0106] In this example, an information entropy-driven covariance modulation method is further proposed to quantify the uncertainty of the current output of each sensor.

[0107] For the There are 1 sensor, and its original covariance is . .

[0108] Based on the health indicators of each sensor, the information entropy of each sensor is constructed using the following formula.

[0109] .

[0110] in, For the sensor The reliable probability distribution obtained by normalizing all health states For sensors Information entropy.

[0111] In one example, the information entropy of the visual sensor, laser sensor, GNSS sensor, and map matching information source is calculated separately.

[0112] Specifically, as shown in the following formula, the information entropy of the visual sensor is determined by the number of effective feature points, reprojection error, optical flow consistency, and the proportion of semantic dynamic regions.

[0113] .

[0114] in, The information entropy of the visual sensor. This is the function for calculating the information entropy of a visual sensor. The number of feature points in the visual sensor. This refers to the reprojection error of the visual sensor. For optical flow consistency of visual sensors, This represents the semantic dynamic region ratio of the visual sensor.

[0115] As shown in the following formula, the information entropy of the laser sensor is determined by the point cloud density, matching residual, Hessian eigenvalue, and number of degradation directions.

[0116] .

[0117] in, The information entropy of the laser sensor, This is the function for calculating the information entropy of the laser sensor. The point cloud density of the laser sensor. For the matching residual of the laser sensor, This represents the Hessian eigenvalue of the laser sensor. This represents the number of degradation directions for the laser sensor.

[0118] As shown in the following formula, the entropy of GNSS information is determined by RTK solution status, PDOP, HDOP, number of satellites, and multipath index.

[0119] .

[0120] in, For GNSS information entropy, This is the entropy calculation function for GNSS information. For GNSS RTK solution status, For GNSS observations, PDOP For GNSS observations, HDOP The number of available observation satellites for GNSS. This is a multipath indicator for GNSS.

[0121] Traditional HD maps are usually used as passive matching objects. This application introduces map prior data into the fusion closed loop, so that the map not only participates in positioning correction, but also influences the sensor weight allocation in reverse.

[0122] The laser point cloud and visual semantic features are matched with lane lines, curbs, traffic signs, and passable areas in the HD map to obtain the map matching residual as shown in the following formula.

[0123] .

[0124] in, To match residuals to the map, For multi-source positioning sensor data, This is the prior data for the map.

[0125] The map confidence index is calculated using the following formula based on the map matching residuals.

[0126] .

[0127] in, This is a map confidence index. Let be the prior covariance of the map.

[0128] When the map is highly consistent with real-time perception Approaching 1 strengthens map constraints; when the map is inconsistent with the real-time environment, such as during construction, lane changes, temporary road closures, or missing signs, The system reduces and automatically weakens map constraints to prevent positioning offsets caused by erroneous maps. This example forms a two-way closed-loop mechanism of "sensor correcting the map and map inversely constraining the sensor," improving positioning reliability in scenarios with complex roads and changing maps.

[0129] In one example, after step 205 above, the vehicle fusion positioning method further includes steps 206 to 208.

[0130] Step 206: Construct a local real-time map based on the fusion positioning result of the target vehicle, the multi-source positioning sensor data, and the motion state parameters.

[0131] Step 207: Calculate map incremental data based on the local real-time map and the map prior data.

[0132] Step 208: Update the map prior data based on the map incremental data.

[0133] This example employs a multi-layered map structure, including an SD base map, an HD lane-level map, a geomagnetic map, and a dynamic event layer. The SD map provides global road network and administrative region information for coarse positioning; the HD map includes lane lines, curbs, traffic signs, traffic lights, stop lines, and passable areas for centimeter-level fine matching; the geomagnetic map is used in scenarios where GNSS is unavailable, such as tunnels and underground parking garages; and the dynamic event layer represents construction, congestion, accidents, temporary road closures, and traffic control information.

[0134] During the map prior data generation process, laser point clouds, visual semantic features, RTK trajectories, and vehicle motion states are integrated to form structured map elements. After the map prior data is uploaded to the cloud, it undergoes anonymization, encryption, review, and version authentication.

