Automatic driving positioning method and system and vehicle

Abstract

The invention provides an automatic driving positioning method and system and a vehicle, and relates to the technical field of automatic driving, and the method comprises the steps: obtaining the initial positioning information of the vehicle, and determining the initial position of the vehicle in a pre-constructed high-precision map based on the initial positioning information; based on a plurality of sensors carried on a vehicle, fusing data acquired by the plurality of sensors to obtain fusion information; and matching the fusion information with a high-precision map to determine the precise position of the vehicle. The vehicle positioning accuracy can be improved.

Description

Technical Field

[0001] The present disclosure relates to the field of autonomous driving technology, and particularly to an autonomous driving positioning method, system and vehicle. Background Art

[0002] Performing high-precision positioning on a vehicle is an important prerequisite for realizing autonomous driving technology, and can provide important guidance for the vehicle's perception, decision-making, and control. From the map level, a high-definition map (HD Map) has higher information content and accuracy. Compared with traditional navigation maps, a high-definition map not only includes road information, but also detailed information such as lane lines, traffic signs, signal lights, and roadside facilities, and can achieve centimeter-level accuracy. It is one of the key technologies for realizing high-level autonomous driving and can help the vehicle effectively understand the surrounding environment. From the sensor level, sensors such as the global navigation satellite system GNSS, inertial unit IMU, vision camera, and lidar can obtain the vehicle's position and real-time data around the vehicle, so as to cooperate with the high-definition map to obtain the vehicle's high-precision positioning information.

[0003] However, in the related art, for the utilization of multi-sensor data, a fixed weight is usually used for sensor data fusion, which is difficult to adapt to the dynamic environment during vehicle operation, thus affecting the positioning accuracy. Summary of the Invention

[0004] To solve the problems existing in the prior art, embodiments of the present disclosure provide an autonomous driving positioning method, system and vehicle. The technical solutions are as follows:

[0005] In a first aspect, an autonomous driving positioning method is provided, including:

[0006] Obtaining the initial positioning information of the vehicle, and determining the initial position of the vehicle in a pre-constructed high-definition map based on the initial positioning information;

[0007] Based on a variety of sensors mounted on the vehicle, fusing the data collected by the variety of sensors to obtain fusion information;

[0008] Matching the fusion information with the high-definition map to determine the precise position of the vehicle;

[0009] Wherein, the data of the variety of sensors is fused by using Kalman filtering, and the measurement noise covariance matrix in Kalman filtering is dynamically adjusted by the confidence levels determined by each sensor according to the collected data.

[0010] In a second aspect, an autonomous driving positioning system is provided, including:

[0011] An initial positioning information determination module, configured to obtain the initial positioning information of a vehicle and determine the initial position of the vehicle in a pre-constructed high-precision map based on the initial positioning information;

[0012] A multi-sensor data fusion module, configured to fuse the data collected by a variety of sensors mounted on the vehicle to obtain fusion information; wherein, the Kalman filter is used to fuse the data of the variety of sensors, and the measurement noise covariance matrix in the Kalman filter is dynamically adjusted by the confidence levels determined by each sensor according to the collected data;

[0013] A high-precision map matching module, configured to match the fusion information with the high-precision map to determine the precise position of the vehicle.

[0014] In a third aspect, an electronic device is provided, including a memory, a processor, and a computer program stored on the memory, and the processor executes the computer program to complete the steps of the above-mentioned autonomous driving positioning method.

[0015] In a fourth aspect, a computer-readable storage medium is provided for storing computer instructions, and when the computer instructions are executed by a processor, the steps of the above-mentioned autonomous driving positioning method are completed.

[0016] In a fifth aspect, a computer program product is provided, including a computer program / instructions, and when the computer program / instructions are executed by a processor, the steps of the above-mentioned autonomous driving positioning method are implemented.

