High-precision navigation positioning method and system fused with multi-source sensing data

Abstract

The invention provides a high-precision navigation positioning method and system fusing multi-source sensing data, and belongs to the technical field of vehicle navigation positioning, and the method comprises the following steps: activating a multi-source sensing collector of a vehicle to execute data collection, establishing a multi-source data set comprising GNSS, inertial measurement, a wheel speed meter, a visual image, a laser radar and Internet of Vehicles data, mapping the initial pose information into initial pose information; constructing a scene weight factor by using a scene perception model comprising a road type analysis layer, a traffic environment perception layer and a vehicle motion state evaluation layer; after weighted fusion, generating a compensation pose through vehicle dynamics, road geometry and traffic flow constraint compensation; performing local-global verification on the compensation pose to generate a positioning result; the technical problem that high-precision positioning is difficult to achieve due to the fact that positioning is poor in scene adaptability and incomplete in compensation mechanism in the prior art is solved, and the technical effect that high-precision navigation positioning of the vehicle is achieved by improving the positioning scene adaptability and improving the compensation mechanism is achieved.

Description

Technical Field

[0001] This invention belongs to the field of vehicle navigation and positioning technology, specifically relating to a high-precision navigation and positioning method and system that integrates multi-source sensing data. Background Technology

[0002] In intelligent transportation systems, high-precision vehicle navigation and positioning is the fundamental support for core functions such as path planning, automatic obstacle avoidance, and lane keeping. Its accuracy directly determines the safety and reliability of autonomous driving systems. Currently, the industry generally adopts multi-source perception solutions that integrate GNSS information, inertial measurement data, wheel speed sensor data, LiDAR data, visual images, and vehicle-to-everything (V2X) data to improve positioning performance. However, existing multi-source fusion positioning technologies still have the following problems: First, insufficient scene adaptability. Most solutions use a fixed weight allocation strategy to process multi-source data without dynamically adjusting according to the actual driving scenario, resulting in increased positioning deviations in high-interference scenarios. Second, imperfect compensation mechanisms. Existing solutions mostly rely on single-dimensional constraints, making it difficult to correct deviations in the fusion results caused by sensor errors and external environmental interference. Ultimately, this makes it difficult for existing positioning technologies to stably achieve high-precision positioning in complex scenarios.

[0003] Existing positioning technologies suffer from poor scene adaptability and imperfect compensation mechanisms, making it difficult to achieve high-precision positioning. Summary of the Invention

[0004] This invention provides a high-precision navigation and positioning method and system that integrates multi-source sensing data. This solves the technical problems in the prior art, such as poor scene adaptability and imperfect compensation mechanism, which make it difficult to achieve high-precision positioning. It achieves the technical effect of improving the adaptability of positioning scenes and perfecting the compensation mechanism to realize high-precision navigation and positioning of vehicles.

[0005] In view of the above problems, the present invention adopts the following technical solution:

[0006] In a first aspect, the present invention provides a high-precision navigation and positioning method that integrates multi-source perception data. The method includes: activating the multi-source perception collector of a vehicle to perform data acquisition, establishing a multi-source dataset, and mapping the multi-source dataset to initial pose information. The multi-source dataset includes GNSS information, inertial measurement data, wheel speedometer data, visual images, lidar data, and vehicle-to-everything (V2X) data. A scene perception model is used to perform scene trust perception on the multi-source dataset, establishing scene-based weighting factors. The scene perception model includes a road type analysis layer, a traffic environment perception layer, and a vehicle motion state evaluation layer. Based on the scene-based weighting factors, the initial pose information is weighted and fused, and then the weighted fusion result is compensated through vehicle dynamics constraints, road geometric constraints, and traffic flow constraints to generate a compensated pose. The compensated pose is then subjected to local-global fusion verification, and a pose navigation and positioning result is generated based on the fusion verification result.

[0007] On the other hand, the present invention also provides a high-precision navigation and positioning system that integrates multi-source perception data. The system includes: a data acquisition and pose mapping module, used to activate the vehicle's multi-source perception collector to perform data acquisition, establish a multi-source dataset, and map the multi-source dataset to initial pose information. The multi-source dataset includes GNSS information, inertial measurement data, wheel speedometer data, visual images, lidar data, and vehicle network data. A scene-based weighting factor establishment module, used to use a scene perception model to perform scene trust perception of the multi-source dataset and establish scene-based weighting factors. The scene perception model includes a road type analysis layer, a traffic environment perception layer, and a vehicle motion state evaluation layer. A weighted fusion and constraint compensation module, used to perform weighted fusion of the initial pose information based on the scene-based weighting factors, and then compensate the weighted fusion result through vehicle dynamics constraints, road geometric constraints, and traffic flow constraints to generate a compensated pose. A fusion verification and positioning result generation module, used to perform local-global fusion verification on the compensated pose and generate a pose navigation and positioning result based on the fusion verification result.

[0008] One or more technical solutions provided in this invention have at least the following technical effects or advantages:

[0009] The vehicle's multi-source perception and acquisition system is activated to collect data and establish a multi-source dataset. This dataset is then mapped to initial pose information. The multi-source dataset includes GNSS information, inertial measurement data, wheel speedometer data, visual images, lidar data, and vehicle-to-everything (V2X) data. A scene perception model is used to perform scene trust perception on the multi-source dataset, establishing scene-based weighting factors. This scene perception model includes a road type analysis layer, a traffic environment perception layer, and a vehicle motion state evaluation layer. Based on the scene-based weighting factors, the initial pose information is weighted and fused. The fusion result is then compensated using vehicle dynamics constraints, road geometry constraints, and traffic flow constraints to generate a compensated pose. The compensated pose undergoes local-global fusion verification, and a pose navigation and positioning result is generated based on the verification result. This achieves the technical effect of improving the adaptability of the positioning scene and perfecting the compensation mechanism to realize high-precision vehicle navigation and positioning. Attached Figure Description

[0010] Figure 1 This is a flowchart illustrating the high-precision navigation and positioning method that integrates multi-source sensing data according to the present invention.