[0135] Each time the map prior data is updated, the following unique authentication digest is generated.

[0136] .

[0137] .

[0138] in, The authentication digest generated for the a-th report of incremental map data. For incremental map data, For the timestamp of incremental map data upload, for The corresponding unique identifier, The authentication digest generated for the b-th update of the map prior data. To update the map prior data, To update the timestamp of the map prior data, for The corresponding unique identifier, This is a function for the Secure Hash Algorithm 256 (SHA-256).

[0139] This step records each update through the map's prior data version chain, enabling traceability of map origin, verification of version, and auditability of updates.

[0140] When a vehicle detects an area of ​​map inconsistency during operation, incremental map data is generated as shown in the following formula.

[0141] .

[0142] in, This is incremental map data, including newly added lane lines, missing lane lines, changes in traffic signs, changes in road boundaries, changes in construction areas, and temporary road closure information. This is a partial real-time map. This is the prior data for the map.

[0143] In this example, incremental map data is uploaded to the cloud for review, and the prior map data is updated after the review is approved. Before review, incremental map data only temporarily participates in positioning as a dynamic event layer and does not change the prior map data.

[0144] In one example, such as Figure 4 As shown, a scene adaptive redundancy mechanism is set up, which automatically enters the following different redundancy modes based on the results of the scene state parameters.

[0145] When GNSS fails, reduce the GNSS weight and enable the fusion of laser, visual, IMU, and HD maps.

[0146] When visual degradation occurs, enhance laser structural features, IMU, and road boundary constraints (map constraints).

[0147] When laser degradation occurs, enhance visual semantics, IMU, wheel speed, and map topology constraints.

[0148] When both GNSS and visual maps fail, laser, IMU, wheel speed, and geomagnetic map fusion is enabled.

[0149] In special scenarios such as underground parking garages and tunnels, GNSS is turned off, and the system is switched to a joint positioning mode using geomagnetic map, wheel speed, IMU, and local semantic map.

[0150] In adverse weather conditions, visual perception is weakened while topological constraints are enhanced.

[0151] Finally, based on the above-mentioned scenario adaptive redundancy mechanism, a scenario fusion strategy is output.

[0152] In addition, anomaly removal can be performed based on residual consistency detection. Specifically, the observation residuals are calculated using the following formula.

[0153] .

[0154] in, Let k be the residual vector observed at time k. Let k be the target vehicle state vector output by filtering at time k.

[0155] If the following formula is met, the sensor is determined to be in abnormal measurement mode, and weight reduction, temporary isolation or redundant positioning mode is triggered.

[0156] .

[0157] Among them, S k Let covariance be the matrix corresponding to the residual. A preset judgment threshold is set for the chi-square test.

[0158] In one example, the multi-sensor fusion filtering method is the Extended Kalman Filter (EKF) method.

[0159] The EKF method receives visual, lidar, GNSS, and map measurements after information entropy modulation and inverse constraints based on map confidence indices, and constructs a unified state estimation model. The state prediction formula of the state estimation model is as follows.

[0160] .

[0161] .

[0162] in, This is the posterior vehicle state vector obtained after filtering and updating at time k-1. The prior vehicle state prediction vector at time k is obtained by extrapolating the result at time k-1. This is the system control input at time k-1. The noise at time k-1 is the process noise. It is a nonlinear state transition function. Let be the posterior covariance matrix at time k-1. Let k be the prior covariance matrix. for exist The Jacobian matrix obtained by solving for the given value. Let be the covariance matrix corresponding to the process noise at time k-1.

[0163] The measurement update formula is as follows.

[0164] .

[0165] .

[0166] .

[0167] in, Here is the Kalman gain matrix. For the measurement model Jacobian matrix, To measure the noise covariance matrix, Let k be the updated posterior vehicle state vector measured at time k. Let k be the multi-source localization sensor data vector at time k. Let k be the posterior covariance matrix updated at time k. It is an identity matrix.