[0017] In a sixth aspect, a vehicle is provided, and the vehicle uses the above-mentioned autonomous driving positioning method for vehicle positioning, or the vehicle includes the above-mentioned autonomous driving positioning system, and vehicle positioning is realized based on this system.

[0018] The beneficial effects brought by the technical solutions provided in the embodiments of the present disclosure are as follows: In the embodiments of the present disclosure, aiming at the problem of insufficient adaptability in the traditional multi-sensor fixed-weight fusion, the idea of environment-adaptive weight adjustment is introduced. By collecting data from sensors, the environmental parameters are sensed in real time, and the confidence levels, that is, the reliability, of each sensor are evaluated. Then, the measurement noise covariance matrix of the Kalman filter is dynamically adjusted according to the confidence levels, realizing positioning adaptability in a dynamic environment, improving the positioning accuracy, and at the same time being able to quickly switch the positioning method when a single sensor fails, improving the positioning continuity.

[0019] Some of the advantages of the additional aspects of the present disclosure will be given in the following description, some will become obvious from the following description, or will be understood through the practice of the present disclosure. Description of the Drawings

[0020] To more clearly illustrate the technical solutions in the embodiments of the present disclosure, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present disclosure. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0021] Figure 1 is a flowchart of an autonomous driving positioning method provided by an embodiment of the present disclosure;

[0022] Figure 2 is a structural block diagram of an autonomous driving positioning system provided by an embodiment of the present disclosure;

[0023] Figure 3 is a structural block diagram of an electronic device provided by an embodiment of the present disclosure. Detailed implementation manners

[0024] To make the objectives, technical solutions, and advantages of the present disclosure clearer, the following will further describe the embodiments of the present disclosure in detail with reference to the accompanying drawings.

[0025] Figure 1 is a flowchart of an autonomous driving positioning method provided by an embodiment of the present disclosure. Refer to Figure 1 and the method includes:

[0026] Step 101, obtain the initial positioning information of the vehicle, and determine the initial position of the vehicle in a pre-constructed high-precision map based on the initial positioning information;

[0027] Step 102, based on a variety of sensors mounted on the vehicle, fuse the data collected by the variety of sensors to obtain fused information; among them, the Kalman filter is used to fuse the data of the variety of sensors, and the measurement noise covariance matrix in the Kalman filter is dynamically adjusted according to the confidence levels determined by each sensor based on the collected data;

[0028] Step 103, match the fused information with the high-precision map to determine the accurate position of the vehicle.

[0029] In step 101, GNSS (Global Navigation Satellite System) and IMU (Inertial Measurement Unit) are installed on the vehicle to achieve preliminary positioning. Based on the real-time data of GNSS and IMU, an initial position judgment is made to determine the initial position of the vehicle, that is, low-precision positioning information.

[0030] Before the autonomous driving system is started, a pre-constructed high-precision map is loaded. These maps usually contain environmental data such as detailed road geometries, traffic signs, intersection information, obstacles, etc.

[0031] Initialize the high-precision map. Determine the local search range of the high-precision map based on the low-precision positioning information for subsequent further high-precision positioning.

[0032] In step 102, various sensors for sensing environmental data (such as lidar, cameras, millimeter-wave radars, inertial unit IMU, etc.) mounted on the autonomous vehicle start to work and collect surrounding environmental data in real time. This data includes key information such as distance, speed, acceleration, and direction.

[0033] Preprocess the collected sensor data, including data cleaning, noise elimination, outlier processing, etc., to improve the accuracy and reliability of the data. The inertial navigation solution operator module based on IMU provides the initial position and attitude information of the vehicle. The GNSS positioning sub-module based on the ground base station and vehicle-mounted antenna provides the absolute position information of the vehicle. Other sensors such as lidar collect surrounding environmental data in real time, including distance, speed, acceleration, etc. Fuse these sensor data to obtain a more comprehensive and accurate vehicle position and attitude information.