[0011] Figure 2 This is a schematic diagram of the high-precision navigation and positioning system that integrates multi-source sensing data according to the present invention;

[0012] Figure labeling: 11 Data acquisition and pose mapping module, 12 Scene-based weight factor establishment module, 13 Weighted fusion and constraint compensation module, 14 Fusion verification and positioning result generation module. Detailed Implementation

[0013] This invention provides a high-precision navigation and positioning method and system that integrates multi-source sensing data. This solves the technical problems in the prior art, such as poor scene adaptability and imperfect compensation mechanism, which make it difficult to achieve high-precision positioning. It achieves the technical effect of improving the adaptability of positioning scenes and perfecting the compensation mechanism to realize high-precision navigation and positioning of vehicles.

[0014] like Figure 1 As shown, this invention provides a high-precision navigation and positioning method that fuses multi-source sensing data, the method comprising:

[0015] S100: Activate the vehicle's multi-source perception and acquisition device to perform data acquisition, establish a multi-source dataset, and map the multi-source dataset to initial pose information. The multi-source dataset includes GNSS information, inertial measurement data, wheel speedometer data, visual images, lidar data, and vehicle network data.

[0016] Specifically, in high-precision navigation and positioning by integrating multi-source sensing data, the first step is to activate the multi-source sensing data acquisition unit on the vehicle, enabling it to enter working mode and collect data. The collected data covers six core categories: GNSS information, inertial measurement data, wheel speedometer data, visual images, LiDAR data, and vehicle-to-everything (V2X) data. These collected data from different sources are then integrated and summarized to form a multi-source dataset. Specific data processing methods are then used to transform this multi-source dataset into initial pose information reflecting the vehicle's current position and attitude. These specific data processing methods include first using Kalman filtering to denoise the GNSS information, then performing integral calculations on the inertial measurement data to determine motion parameters, calibrating speed deviations using wheel speedometer data, then completing details through feature matching of visual images and LiDAR, and finally unifying the coordinate system to transform the multiple types of data into initial pose information containing latitude, longitude, and heading angle.

[0017] The multi-source perception data acquisition unit for a vehicle is a collective term for various devices installed on the vehicle to collect the six different types of perception data mentioned above. These include a GNSS receiver for collecting GNSS information, an inertial measurement unit for acquiring inertial measurement data, a wheel speed sensor for collecting wheel speed data, an onboard camera for capturing visual images, an onboard LiDAR device for acquiring LiDAR data, and an onboard LiDAR communication module for receiving vehicle network data. The multi-source dataset is the dataset formed after preliminary integration of the six types of raw perception data acquired from the different acquisition devices. Initial pose information is the basic information obtained after preliminary processing of the multi-source dataset, which can initially reflect the vehicle's position and attitude at the current moment; position includes longitude and latitude coordinates; attitude includes heading angle, pitch angle, and roll angle. Mapping the multi-source dataset to initial pose information involves using data processing algorithms to transform the originally scattered and differently formatted multi-source perception data into initial pose information in a unified format that directly reflects the vehicle's position and attitude.

[0018] S200: Use a scene perception model to perform scene trust perception on the multi-source dataset and establish scene-based weighting factors. The scene perception model includes a road type analysis layer, a traffic environment perception layer, and a vehicle motion state evaluation layer.

[0019] Specifically, after establishing the multi-source dataset and mapping the initial pose information, to ensure that the subsequent weighted fusion of the initial pose information conforms to the actual positioning scenario and improves the rationality of the fusion, a scene perception model is used to perform scene trust perception on the multi-source dataset, and a scene-based weighting factor is established based on the scene trust perception results. The scene perception model achieves comprehensive perception of the positioning scenario and data trust assessment through the collaborative action of the road type analysis layer, traffic environment perception layer, and vehicle motion state evaluation layer. The data trust assessment combines the three layers of the scene perception model to judge the reliability of multi-source data in the current scenario; for example, the road type analysis layer determines that the GNSS information is unobstructed in a straight road scenario, and the trust level is set to 0.8; when the traffic environment layer identifies congestion, the LiDAR data is anti-interference, and the trust level is set to 0.9; when the vehicle motion layer detects rapid acceleration, the inertial measurement data is accurate, and the trust level is set to 0.85.

[0020] The scene perception model, comprised of a road type analysis layer, a traffic environment perception layer, and a vehicle motion state evaluation layer, is used to process multi-source datasets and assess the trustworthiness of scene-dimensional data. Scene trust perception combines the actual characteristics of the positioning scene to determine the reliability and applicability of various data types in the multi-source dataset within the current scene, clarifying the relative strength of different data in supporting positioning. Multi-source dataset scene trust perception, based on the three-layer structure of the scene perception model, analyzes the trustworthiness of various data types in the current scene within the constructed multi-source dataset, filtering out data more effective for positioning. Scene-specific weighting factors, based on the scene trust perception results of the multi-source dataset, assign weight values ​​to each type of data in the multi-source dataset, adapted to the current positioning scene. These weights are subsequently used in the initial pose information weighted fusion to ensure a higher proportion of high-trust data in the fusion process.

[0021] S300: After weighted fusion of the initial pose information based on the scenario-based weighting factors, the weighted fusion result is compensated by vehicle dynamics constraints, road geometry constraints, and traffic flow constraints to generate a compensated pose.