[0168] In this example, ,in, For the second-order modulation covariance of the vision sensor, The second-order modulation covariance of the laser sensor, The second-order modulation covariance of the GNSS sensor. The second-order modulation covariance is used to match the information sources to the map. Through this method, the fusion weights can be dynamically adjusted based on the real-time scene, sensor health, and map confidence.

[0169] In one example, to verify the effectiveness of the proposed method, multi-scenario simulation experiments were conducted based on the ROS and Gazebo / Carla autonomous driving simulator co-simulation platform, and compared with VINS-Fusion, FAST-LIO2, and the traditional fixed covariance EKF fusion method. In the urban canyon scenario, the average positioning error of the traditional VINS+RTK fusion method was approximately 48 cm, FAST-LIO2 was approximately 36 cm, and the fixed covariance EKF fusion was approximately 31 cm. The proposed vehicle fusion positioning method, through dynamic scene recognition and map inverse constraints, reduced the average positioning error to 18 cm. In the long tunnel scenario, due to the complete failure of GNSS and the impact of low light on vision, the traditional VINS-Fusion accumulated drift reached 2.8 m, and the FAST-LIO2 drift was approximately 1.6 m. The proposed vehicle fusion positioning method, through laser, IMU, and geomagnetic redundancy fusion, controlled the maximum drift to within 0.45 m.

[0170] In underground parking garages with automated parking systems, traditional methods often result in parking space errors exceeding 20cm due to repetitive structures and the absence of GNSS. The vehicle fusion localization method proposed in this application, through the fusion of local topology maps, geomagnetism, and wheel speed, reduces this error to less than 9cm. In tests conducted in rain, snow, fog, and low-light environments, the localization availability of traditional visual-inertial methods is approximately 81.5%, FAST-LIO2 is approximately 92.3%, and fixed EKF fusion is approximately 95.1%. The vehicle fusion localization method proposed in this application, through information entropy covariance modulation and scene adaptive fusion mechanisms, improves the system's localization availability to over 99%.

[0171] In summary, this application significantly improves the vehicle's positioning continuity, robustness, fault tolerance, and map reliability in complex scenarios through the collaborative design of scene confidence state parameters, information entropy covariance modulation, map confidence inverse constraints, and multi-layer positioning architecture. It is particularly suitable for full-scenario autonomous driving applications such as complex urban roads, tunnels, underground parking garages, severe weather, and dynamic construction areas.

[0172] Based on the same inventive concept, this application also provides a vehicle fusion positioning device for implementing the vehicle fusion positioning method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more vehicle fusion positioning device embodiments provided below can be found in the limitations of the vehicle fusion positioning method described above, and will not be repeated here.

[0173] In one exemplary embodiment, such as Figure 5 As shown, a vehicle fusion positioning device is provided, including a data acquisition module 001, a motion state calculation module 002, a scene state evaluation module 003, a covariance modulation module 004, and a vehicle fusion positioning module 005.

[0174] The data acquisition module 001 is used to acquire environmental perception data, map prior data, and multi-source positioning sensor data of the target vehicle. The multi-source positioning sensor data is collected by multiple sensors of different types.

[0175] The motion state calculation module 002 is used to calculate the motion state parameters of the target vehicle by processing the map prior data and the multi-source positioning sensor data respectively. The motion state parameters include visual pose, lidar pose, and global correction constraints.

[0176] The scene state assessment module 003 is used to perform confidence quantification assessment based on the environmental perception data, the multi-source positioning sensor data, and the motion state parameters to obtain scene state parameters. The scene state parameters include observation confidence index, weather confidence index, and obstacle confidence index.

[0177] The covariance modulation module 004 is used to modulate the original covariance of each sensor in stages according to the scene state parameters, the multi-source positioning sensing data and the map prior data, so as to obtain the modulation covariance of each sensor.

[0178] The vehicle fusion positioning module 005 is used to obtain the fusion positioning result of the target vehicle by using a multi-sensor fusion filtering method based on the modulation covariance of each sensor, the multi-source positioning sensing data and the motion state parameters.