[0034] To achieve multi-sensor data fusion, the Kalman filter algorithm is used for fusion in this embodiment.

[0035] First, perform system modeling, including the state model and the observation model. The state model describes how the system state changes over time, and the corresponding state equation is expressed as:

[0036] x k = Ax k-1 + Bu k + ω k ;

[0037] Among them, x k represents the system state at time k, here representing the vehicle state (position, speed, attitude, etc.); A is the state transition matrix, obtained from the vehicle motion model constructed based on IMU, B is the control input matrix, u k is the control vector, ω k is the process noise, and the corresponding process noise covariance matrix Q k is dynamically adjusted.

[0038] The observation equation corresponding to the observation model is expressed as:

[0039] z k = Hx k + v k ;

[0040] Among them, z k represents the observation data at time k, here representing the observation data of each sensor (GNSS position, lidar point cloud, camera pose, etc.), vk To measure noise, the corresponding measurement noise covariance matrix is R k , which is dynamically adjusted by the sensor confidence.

[0041] In this embodiment, a dynamic evaluation model of sensor reliability based on multi-modal data (camera rain and fog recognition confidence plus lidar occlusion rate) is proposed, and a non-linear dynamic adjustment formula for the measurement noise covariance matrix is designed to be adaptive according to environmental factors and sensor states. The dynamic adjustment formula of the measurement noise covariance matrix is expressed as:

[0042] R k = α·R base + β·(1 - Confidence sensor ) + γ·Env score ;

[0043] Where α, β, and γ are weight coefficients, R base is the sensor reference noise, Confidence sensor is the sensor confidence, which is updated in real time according to the environmental perception result, and Env score is the environmental complexity score, and the weights are dynamically corrected through real-time perception data (such as dynamic obstacle density), breaking through the limitations of traditional static parameters.

[0044] The calculation of the sensor confidence Confidence sensor needs to combine the sensor's own performance indicators and real-time environmental perception data, and the calculation formula is:

[0045] Confidence sensor = ω 1 ·S intrinsic + ω 2 ·S environment + ω 3 ·S history ;

[0046] Where S intrinsic is the sensor's inherent reliability score (such as calibration accuracy, noise level); S environment is the score of the impact of environmental interference on the sensor (such as rain and fog, occlusion); S history is the stability score statistically from historical data (such as the error variance of the past N frames); ω 1 , ω 2 , ω 3 are dynamic weights, which are adaptively adjusted by the environmental complexity Env score .

[0047] The sensor's inherent reliability score S intrinsicBased on the factory calibration parameters of the sensors (such as the zero-bias stability of the IMU and the angular resolution of the lidar), it is calculated by quantifying through a predefined scoring table:

[0048]

[0049] The influence score S of environmental interference on the sensor environment , including the camera interference score The lidar interference score and the GNSS interference score

[0050] The camera interference score According to the output probability (rain and fog probability) of the rain and fog recognition model

[0051] P rain / fog and the visibility estimate value V visibility Calculate:

[0052]

[0053] The lidar interference score Is dynamically calculated according to the point cloud occlusion rate O occlusion (The number of occluded points / the total number of points) to suppress the influence of a high occlusion rate:

[0054]

[0055] The GNSS interference score Based on the signal strength S of the GNSS signal and the multipath error E multipath Calculate to get:

[0056]

[0057] Among them, S max Represents the maximum signal strength, and E threshold Represents the allowable threshold of the multipath error.

[0058] The stability score S statistically calculated from historical data history Is calculated according to the standard deviation of the historical positioning results. Using the sliding window statistical method, calculate the standard deviation σ of the positioning results of the past N frames and map it to the confidence interval, and use exponential decay to suppress burst noise:

[0059]

[0060] Among them, σ threshold Represents the allowable threshold of the standard deviation.