[0022] Specifically, after establishing scenario-based weighting factors, the initial pose information acquired earlier is weighted and fused based on these factors. This allows data with higher credibility in the current scenario to play a greater role in the fusion process, initially improving the scenario fit of the pose information. However, due to external environmental interference and other factors, the weighted fusion results may have deviations. Therefore, targeted compensation and correction are performed using vehicle dynamics constraints, road geometry constraints, and traffic flow constraints to ultimately generate a more accurate compensated pose.

[0023] Among these, vehicle dynamics constraints are reasonable limitations imposed on the weighted fusion results based on the vehicle's own physical motion laws, ensuring that the pose information conforms to the vehicle's mechanical performance and motion logic. Road geometry constraints are adaptive adjustments to the weighted fusion results based on the inherent geometric features of the road, ensuring that the pose information is consistent with the actual road shape. Traffic flow constraints are coordinated corrections to the weighted fusion results based on the current traffic flow status, such as the average flow velocity and traffic direction, ensuring that the pose information conforms to the actual traffic flow situation. Compensated pose is the pose information with higher accuracy and more consistent with the actual scenario obtained after correcting and compensating the weighted fusion results through vehicle dynamics constraints, road geometry constraints, and traffic flow constraints.

[0024] S400: Perform local-global fusion verification on the compensated pose, and generate pose navigation and positioning results based on the fusion verification results.

[0025] Specifically, after acquiring the compensated pose, to avoid potential issues such as local trajectory inconsistencies or global position drift, and to ensure that the positioning results satisfy both short-term motion continuity and long-term position accuracy, a local-global fusion verification is performed on the compensated pose. At the local level, the short-term change patterns of the compensated pose are examined to determine whether the pose transitions between adjacent moments conform to the normal vehicle motion logic. At the global level, the overall positional accuracy of the compensated pose is verified, confirming its matching degree with a large-scale positioning reference. Based on the fusion verification results from these two aspects, the compensated pose is then finely corrected, ultimately generating a pose navigation and positioning result that can directly serve vehicle navigation. The local-global fusion verification combines local short-term pose continuity checks with global long-term pose accuracy verification to perform a two-way check on the compensated pose to eliminate deviations. The pose navigation and positioning result, after local-global fusion verification and correction, is the final output data that accurately reflects the vehicle's real-time position and attitude.

[0026] Furthermore, the present invention also provides a method for scene trust perception of the multi-source dataset using a scene perception model and for establishing scene-based weighting factors, including: reading navigation map data from the vehicle's infotainment system and synchronously inputting the navigation map data as auxiliary data into the scene perception model; extracting road category features from the navigation map data using the road type analysis layer, and identifying lane width features, lane curvature features, and intersection structure features from LiDAR data and visual images to establish a lane feature dataset; performing lane scene adaptation perception based on the lane feature dataset and establishing a first weighting factor; and establishing scene-based weighting factors using the first weighting factor.

[0027] Specifically, in the process of using a scene-aware model to perform scene trust perception on multi-source datasets and then establish scene-based weighting factors, the navigation map data stored in the vehicle's infotainment system or acquired online is first read. This navigation map data is used as auxiliary information and is synchronously input into the scene-aware model along with the multi-source datasets, providing a map dimension reference for subsequent scene feature extraction. Next, the road type analysis layer in the scene-aware model is invoked. On one hand, road category features that can distinguish road types are extracted from the navigation map data. On the other hand, lane width features reflecting the lateral dimensions of the lanes, lane curvature features reflecting the degree of lane curvature, and intersection structure features describing the intersection layout are identified from the LiDAR data and visual images in the multi-source datasets. These extracted features are then integrated to establish a lane feature dataset that comprehensively reflects the current lane scene. Subsequently, based on this lane feature dataset, the current lane scene of the vehicle is analyzed, such as straight roads, curves, and areas near intersections, to determine scene adaptability. Based on the adaptability results, initial weights are assigned to different types of data in the multi-source datasets, establishing the first weighting factor. Finally, based on the first weight factor, a scenario-specific weight factor that fits the current scenario is further constructed, providing a basis for the weighted fusion of the initial pose information in the future.

[0028] The navigation map data consists of electronic map data stored within the vehicle's infotainment system or acquired via the network, containing information such as road location, road attributes, and road network structure. Examples include map data storing the location coordinates and related parameters of urban ring roads, suburban expressways, and rural cement roads. Road types are categorized according to road function and construction standards, such as expressways, urban expressways, ordinary municipal roads, and rural roads. Road category features are specific attribute information used to distinguish different road types, such as the speed limit of 120 km / h, four lanes in both directions, and the presence of a central median on expressways. Lane width features describe the specific data characteristics of the lateral width of a single lane, such as the standard width of a standard urban motor vehicle lane being 3.5 meters. Lane curvature features reflect the degree of curvature of the lane line, usually expressed as the radius of curvature or curvature value. Intersection structure features describe the shape of road intersections, lane connection methods, and traffic organization patterns. The lane feature dataset is a dataset formed by summarizing and integrating the extracted road category features, lane width features, lane curvature features, and intersection structure features.

[0029] The first weighting factor is the initial weight value assigned to different perception data in the multi-source dataset after adapting the perception to the current lane scene based on the lane feature dataset. For example, in a straight road scenario, the first weighting factor established based on the lane feature dataset might assign 0.4 to GNSS data, 0.3 to LiDAR data (due to less GNSS signal obstruction and higher reliability), and 0.2 to visual image data. The scenario-based weighting factor is a weighting factor built upon the first weighting factor that accurately fits the current lane scene. It is used for subsequent weighted fusion of the initial pose information. For example, based on the first weighting factor for the straight road segment, the scenario-based weighting factor, after fine-tuning, still maintains a higher weight for GNSS data, ensuring that high-reliability data is prioritized during fusion, thus improving the scenario adaptability of the fusion result.