[0179] In another example, such as Figure 6 As shown, the vehicle fusion positioning device consists of a front-end multi-source perception layer, an adaptive scene state machine (responsible for scene recognition, environmental analysis, and sensor health assessment), an information entropy covariance modulation middleware (used for dynamic covariance adjustment and fusion weight allocation), a multi-layer fusion positioning module (used to complete layered filtering operations according to a three-layer architecture of high-frequency layer, mid-frequency layer, and global layer), and an authentication map matching and reverse constraint module (relying on HD map, dynamic event layer, and geomagnetic map to perform map matching and constraint correction), ultimately outputting a high-precision positioning result for the target vehicle. Specifically, the front-end multi-source perception layer accesses four types of sensor data sources—VINS-Fusion, FAST-LIO2, RTK, and IMU—for layer-by-layer processing, ultimately outputting a high-precision vehicle positioning result. The device uses the Robot Operating System (ROS) middleware to complete multi-sensor data synchronization, state estimation, covariance propagation, anomaly detection, and map updates, forming a closed-loop system of "perception-fusion-localization-mapping-map inverse constraint". The multi-layer fusion localization module uses the EKF method for fusion localization. The authentication map is a specially made prior map based on prior map data, with hash fingerprint anti-counterfeiting, cloud incremental review, and version full-link audit and control mechanisms superimposed.

[0180] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 7As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs in the non-volatile storage media to run. The database stores environmental perception data, map prior data, and multi-source positioning sensor data of the target vehicle. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a vehicle fusion positioning method.

[0181] Those skilled in the art will understand that Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer equipment to which the present application is applied. Specific computer equipment may include, for example, [the following is a list of possible additional structures]. Figure 7 The diagram shows more or fewer components, or combinations of certain components, or different component arrangements.

[0182] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0183] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0184] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0185] Compared to existing vehicle multi-sensor fusion localization and mapping technologies, this application significantly improves the continuity, robustness, map reliability, and all-scenario adaptability of vehicle localization in complex scenarios by introducing scene state parameters, information entropy covariance modulation mechanisms, map confidence inverse constraint mechanisms, and a multi-layer localization architecture. Existing technologies typically employ fixed weights or empirical covariance adjustment methods, which are prone to localization drift, localization jumps, and even system lock-out problems in scenarios with GNSS occlusion, visual degradation, point cloud geometric degradation, or map mismatch. This application, however, can dynamically adjust the fusion strategy according to the real-time environment, thereby effectively improving the stability and reliability of localization under complex conditions.

[0186] This application constructs a scene confidence state machine to perform real-time identification of the vehicle's environment, including scenarios such as urban canyons, tunnels, underground parking garages, highways, rainy, snowy, and foggy weather, and dynamic construction areas. The system adaptively switches between different fusion modes based on indicators such as GNSS availability, visual texture quality, laser point cloud structure stability, dynamic obstacle density, and map matching confidence. For example, in urban canyon environments, it automatically reduces GNSS weights and increases visual semantic and HD map matching weights; in tunnel and underground parking garage environments, it disables GNSS constraints and enables laser, IMU, wheel speed, and geomagnetic fusion; in severe weather environments such as rain, snow, and fog, it weakens visual texture features and enhances laser structural features and road topology constraints. Compared to traditional fixed fusion structures, this application can adjust the system state in real time according to scene changes, thereby significantly improving continuous positioning capabilities in complex environments.

[0187] This application further proposes a covariance modulation mechanism based on information entropy, which unifies the visual, laser, GNSS, and map matching states into quantifiable uncertainty indicators, achieving unified information entropy modeling. Compared with existing covariance adjustment methods based on a single threshold or empirical parameter, this application can more accurately describe the real-time health status of sensors and dynamically adjust sensor weights according to the degree of degradation. When the number of visual feature points decreases, reprojection errors increase, laser point clouds exhibit geometric degradation, or GNSS signal quality deteriorates, the corresponding covariance is automatically expanded, reducing the impact of abnormal observations on the overall fusion result. This effectively avoids fusion divergence problems caused by the failure of a single sensor and improves the overall robustness of the method.