[0061] The dynamic weight ω 1 、ω 2 、ω3 , adjusted adaptively through the environmental complexity Env score , and the environmental complexity is calculated by real-time sensing data:

[0062] Env score = α·D obstacle + β·I weather + γ·C dynamic ;

[0063] Among them, α, β, and γ are weight coefficients, and D obstacle is the density of dynamic obstacles (the number of moving targets per unit area), and the calculation formula is as follows:

[0064]

[0065] For example, for an urban crossroads, D obstacle = 0.8, and for a rural road, D obstacle = 0.2;

[0066] I weather is the intensity of weather interference (such as rainfall level, fog concentration), and the calculation formula is as follows:

[0067] I weather = α I ·P rain + β I ·C fog + γ I ·V visibility ;

[0068] Among them, P rain represents the rainfall probability (normalized to 0 - 1), C fog represents the fog concentration (unit: g / m 3 ), V visibility represents the visibility measured by the camera (unit: m), and α I , β I , γ I are weight coefficients, calibrated according to the characteristics of the sensor;

[0069] C dynamic is the scene change rate, calculated by the difference between adjacent frame point clouds, and the calculation formula is as follows:

[0070]

[0071] Among them, ‖PointCloud t - PointCloud t-1 ‖ 2Represents the Euclidean distance between the current frame point cloud and the previous frame point cloud. The scene change rate is calculated by the average Euclidean distance of adjacent frame lidar point clouds, which can reflect the change speed of the road structure or obstacles.

[0072] The environmental complexity Env is calculated. score After that, according to Env score value, the dynamic weights ω sensor in the sensor confidence Confidence 1 、ω 2 、ω 3 are adjusted. A high environmental complexity threshold θ high and a low environmental complexity threshold θ low are preset. θ high and θ low are obtained through training and optimization with a large amount of measured data (such as different cities, weather). According to the relationship between Env score and θ high and θ low the dynamic weights are modified.

[0073] If Env score ≥θ high , it is judged that it is in a high complexity environment at this time, and strategies such as dynamic weight allocation and tight coupling fault tolerance are triggered, that is, ω 2 (environmental interference weight) is increased; a highly complex environment means an environment with dense dynamic obstacles, bad weather, and frequent scene changes.

[0074] In some embodiments, the weight calculation formula in a high complexity environment is as follows: The weight allocation is dominated by ω 2 , while suppressing ω 1 and ω 3 :

[0075]

[0076] If Env score ≤θ low , it is judged that it is in a stable environment at this time, and the default sensor fusion mode is adopted, giving priority to relying on GNSS or high-precision maps, that is, giving priority to ω 1 and ω 3 (inherent performance and historical stability); a stable environment means an environment with sparse obstacles, good weather, and static scenes.

[0077] In some embodiments, the weight calculation formula in a stable environment is as follows: The weight allocation is dominated by ω 1 and ω 3 , and ω 2 is reduced to the lowest:

[0078]

[0079] If θ low <Env score <θ high , it is determined that it is in a transitional environment at this time, and a progressive strategy adjustment is enabled. In some embodiments, the weight calculation formula in the transitional environment is as follows: The weight is smoothly transitioned through bilinear interpolation to avoid mutations:

[0080]

[0081] The following uses three example scenarios to illustrate the calculation of Env score . For ease of description, the calculation is carried out with α = 0.5, β = 0.3, and γ = 0.2 as an example, and the calibrated θ high = 0.5, θ low = 0.3.

[0082] (1) Urban tunnel scenario:

[0083] D obstacle = 0.6 (heavy traffic), I weather = 0 (no rain or fog), C dynamic = 0.4 (slight point cloud jitter);

[0084] Then Env score = 0.5·0.6 + 0.3·0 + 0.2·0.4 = 0.38;

[0085] It is determined that it is in a stable environment at this time, but the loss of GNSS signals needs to be monitored.