[0030] Furthermore, the present invention also provides a method for establishing a scenario-based weighting factor using the first weighting factor, comprising: extracting visual images, LiDAR data, and vehicle network data from a multi-source dataset using a traffic environment perception layer; establishing a traffic environment feature set based on the extraction results, wherein the traffic environment feature set includes dynamic state features of vehicles, pedestrians, and obstacles, traffic density features, traffic flow direction features, lane occupancy features, traffic light state features, and landmark features; establishing a second weighting factor using the traffic environment feature set; extracting inertial measurement data and wheel speedometer data from a multi-source dataset using a vehicle motion state evaluation layer; establishing a vehicle motion state feature set, wherein the vehicle motion state feature set includes rapid acceleration features, sharp turning features, skidding features, and braking abrupt change features; establishing a third weighting factor based on the vehicle motion state feature set; and constructing a scenario-based weighting factor using the first weighting factor, the second weighting factor, and the third weighting factor.

[0031] Specifically, when constructing scenario-based weighting factors using the first weighting factor, the traffic environment perception layer in the scenario perception model is first activated. It filters and extracts three types of data directly related to the traffic environment from multi-source datasets: visual images, LiDAR data, and vehicle-to-everything (V2X) data. These data undergo preliminary analysis and feature extraction to obtain the extraction results. Then, all extraction results are integrated and categorized to establish a traffic environment feature set containing multi-dimensional information such as the dynamic states of vehicles, pedestrians, and obstacles. Subsequently, based on this traffic environment feature set, the reliability of different perception data in the current traffic environment is analyzed. For example, LiDAR data is more stable than GNSS data in congested environments. Corresponding weights are assigned to each type of data, thus establishing the second weighting factor. Next, the vehicle motion state evaluation layer of the scenario perception model is invoked. Inertial measurement data and wheel speed measurement data that reflect the vehicle's own motion are extracted from the multi-source datasets. These two types of data are processed to identify the vehicle's motion state, thereby establishing a vehicle motion state feature set including features such as rapid acceleration and sharp turns. Then, based on this feature set, the adaptability of different perception data to the vehicle motion state is judged. For example, inertial measurement data is more accurate during rapid acceleration, thus establishing the third weighting factor. Finally, the obtained first, second, and third weight factors are integrated and their weights are adjusted to construct a scenario-based weight factor that can fully adapt to the current scenario.

[0032] The traffic environment perception layer is a functional level within the scene perception model used to extract and analyze traffic environment-related data from multi-source datasets, thereby capturing traffic environment characteristics. For example, when a vehicle approaches an intersection, this layer focuses on extracting pedestrian and traffic light data from visual images to determine the complexity of the intersection's traffic environment. Visual images are real-time images captured by the vehicle's onboard cameras that reflect visual information about the surrounding environment, such as clear images of pedestrians crossing the road at an intersection or roadside traffic signs captured by the camera. LiDAR data is 3D point cloud data obtained by the onboard LiDAR device through scanning the surrounding environment with laser beams, containing information such as target distance, contour, and position, such as the 3D contour of a guardrail 100 meters ahead or the position data of an obstacle 5 meters away from the vehicle scanned by LiDAR. Vehicle-to-everything (V2X) data is data related to traffic conditions and roadside equipment information transmitted through vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communication technologies. The extraction result is the effective information related to the traffic environment extracted by the traffic environment perception layer after analyzing visual images, LiDAR data, and V2X data. A traffic environment feature set is a dataset that comprehensively reflects the current traffic environment by summarizing all extracted results by category. It includes features such as vehicle dynamics, pedestrian dynamics, traffic density, traffic flow direction, lane occupancy, traffic light status, and landmarks.

[0033] The dynamic state characteristics of vehicles, pedestrians, and obstacles are concentrated traffic environment features, describing the real-time movement of vehicles, pedestrians, and obstacles. Traffic density characteristics reflect the number of vehicles and pedestrians per unit length of road segment or unit area. Traffic flow direction characteristics reflect the travel direction of most vehicles in the current road segment. Lane occupancy characteristics describe the occupancy of each lane by vehicles and obstacles. Traffic light status characteristics reflect information such as the current color and countdown timer of the traffic lights. Landmark characteristics reflect information about various traffic signs, fixed facilities, and other landmarks on the road.

[0034] The second weighting factor is the weight value assigned to various types of data after analyzing the credibility of different perception data in the current traffic environment based on the traffic environment feature set. For example, in traffic congestion and pedestrian-heavy environments, LiDAR can accurately identify obstacles, so the weight of LiDAR data is set to 0.4; visual image data can clearly see traffic lights and pedestrians, so the weight of visual image data is set to 0.3; GNSS data is easily obstructed and has low accuracy, so the weight of GNSS data is set to 0.1; and vehicle-to-everything (V2X) data provides road conditions, so the weight of V2X data is set to 0.2. The vehicle motion state evaluation layer is a functional level in the scene perception model specifically used to extract and analyze the vehicle's own motion data, including inertial measurement data and wheel speed measurement data, to determine the vehicle's motion state. For example, by analyzing inertial measurement data, it can be determined whether the vehicle is in a state of rapid acceleration or sharp turning. Inertial measurement data is collected by the inertial measurement unit (IMU) on the vehicle and includes vehicle acceleration, including longitudinal, lateral, and vertical axes, and angular velocities including roll, pitch, and heading, which are physical quantities reflecting the vehicle's motion state. Wheel speed sensor data, collected by the vehicle's wheel speed sensors, reflects real-time wheel rotation speed. The approximate vehicle speed can be calculated from the wheel speed. The vehicle motion state feature set is a set of features reflecting the core motion state of the vehicle, extracted by the vehicle motion state evaluation layer after processing the inertial measurement data and wheel speed sensor data. Rapid acceleration features describe vehicle acceleration exceeding a preset threshold. Sharp turning features describe steering angular velocity exceeding a preset threshold. Slippage features describe a mismatch between wheel speed and actual vehicle speed. Sudden braking features describe deceleration exceeding a preset threshold during braking.