[0188] This application also proposes a map confidence reverse constraint mechanism, changing the traditional approach where HD maps are only used as static priors in positioning. The system dynamically evaluates map confidence by calculating the matching residuals between sensor observations and prior map data in real time. When the map is highly consistent with the real road environment, the system strengthens map constraints to improve global positioning accuracy; conversely, when road construction, lane changes, traffic sign alterations, or temporary road closures cause inconsistencies between the map and reality, the system automatically reduces map constraint weights to prevent erroneous map priors from misleading the positioning results. Compared to existing HD map positioning technologies, this application effectively improves positioning stability and map adaptability in complex and dynamic road environments.

[0189] Furthermore, this application employs a multi-layer positioning architecture, combining the IMU, wheel velocity sensor, and VINS to form a local high-frequency positioning layer for short-term continuous pose prediction; combining FAST-LIO2 and point cloud matching to form a mid-frequency structural constraint layer for providing stable structured positioning; and combining RTK, HD map, geomagnetic map, and dynamic event layer to form a low-frequency global correction layer for providing global consistency constraints. Through this multi-layer positioning collaboration mechanism, this application can balance high-frequency real-time performance, mid-frequency robustness, and low-frequency global accuracy, overcoming the difficulty of balancing real-time performance and global consistency in traditional single-layer fusion structures.

[0190] This application also constructs a multi-level redundant positioning mechanism. When GNSS fails, the system automatically switches to a fusion mode of laser, vision, and IMU; when laser degradation occurs, visual semantics, wheel speed, and map topology constraints are enhanced; and when both GNSS and vision fail simultaneously, geomagnetic, IMU, and wheel speed fusion positioning is activated. Through this multi-level redundancy fallback mechanism, this application can maintain continuous and stable positioning output in tunnels, underground parking garages, severe weather, and complex dynamic traffic scenarios, significantly improving the system's fault tolerance and engineering reliability.

[0191] Regarding map generation and updates, this application utilizes map version hashing, timestamp signing, incremental map review, and dynamic event layer management mechanisms to achieve traceable, verifiable, and auditable updates of map data. Compared to traditional map systems, this application not only enables high-precision map construction but also ensures the reliability and security of map updates, making it more suitable for long-term, large-scale operation scenarios of autonomous logistics vehicles.

[0192] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0193] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0194] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0195] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0196] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A vehicle fusion positioning method, characterized in that, The method includes: The system acquires environmental perception data, map prior data, and multi-source positioning sensor data of the target vehicle; the multi-source positioning sensor data is collected by multiple sensors of different types. The motion state parameters of the target vehicle are obtained by solving the map prior data and the multi-source positioning sensor data respectively; the motion state parameters include visual pose, lidar pose and global correction constraints. Based on the environmental perception data, the multi-source positioning sensor data, and the motion state parameters, a confidence quantification assessment is performed to obtain scene state parameters; the scene state parameters include observation confidence index, weather confidence index, and obstacle confidence index; Based on the scene state parameters, the multi-source positioning sensor data, and the map prior data, the original covariance of each sensor is modulated in stages to obtain the modulation covariance of each sensor. Based on the modulation covariance of each sensor, the multi-source positioning sensor data, and the motion state parameters, a multi-sensor fusion filtering method is used to obtain the fusion positioning result of the target vehicle.

2. The vehicle fusion positioning method according to claim 1, characterized in that, The multi-source positioning sensor data includes inertial observation data, odometer observation data, visual observation data, lidar observation data, and satellite observation data; the visual pose includes visual local position coordinates, visual local attitude angles, and velocity components; the lidar pose includes lidar local position coordinates and lidar local attitude angles; the global correction constraints include global coordinate constraints, road topology constraints, and lane geometry constraints. The motion state parameters of the target vehicle are obtained by separately processing the map prior data and the multi-source positioning sensor data, specifically including: The visual pose is obtained by fusing and solving the inertial observation data, the odometer observation data, and the visual observation data. The lidar observation data is processed into point cloud front-end to obtain an initial pose estimate; The initial pose estimation is matched with point cloud using a point cloud registration method to obtain the matched pose. The LiDAR pose is obtained by performing local map matching correction on the matched pose. The global coordinate constraints are constructed based on the satellite observation data. The road topology constraints and lane geometry constraints are constructed based on the prior map data.