[0086] (2) Rainstorm construction area:

[0087] D obstacle = 0.3 (cones are dense), I weather = 0.9 (rainstorm), C dynamic = 0.8 (road structure change);

[0088] Then Env score = 0.5·0.3 + 0.3·0.9 + 0.2·0.8 = 0.58;

[0089] It is determined that it is in a transitional environment at this time, and a partial fault tolerance mechanism is enabled.

[0090] (3) Dense fog on the highway:

[0091] D obstacle = 0.1 (sparse traffic), I weather = 1.0 (visibility < 50 meters), C dynamic = 0.2 (slight point cloud jitter);

[0092] Then Envscore = 0.5·0.1 + 0.3·1.0 + 0.2·0.2 = 0.39;

[0093] It is determined that it is in a stable environment at this time, but positioning depends on lidar and high-precision maps.

[0094] Through the above solution, the sensor confidence not only quantifies its own performance, but also deeply integrates environmental and historical factors, providing highly robust inputs for dynamic covariance adjustment, significantly superior to traditional static or single-dimensional confidence calculation solutions. In the dynamic adjustment calculation of the measurement noise covariance matrix R k , through non-linear mappings such as tanh and exponential functions, the sensitivity to environmental mutations of traditional linear weighting is avoided; by combining physical sensor data (such as point cloud occlusion rate) and semantic information (such as weather recognition), the comprehensiveness of evaluation is improved; based on the environmental complexity, the scoring weights are dynamically adjusted to achieve "sensor-environment" collaborative optimization.

[0095] In some embodiments, when GNSS fails, a tight coupling mode of IMU and lidar is adopted, combined with sequential sliding window optimization (SWO) to compensate for cumulative errors in real time, and through NDT algorithm and high-precision map matching, the error control is reduced from 30 cm of traditional methods to 10 cm.

[0096] By adopting the above solution, the positioning accuracy in rainy and foggy weather is improved to 0.2 m (0.5 m for traditional methods), and the continuous positioning time in tunnel scenarios reaches 8 minutes (only 1 minute for traditional methods).

[0097] In step 103, the fused sensor data and the pre-made high-precision map data are transformed into the same coordinate system for matching. During the matching process, common feature map matching positioning algorithms include ICP (Iterative Closest Point) algorithm, NDT (Normal Distribution Transformation) algorithm, etc. These algorithms find the most matching map position by comparing the differences between sensor data and map data. After successful matching, the vehicle positioning information can be confirmed.

[0098] In addition, according to the results of high-precision map matching, the vehicle position is continuously optimized and corrected to ensure positioning accuracy and stability. The fusion results are fed back to the GNSS positioning and point cloud positioning sub-modules to improve the accuracy of the two positioning modules.

[0099] In some embodiments, considering that traditional ICP or NDT algorithms rely on geometric features, have a high false matching rate (error up to 50 cm) under the interference of dynamic obstacles, it is difficult to align the features of cross-modal data (point cloud, image), and the matching efficiency is low. A cross-modal attention module (CM-ATT) is further designed for matching. The lidar point cloud features (extracted by PointNet++) and the image semantic features (extracted by ResNet-50) are dynamically fused through attention weights, and the loss function is further improved to:

[0100]

[0101] where p i and i j are the matched point cloud-image pairs, ⊕ represents feature concatenation, f(p i ) represents the extracted lidar point cloud features, f(i j ) represents the extracted image semantic features, ‖·‖ 2 represents calculating the Euclidean distance, λ is the weight coefficient, KL represents calculating the KL divergence, and the KL divergence term is used to enhance the distinguishability of static elements. α static represents static elements, and α dynamic represents dynamic elements.

[0102] Introduce spatio-temporal consistency verification, combine the attention mechanism of Transformer and Kalman prediction, dynamically suppress the influence of short-term moving obstacles (such as pedestrians) on matching, and retain long-term static elements.