[0035] The third weighting factor is the weight value assigned to various types of data after analyzing the adaptability of different perception data to the current vehicle motion state based on the vehicle motion state feature set. For example, when the vehicle is in a state of rapid acceleration, inertial measurement data can accurately capture acceleration, so the weight is set to 0.4; wheel speed meter data assists in judging vehicle speed, so the weight is set to 0.3; lidar data has low adaptability to motion state, so the weight is set to 0.15; and visual image data has a weight of 0.15.

[0036] Contextualized weighting factors are the final weight values ​​obtained by comprehensively weighting and integrating the first, second, and third weighting factors. These weights fully fit the current overall scenario and are used for subsequent weighted fusion of initial pose information. For example, in a straight urban road segment, with the first weighting factor being GNSS 0.3, LiDAR 0.25, and traffic congestion, the second weighting factor being LiDAR 0.4, vision 0.3, and rapid vehicle acceleration, and the third weighting factor being inertial measurement 0.4 and wheel speed measurement 0.3, the integrated contextualized weighting factors might be GNSS 0.2, LiDAR 0.35, vision 0.2, inertial measurement 0.15, wheel speed measurement 0.1, and vehicle network data 0.0, ensuring that highly adaptable data plays a greater role in the fusion process.

[0037] Furthermore, the present invention also provides a local-global fusion verification of the compensated pose, and generates a pose navigation and positioning result based on the fusion verification result, including: after configuring an adaptive sliding window, performing local trajectory fitting on the compensated pose based on the adaptive sliding window to establish a local trajectory reference; after extracting GNSS reference points from GNSS information, configuring a global positioning reference using the GNSS reference points to establish a global trajectory reference; performing multi-scale residual analysis on the local trajectory reference and the global trajectory reference, wherein the multi-scale residual analysis includes short-time translation residuals, long-time curvature residuals, and topology matching residuals; and establishing a fusion verification result after weighted correction of the multi-scale residual analysis results.

[0038] Specifically, when performing local-global fusion verification of the compensated pose and generating pose navigation and positioning results based on the verification results, an adaptive sliding window is first configured. This window automatically adjusts its temporal or spatial range according to the frequency of changes in the compensated pose. Then, using the adaptive sliding window as the analysis range, continuous compensated poses within the window are fitted to generate a local trajectory reference reflecting the vehicle's short-term motion trajectory. Subsequently, stable signal and low-error position data are selected from GNSS information in multi-source datasets as GNSS reference points. These reference points are used to construct a global positioning reference frame covering a large area, which is then connected to form a global trajectory reference reflecting the vehicle's long-term position trend. Next, multi-scale residual analysis is performed on the local and global trajectory references, focusing on calculating the deviations between them in three dimensions: short-term position offset, long-term trajectory curvature, and road network structure matching degree—namely, short-term translational residuals, long-term curvature residuals, and topology matching residuals. Finally, the analysis results are corrected and adjusted according to the influence weights of different residuals to establish a fusion verification result that eliminates local deviations and global drift. Based on this result, accurate pose navigation and positioning results are output.

[0039] The fusion verification result is the pose verification conclusion obtained after multi-scale residual analysis and weighted correction, eliminating local anomalies and global drift. For example, if the compensated pose had a short-term offset of 0.6 meters and a global drift of 1 meter before correction, the verification result adjusts it to a lateral offset of 0.1 meters and a global deviation of 0.2 meters, ensuring a reasonable pose. The pose navigation and positioning result is the core data that accurately reflects the vehicle's real-time position and attitude, based on the fusion verification result. The adaptive sliding window is a dynamic window that automatically adjusts the analysis range according to the rate of change of the compensated pose.

[0040] Local trajectory fitting is the process of fitting data to continuously compensated poses within an adaptive sliding window to generate a short-term trajectory. GNSS information refers to data collected by the GNSS receiver on the vehicle, including the vehicle's latitude, longitude, elevation, and other location information. GNSS reference points are high-confidence location points selected from the GNSS information that have unobstructed signals and an error of less than 0.3 meters. Global positioning reference is a reference framework covering a large area, built based on multiple GNSS reference points, used to calibrate global position. For example, five GNSS reference points are selected within a 10-kilometer road segment, and a position reference line covering that segment is drawn as the global positioning reference. Global trajectory reference is a trajectory generated based on the global positioning reference and combined with the vehicle's driving trend, reflecting the vehicle's long-term position changes. For example, five GNSS reference points are connected in driving order to form a smooth curve as the global trajectory reference. Multi-scale residual analysis is the process of calculating the deviation between the local trajectory reference and the global trajectory reference at three different scales: short-term position, long-term curvature, and road network matching. Short-time translational residuals, in multi-scale residual analysis, reflect the short-term positional translational deviation between the local and global trajectory references. Long-time curvature residuals, in multi-scale residual analysis, reflect the long-term deviation in trajectory curvature between the local and global trajectory references. Topology matching residuals, in multi-scale residual analysis, reflect the deviation in matching between the local and global trajectory references in terms of road network topology, such as road connectivity and lane distribution.

[0041] Furthermore, the present invention also provides a method for compensating for a compensated pose by weighted fusion results of vehicle dynamics constraints, road geometric constraints, and traffic flow constraints, including: establishing vehicle dynamics constraints, which include yaw rate constraints, longitudinal acceleration constraints, and tire slip constraints on the weighted fusion results; establishing road geometric constraints by performing geometric projection matching on the weighted fusion results of lane lines, curvature, and road boundaries to complete road geometric constraint compensation; and compensating for traffic flow constraints by weighted fusion results of traffic flow velocity distribution and velocity consistency identification.