3. The vehicle fusion positioning method according to claim 1, characterized in that, The multi-source positioning sensor data includes visual observation data and lidar observation data; Based on the environmental perception data, the multi-source positioning sensor data, and the motion state parameters, a confidence quantification assessment is performed to obtain scene state parameters, specifically including: The scene confidence level is evaluated on the motion state parameters to obtain the observation confidence index; The environmental perception data is quantitatively evaluated to obtain the weather confidence index; Based on the lidar observation data and the visual observation data, obstacle detection is performed to obtain the obstacle distribution density, and the confidence level of the obstacle distribution density is quantitatively evaluated to obtain the obstacle confidence index.

4. The vehicle fusion positioning method according to claim 1, characterized in that, Based on the scene state parameters, the multi-source positioning sensor data, and the map prior data, the original covariance of each sensor is modulated in stages to obtain the modulation covariance of each sensor, specifically including: Based on the scene state parameters, the health indicators of each sensor are calculated; For any given sensor, construct the sensor's information entropy based on the sensor's health indicators; Based on the information entropy, the original covariance of the sensor is initially modulated using the following formula to obtain the initial modulation covariance: ； ； in, Let be the covariance inflation coefficient of the i-th sensor. Let be the information entropy of the i-th sensor. Let be the information entropy sensitivity coefficient of the i-th sensor. Let be the initial modulation covariance of the i-th sensor. Let be the original covariance of the i-th sensor; The map matching residual is calculated based on the multi-source positioning sensor data and the map prior data. The map confidence index is calculated based on the map matching residuals. Based on the map confidence index, the initial modulation covariance is modulated a second time using the following formula to obtain the second modulation covariance: ； in, Let be the quadratic modulation covariance of the i-th sensor. The confidence index for the map is... The constant of the stability term; The modulation covariance is obtained based on the second-order modulation covariance.

5. The vehicle fusion positioning method according to claim 1, characterized in that, The multi-sensor fusion filtering method is the extended Kalman filter method.

6. The vehicle fusion positioning method according to claim 1, characterized in that, After obtaining the fused positioning result of the target vehicle by using a multi-sensor fusion filtering method based on the modulation covariance of each sensor, the multi-source positioning sensor data, and the motion state parameters, the vehicle fusion positioning method further includes: Based on the fusion positioning results of the target vehicle, the multi-source positioning sensor data, and the motion state parameters, a local real-time map is constructed. Based on the local real-time map and the prior map data, the map incremental data is calculated. The map prior data is updated based on the incremental map data.

7. A vehicle fusion positioning device for implementing the vehicle fusion positioning method according to any one of claims 1-6, characterized in that, The vehicle fusion positioning device includes: The data acquisition module is used to acquire environmental perception data, map prior data, and multi-source positioning sensor data of the target vehicle; the multi-source positioning sensor data is collected by multiple different types of sensors; The motion state calculation module is used to calculate the motion state parameters of the target vehicle by solving the map prior data and the multi-source positioning sensor data respectively; the motion state parameters include visual pose, lidar pose and global correction constraints. The scene state assessment module is used to perform confidence quantification assessment based on the environmental perception data, the multi-source positioning sensing data, and the motion state parameters to obtain scene state parameters; the scene state parameters include observation confidence index, weather confidence index, and obstacle confidence index; The covariance modulation module is used to modulate the original covariance of each sensor in stages according to the scene state parameters, the multi-source positioning sensing data and the map prior data, so as to obtain the modulation covariance of each sensor. The vehicle fusion positioning module is used to obtain the fusion positioning result of the target vehicle by using a multi-sensor fusion filtering method based on the modulation covariance of each sensor, the multi-source positioning sensing data, and the motion state parameters.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the vehicle fusion positioning method according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the vehicle fusion positioning method according to any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the vehicle fusion positioning method according to any one of claims 1-6.

Images