[0103] After the above improvements, the dynamic scene matching error is reduced to 10 cm, and the calculation time is reduced by 40% (compared with the traditional ICP algorithm, its dynamic scene matching error is 50 cm, and the algorithm takes about 100 ms).

[0104] The accuracy of the high-precision map itself has a direct impact on the vehicle positioning accuracy. Currently, the high-precision map relies on offline updates (quarterly), and cannot reflect dynamic changes such as temporary construction and obstacles, resulting in positioning deviations.

[0105] To overcome the above problems, in this embodiment, based on the collection of environmental data by vehicles on the road, the high-precision map is updated in real time. As the vehicle travels, new environmental data is continuously collected through sensors and the high-precision map is updated in real time. Map updates include updates of road changes, new buildings, traffic signs, etc., to ensure the accuracy and real-time nature of the map.

[0106] The update of the high-precision map includes local map correction and map-sensor two-way calibration.

[0107] Local Map Correction: By analyzing the ICP residuals between lidar point clouds and high-precision maps, detect areas with road changes (such as newly added cones, etc.), and use a sliding window optimization (BA) to update the local map.

[0108] Map-Sensor Bidirectional Calibration: If multiple vehicles detect position deviations of map signs, trigger the correction process:

[0109]

[0110] where N is the number of cooperative vehicles, x sensor，i represents the position of the map sign detected by the i-th vehicle, and x map represents the position of the map sign in the high-precision map. The correction amount Δx is fed back to the map database in real time to correct the high-precision map.

[0111] In addition, in the multi-vehicle cooperation mode, there are often data conflicts in crowdsourcing updates, and it is difficult to ensure global consistency. Based on this, the Raft protocol is used to elect a master node (such as preferentially selecting vehicles with high confidence) to merge local update data to avoid the bottleneck of the centralized server.

[0112] Further design a bidirectional residual feedback mechanism: The map correction amount Δx not only updates the database, but also reversely optimizes the sensor calibration parameters to achieve closed-loop calibration:

[0113]

[0114] where represents the updated map data, represents the map data before update, represents the optimized sensor calibration parameters, represents the sensor calibration parameters before optimization, and η and κ are adaptive learning rates, which are dynamically adjusted according to the historical error gradient.

[0115] Based on the real-time update of the high-precision map with multi-vehicle cooperation, the map update delay is reduced from the hour level to the second level (it can be shortened to 200 ms), and the positioning success rate in the temporary construction scenario is increased from 70% to 98%.

[0116] In some embodiments, during the vehicle positioning process, if a single positioning mode (such as GNSS, SLAM) is used in different scenarios, it is difficult to adapt to variable scenarios such as cities, tunnels, and rural areas, the cumulative error is significant, and the multi-coordinate system conversion error will cause positioning drift. Based on this, on the basis of the foregoing positioning scheme, further optimize the parameters in specific scenarios for different scenarios, determine the positioning model for different scenarios, and jointly determine the scenario mode through multi-dimensional data such as lidar point cloud density, GNSS signal strength, and camera semantic tags to match the best positioning model for the current scenario.

[0117] Furthermore, during the scene switching process, to reduce the coordinate system conversion error and achieve seamless scene switching, Lie group theory is adopted for error compensation optimization, and a Lie algebra incremental optimizer is designed. By calculating the pose residuals of adjacent frames in real time, the Lie algebra parameter ξ is dynamically corrected to eliminate the cumulative error:

[0118]

[0119] where ε is the pose residual function and μ is the adaptive step size.

[0120] Through the above scheme, the positioning continuity of the tunnel scene reaches 99.5%, and the cumulative error of coordinate system conversion is reduced to 0.05 m / km (0.5 m / km for the traditional method).

[0121] The entire positioning system is monitored in real time to ensure the normal operation of each sensor and algorithm. The accuracy and stability of the positioning system are evaluated regularly to detect and solve problems in a timely manner, achieving high-precision, real-time, and stable vehicle positioning, and providing strong support for the safe and efficient operation of the autonomous driving system.