[0042] Specifically, after obtaining the weighted fusion result of the initial pose information, to eliminate potential deviations in physical rationality, road adaptability, and traffic coordination, constraints are established from three dimensions for compensation. First, vehicle dynamics constraints are constructed, limiting yaw rate, longitudinal acceleration, and tire slip ratio within the vehicle's physical performance range to ensure the fusion result conforms to vehicle motion laws. Second, road geometric constraints are established, projecting the pose information of the fusion result onto geometric features such as lane lines, road curvature, and road boundaries to correct deviations from the actual road shape. Finally, traffic flow constraints are applied, adjusting speed or position information in the fusion result that does not match the traffic flow state based on traffic flow velocity distribution and vehicle speed consistency characteristics. Ultimately, the combined effect of these three types of constraints generates a more accurate compensated pose.

[0043] Among these, the weighted yaw rate constraint is a limitation on the vehicle steering angular velocity reflected in the fusion result within the vehicle dynamics constraints, ensuring it does not exceed the mechanical limits of the vehicle steering system. The longitudinal acceleration constraint is a limitation on the longitudinal acceleration or deceleration of the vehicle reflected in the fusion result within the vehicle dynamics constraints, ensuring it does not exceed the performance limits of the vehicle's power or braking system. The tire slip constraint is a limitation on the degree of tire slip reflected in the fusion result within the vehicle dynamics constraints, preventing speed or position deviations caused by slip. The degree of tire slip is represented by the slip ratio.

[0044] Geometric projection matching, within road geometric constraints, involves projecting the pose information of the weighted fusion result onto a road geometric model composed of lane lines, road curvature, and road boundaries, and determining the degree of matching between the two. Road geometric constraint compensation, after identifying deviations through geometric projection matching, corrects the pose information of the weighted fusion result to a range that matches the road geometric model. Traffic flow velocity distribution, within traffic flow constraints, reflects the characteristics of vehicle speed distribution at different locations on the current road segment. Velocity consistency, within traffic flow constraints, reflects whether the speed of a target vehicle is consistent with the speed of other vehicles in the same or adjacent lanes. Weighted fusion result compensation, through three types of constraints—vehicle dynamics, road geometry, and traffic flow—corrects pose information in the weighted fusion result that does not conform to physical laws, road morphology, or traffic flow state.

[0045] Furthermore, the present invention also provides a method for establishing a multi-source dataset, including: after establishing the multi-source dataset, performing data redundancy monitoring to establish a redundant dataset; using the redundant dataset to verify the multi-source dataset, and compressing and storing the redundant data.

[0046] Specifically, after collecting various data such as GNSS information and inertial measurement data through vehicle multi-source sensing and acquisition devices and integrating them to establish a multi-source dataset, to ensure the accuracy and storage efficiency of the dataset, a data redundancy monitoring operation is first performed to filter out redundant data that is duplicated, substitutable, or exceeds the necessary accuracy range. This redundant data is then separately organized into a redundant dataset. Next, the information in the redundant dataset is compared and verified with the corresponding core data in the multi-source dataset to determine whether there are any data errors or deviations in the multi-source dataset, thus completing the validity verification of the multi-source dataset. Finally, to avoid redundant data consuming excessive storage resources, the redundant data used for verification is compressed and stored so that it can be retrieved at any time if secondary verification is required.

[0047] Data redundancy monitoring involves scanning and analyzing established multi-source datasets to identify redundant data that is duplicated, can be derived from other data, or has an accuracy exceeding positioning requirements. The redundant dataset is formed by categorizing and summarizing all redundant data identified during data redundancy monitoring according to data type or collection time. Multi-source dataset verification involves comparing the redundant data in the redundant dataset with the core data in the multi-source dataset to determine whether the core data is accurate and unbiased.

[0048] Furthermore, the present invention also provides a method for generating pose navigation and positioning results based on fusion verification results, including: performing anomaly identification of the acquisition state of the multi-source sensing acquisition device and establishing a perception trust level; and outputting the pose navigation and positioning results after trust compensation of the fusion verification results through the perception trust level.

[0049] Specifically, when generating pose navigation and positioning results based on the fusion verification results, the working status of the multi-source sensing data acquisition devices is first checked to identify any abnormal acquisition conditions such as signal obstruction, equipment failure, data interruption, or distortion. Then, based on the degree of abnormality and data quality, a quantitative assessment of the reliability of each data acquisition device is assigned, establishing a perception trust level. Subsequently, the fusion verification results are adjusted based on this perception trust level. If a data acquisition device has a low trust level, the weight of its corresponding data in the verification results is reduced to minimize interference from abnormal data. Finally, after trust compensation, accurate pose navigation and positioning results are output.

[0050] The perception trust level is a quantitative indicator that evaluates the reliability of the output data based on the acquisition status of the multi-source sensing acquisition device. It is represented by a value in the range of 0-1, where 1 represents that the acquisition device is in good condition and the data is completely reliable. For example, if the LiDAR is fault-free and the data is clear, the perception trust level is set to 0.9; if the data of the vehicle camera is distorted due to the blurry lens in the rain, the perception trust level is set to 0.4.

[0051] This invention provides a multi-source perception collector for activated vehicles to perform data acquisition, establish a multi-source dataset, and map the multi-source dataset to initial pose information. The multi-source dataset includes GNSS information, inertial measurement data, wheel speedometer data, visual images, lidar data, and vehicle-to-everything (V2X) data. A scene perception model is used to perform scene trust perception on the multi-source dataset, establishing scene-based weighting factors. The scene perception model includes a road type analysis layer, a traffic environment perception layer, and a vehicle motion state evaluation layer. Based on the scene-based weighting factors, the initial pose information is weighted and fused. The weighted fusion result is then compensated through vehicle dynamics constraints, road geometry constraints, and traffic flow constraints to generate a compensated pose. The compensated pose is then subjected to local-global fusion verification, and a pose navigation and positioning result is generated based on the fusion verification result. This achieves the technical effect of improving the adaptability of the positioning scene and perfecting the compensation mechanism to realize high-precision vehicle navigation and positioning.

[0052] Based on the same inventive concept as the aforementioned high-precision navigation and positioning method that integrates multi-source sensing data, this invention also provides a high-precision navigation and positioning system that integrates multi-source sensing data, such as... Figure 2 As shown, the system includes:

[0053] The data acquisition and pose mapping module 11 is used to activate the vehicle's multi-source perception acquisition device to perform data acquisition, establish a multi-source dataset, and map the multi-source dataset to initial pose information. The multi-source dataset includes GNSS information, inertial measurement data, wheel speedometer data, visual images, lidar data, and vehicle network data.

[0054] The scenario-based weight factor establishment module 12 is used to establish scenario-based weight factors by using the scenario perception model to perform scenario trust perception of the multi-source dataset. The scenario perception model includes a road type analysis layer, a traffic environment perception layer, and a vehicle motion state evaluation layer.

[0055] The weighted fusion and constraint compensation module 13 is used to perform weighted fusion of the initial pose information based on the scenario-based weighting factors, and then perform weighted fusion result compensation through vehicle dynamics constraints, road geometry constraints, and traffic flow constraints to generate a compensated pose.

[0056] The fusion verification and positioning result generation module 14 is used to perform local-global fusion verification on the compensated pose and generate pose navigation and positioning results based on the fusion verification results.

[0057] Furthermore, the present invention also provides a method for scene trust perception of the multi-source dataset using a scene perception model and for establishing scene-based weighting factors, including: reading navigation map data from the vehicle's infotainment system and synchronously inputting the navigation map data as auxiliary data into the scene perception model; extracting road category features from the navigation map data using the road type analysis layer, and identifying lane width features, lane curvature features, and intersection structure features from LiDAR data and visual images to establish a lane feature dataset; performing lane scene adaptation perception based on the lane feature dataset and establishing a first weighting factor; and establishing scene-based weighting factors using the first weighting factor.

[0058] Furthermore, the present invention also provides a method for establishing a scenario-based weighting factor using the first weighting factor, comprising: extracting visual images, LiDAR data, and vehicle network data from a multi-source dataset using a traffic environment perception layer; establishing a traffic environment feature set based on the extraction results, wherein the traffic environment feature set includes dynamic state features of vehicles, pedestrians, and obstacles, traffic density features, traffic flow direction features, lane occupancy features, traffic light state features, and landmark features; establishing a second weighting factor using the traffic environment feature set; extracting inertial measurement data and wheel speedometer data from a multi-source dataset using a vehicle motion state evaluation layer; establishing a vehicle motion state feature set, wherein the vehicle motion state feature set includes rapid acceleration features, sharp turning features, skidding features, and braking abrupt change features; establishing a third weighting factor based on the vehicle motion state feature set; and constructing a scenario-based weighting factor using the first weighting factor, the second weighting factor, and the third weighting factor.

[0059] Furthermore, the present invention also provides a local-global fusion verification of the compensated pose, and generates a pose navigation and positioning result based on the fusion verification result, including: after configuring an adaptive sliding window, performing local trajectory fitting on the compensated pose based on the adaptive sliding window to establish a local trajectory reference; after extracting GNSS reference points from GNSS information, configuring a global positioning reference using the GNSS reference points to establish a global trajectory reference; performing multi-scale residual analysis on the local trajectory reference and the global trajectory reference, wherein the multi-scale residual analysis includes short-time translation residuals, long-time curvature residuals, and topology matching residuals; and establishing a fusion verification result after weighted correction of the multi-scale residual analysis results.

[0060] Furthermore, the present invention also provides a method for compensating for a compensated pose by weighted fusion results of vehicle dynamics constraints, road geometric constraints, and traffic flow constraints, including: establishing vehicle dynamics constraints, which include yaw rate constraints, longitudinal acceleration constraints, and tire slip constraints on the weighted fusion results; establishing road geometric constraints by performing geometric projection matching on the weighted fusion results of lane lines, curvature, and road boundaries to complete road geometric constraint compensation; and compensating for traffic flow constraints by weighted fusion results of traffic flow velocity distribution and velocity consistency identification.

[0061] Furthermore, the present invention also provides a method for establishing a multi-source dataset, including: after establishing the multi-source dataset, performing data redundancy monitoring to establish a redundant dataset; using the redundant dataset to verify the multi-source dataset, and compressing and storing the redundant data.

[0062] Furthermore, the present invention also provides a method for generating pose navigation and positioning results based on fusion verification results, including: performing anomaly identification of the acquisition state of the multi-source sensing acquisition device and establishing a perception trust level; and outputting the pose navigation and positioning results after trust compensation of the fusion verification results through the perception trust level.

[0063] This manual uses a progressive approach, focusing on the differences from other parts. The foregoing... Figure 1 The high-precision navigation and positioning method and its specific content that integrates multi-source sensing data are also applicable to the high-precision navigation and positioning system that integrates multi-source sensing data described in this section. Through the foregoing detailed description of the high-precision navigation and positioning method that integrates multi-source sensing data, those skilled in the art can clearly understand the high-precision navigation and positioning system that integrates multi-source sensing data; therefore, for the sake of brevity, it will not be described in detail here. For the disclosed system, since it corresponds to the disclosed method, the description is relatively simple; relevant details can be found in the method section.