[0122] The method provided in this embodiment can comprehensively utilize the advantages of different sensors through the fusion of multi-sensor data, improving the positioning accuracy and stability. Whether it is sensors such as lidar, cameras, or IMUs, they each have different characteristics and advantages. Through fusion processing, the deficiencies of a single sensor can be compensated, and a more accurate and reliable positioning result can be obtained.

[0123] Using real-time sensor data for positioning can reflect the position and attitude information of the vehicle in real time. At the same time, through data fusion and map matching technologies, the positioning calculation can be quickly completed, achieving a rapid response to the vehicle's position. This is crucial for the autonomous driving system and can ensure that the vehicle maintains a stable driving state in various complex environments.

[0124] The autonomous driving positioning method based on high-precision maps and multi-sensor fusion has strong adaptability. Whether it is different road environments such as urban roads, highways, or rural roads, this method can achieve accurate positioning by real-time updating and matching high-precision maps. At the same time, this method can also adapt to different weather and lighting conditions, ensuring stable positioning performance in various complex environments.

[0125] Accurate positioning information is the basis for the safe driving of an autonomous driving system. The high-precision positioning achieved through this method can provide reliable navigation and decision-making support for the autonomous driving system, reducing the risk of traffic accidents. For example, in complex scenarios such as intersections and crosswalks, this method can help the vehicle accurately perceive the surrounding environment, make correct driving decisions, and avoid collisions with pedestrians, other vehicles and other obstacles.

[0126] Traditional positioning methods usually need to rely on expensive sensor devices to achieve high-precision positioning. The autonomous driving positioning method based on high-precision maps and multi-sensor fusion can achieve high-precision positioning by fusing data from different sensors, reducing the dependence on a single high-precision sensor, and thus reducing the system cost. At the same time, this method can also utilize existing high-precision map resources to further reduce the system cost.

[0127] Figure 2 is a structural block diagram of an autonomous driving positioning system provided by an embodiment of the present disclosure, as Figure 2 shown, the system includes: an initial positioning information determination module, a multi-sensor data fusion module, and a high-precision map matching module.

[0128] Among them, the initial positioning information determination module is configured to obtain the initial positioning information of the vehicle and determine the initial position of the vehicle in a pre-constructed high-precision map based on the initial positioning information;

[0129] The multi-sensor data fusion module is configured to fuse the data collected by a variety of sensors based on a variety of sensors mounted on the vehicle to obtain fusion information; among them, the Kalman filter is used to fuse the data of a variety of sensors, and the measurement noise covariance matrix in the Kalman filter is dynamically adjusted by the confidence levels determined by each sensor according to the collected data;

[0130] The high-precision map matching module is configured to match the fusion information with the high-precision map to determine the precise position of the vehicle.

[0131] It should be noted that: for the autonomous driving positioning system provided in the above embodiment, only the above-mentioned division of each functional module is used for illustration. In practical applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the autonomous driving positioning system provided in the above embodiment and the embodiment of the autonomous driving positioning method belong to the same concept. For the specific implementation process, please refer to the method embodiment, which will not be elaborated here.

[0132] Embodiments of the present disclosure also provide a vehicle that uses the above-mentioned autonomous driving positioning method for vehicle positioning, or the vehicle includes the above-mentioned autonomous driving positioning system, and vehicle positioning is achieved based on this system. The vehicle can be a fuel vehicle driven by an internal combustion engine, an electric vehicle driven by electricity, a new energy vehicle driven by other energy forms, or a hybrid vehicle powered by multiple energy forms.

[0133] Figure 3 is a structural block diagram of an electronic device provided by an embodiment of the present disclosure. As Figure 3 shown, the electronic device can be a computer. The electronic device includes: a processor and a memory.