[0064] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0065] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. A high-precision navigation and positioning method that integrates multi-source sensing data, characterized in that, The method includes: The vehicle's multi-source sensing and acquisition device is activated to perform data acquisition, establish a multi-source dataset, and map the multi-source dataset to initial pose information. The multi-source dataset includes GNSS information, inertial measurement data, wheel speedometer data, visual images, lidar data, and vehicle network data. The scene perception model is used to perform scene trust perception of the multi-source dataset and establish scene-based weight factors. The scene perception model includes a road type analysis layer, a traffic environment perception layer and a vehicle motion state evaluation layer. After weighted fusion of initial pose information based on the aforementioned scenario-based weighting factors, the weighted fusion result is compensated by vehicle dynamics constraints, road geometry constraints, and traffic flow constraints to generate a compensated pose. The compensated pose is subjected to local-global fusion verification, and the pose navigation and positioning result is generated based on the fusion verification result.

2. The high-precision navigation and positioning method that integrates multi-source sensing data as described in claim 1, characterized in that, Utilizing a scene-aware model to perform scene trust perception on the multi-source dataset, and establishing scene-based weighting factors, including: Read the navigation map data from the vehicle's infotainment system and synchronously input the navigation map data as auxiliary data into the scene perception model; The road type analysis layer is used to extract road category features from navigation map data, and lane width features, lane curvature features, and intersection structure features are identified from LiDAR data and visual images to establish a lane feature dataset. Based on the lane feature dataset, lane scene adaptation perception is performed, and a first weighting factor is established; Contextualized weighting factors are established using the first weighting factor.

3. The high-precision navigation and positioning method that integrates multi-source sensing data as described in claim 2, characterized in that, Establishing scenario-based weighting factors using the first weighting factor includes: Visual images, LiDAR data, and vehicle network data from multi-source datasets are extracted using the traffic environment perception layer. Based on the extraction results, a traffic environment feature set is established, which includes dynamic state features of vehicles, pedestrians, and obstacles, traffic density features, traffic flow direction features, lane occupancy features, traffic light status features, and landmark features. A second weighting factor is established using the aforementioned traffic environment feature set; Inertial measurement data and wheel speed meter data are extracted from multi-source datasets using the vehicle motion state evaluation layer to establish a vehicle motion state feature set, which includes rapid acceleration features, sharp turning features, skidding features, and braking abrupt change features. A third weighting factor is established based on the vehicle motion state feature set; Contextualized weighting factors are constructed using the first weighting factor, the second weighting factor, and the third weighting factor.

4. The high-precision navigation and positioning method that fuses multi-source sensing data as described in claim 1, characterized in that, The compensated pose is subjected to local-global fusion verification, and pose navigation and positioning results are generated based on the fusion verification results, including: After configuring the adaptive sliding window, the compensated pose is fitted with a local trajectory based on the adaptive sliding window to establish a local trajectory reference. After extracting GNSS reference points from GNSS information, global positioning references are configured using the GNSS reference points to establish a global trajectory benchmark; Multi-scale residual analysis is performed on the local trajectory reference and the global trajectory reference. The multi-scale residual analysis includes short-time translation residual, long-time curvature residual and topology matching residual. After weighted correction of the multi-scale residual analysis results, a fusion verification result is established.

5. The high-precision navigation and positioning method that fuses multi-source sensing data as described in claim 1, characterized in that, The compensated pose is generated by weighted fusion of vehicle dynamics constraints, road geometry constraints, and traffic flow constraints, including: Establish vehicle dynamics constraints, which include yaw rate constraints, longitudinal acceleration constraints, and tire slip constraints on the weighted fusion results; Establish road geometric constraints, and perform geometric projection matching by weighted fusion of lane lines, curvature and road boundaries to complete road geometric constraint compensation; Traffic flow constraints are compensated by weighted fusion results based on traffic flow velocity distribution and velocity consistency identification.

6. The high-precision navigation and positioning method that fuses multi-source sensing data as described in claim 1, characterized in that, Establish a multi-source dataset, including: After establishing the multi-source dataset, perform data redundancy monitoring and establish a redundant dataset; The redundant dataset is used for multi-source dataset verification, and the redundant data is compressed and stored.

7. The high-precision navigation and positioning method that fuses multi-source sensing data as described in claim 1, characterized in that, Based on the fusion verification results, pose navigation and localization results are generated, including: Perform anomaly identification of the acquisition status of multi-source sensing collectors and establish sensing trust level; After the perceived trust level is used to compensate for the trust in the fusion verification result, the pose navigation and positioning result is output.

8. A high-precision navigation and positioning system that integrates multi-source sensing data, characterized in that, The system is used to execute the high-precision navigation and positioning method fusion of multi-source sensing data as described in any one of claims 1 to 7, the system comprising: The data acquisition and pose mapping module is used to activate the vehicle's multi-source perception acquisition device to perform data acquisition, establish a multi-source dataset, and map the multi-source dataset to initial pose information. The multi-source dataset includes GNSS information, inertial measurement data, wheel speedometer data, visual images, lidar data, and vehicle network data. The scenario-based weight factor establishment module is used to establish scenario-based weight factors by using the scenario perception model to perform scenario trust perception of the multi-source dataset. The scenario perception model includes a road type analysis layer, a traffic environment perception layer, and a vehicle motion state evaluation layer. The weighted fusion and constraint compensation module is used to perform weighted fusion of the initial pose information based on the scenario-based weighting factors, and then compensate the weighted fusion result through vehicle dynamics constraints, road geometry constraints, and traffic flow constraints to generate a compensated pose. The fusion verification and positioning result generation module is used to perform local-global fusion verification on the compensated pose and generate pose navigation and positioning results based on the fusion verification results.

Images