[0134] The processor may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor can be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), or PLA (Programmable Logic Array). The processor may also include a main processor and a coprocessor. The main processor is a processor used to process data in the wake state, also known as the CPU (Central Processing Unit); the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, the processor may be integrated with a GPU (Graphics Processing Unit), and the GPU is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor may also include an AI (Artificial Intelligence) processor, and the AI processor is used to process computational operations related to machine learning.

[0135] The memory may include one or more computer-readable media, and the computer-readable media may be non-transitory. The memory may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices and flash storage devices. In some embodiments, the non-transitory computer-readable media in the memory are used to store at least one computer program, and the at least one computer program is used to be executed by the processor to implement an autonomous driving positioning method provided by an embodiment of the present disclosure.

[0136] Those skilled in the art can understand that Figure 3 the structure shown in does not constitute a limitation on the electronic device, and it may include more or fewer components than shown in the figure, or combine certain components, or adopt different component arrangements.

[0137] An embodiment of the present disclosure also provides a computer-readable storage medium for storing computer instructions, which can complete the steps of an automatic driving positioning method provided in the embodiment of the present disclosure when the computer instructions are executed by a processor.

[0138] An embodiment of the present disclosure also provides a computer program product, including computer programs / instructions, which implement the steps of an automatic driving positioning method provided in the embodiment of the present disclosure when the computer programs / instructions are executed by a processor.

[0139] The above are only the preferred embodiments of the present disclosure and are not used to limit the present disclosure. For those skilled in the art, the present disclosure can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. An automatic driving positioning method, characterized in that: include: Acquire initial positioning information of the vehicle, and determine the initial position of the vehicle in a pre-constructed high-precision map based on the initial positioning information; Based on multiple sensors mounted on the vehicle, the data collected by multiple sensors are fused to obtain fusion information; Match the fused information with the high-precision map to determine the exact location of the vehicle; Among them, Kalman filtering is used to fuse the data of multiple sensors, and the measurement noise covariance matrix in the Kalman filtering is dynamically adjusted by the confidence level determined by each sensor according to the collected data.

2. The automatic driving positioning method according to claim 1, characterized in that: The dynamic adjustment formula of the measurement noise covariance matrix is ​​expressed as: R k =α·R base +β·(1-Confidence sensor ) Among them, α and β are the weights of environmental factors, Confidence sensor It is the sensor confidence, which is updated in real time according to the environmental perception results.

3. The automatic driving positioning method according to claim 2, characterized in that: Sensor confidence includes GNSS confidence, camera confidence, and lidar confidence. GNSS confidence is determined based on GNSS positioning accuracy, camera confidence is determined based on weather detection results, and lidar confidence is determined based on occlusion rate.

4. The automatic driving positioning method according to claim 3, characterized in that: The camera confidence is calculated based on the weather detection result and the normalized clarity value.

5. The automatic driving positioning method according to claim 1, characterized in that: The high-precision map is updated in real time based on the collection of environmental data by vehicles on the road.

6. An automatic driving positioning system, characterized in that: include: An initial positioning information determination module is configured to obtain initial positioning information of the vehicle and determine an initial position of the vehicle in a pre-constructed high-precision map based on the initial positioning information; The multi-sensor data fusion module is configured to fuse the data collected by the multiple sensors mounted on the vehicle to obtain fusion information; wherein the data of the multiple sensors are fused by using Kalman filtering, and the measurement noise covariance matrix in the Kalman filtering is dynamically adjusted by the confidence determined by each sensor according to the collected data; The high-precision map matching module is configured to match the fused information with the high-precision map to determine the precise location of the vehicle.

7. An electronic device, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to complete the steps of the method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the steps of the method according to any one of claims 1 to 5.

9. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

10. A vehicle, characterized in that: The vehicle adopts the autonomous driving positioning method as described in any one of claims 1-5 to perform vehicle positioning, or the vehicle includes the autonomous driving positioning system as described in claim 6, and realizes vehicle positioning based on the system.

Images