Position precision evaluation method and device, equipment, vehicle, storage medium and product

Abstract

The invention relates to a position precision evaluation method, device and equipment, a vehicle, a storage medium and a product. The method comprises the following steps: acquiring track data and visual data acquired in a driving process of the vehicle; generating vector data corresponding to the driving environment of the vehicle based on the trajectory data and the visual data; the vector data comprises at least one vector element; and based on the trajectory data, the visual data and the vector data, evaluating the position precision of the vector elements to obtain a position precision evaluation result. The internal evaluation of the position precision is realized according to the characteristics of the trajectory data, the visual data and the vector data, the evaluation of the position precision can be realized without the help of a high-precision map, and the accuracy and efficiency of the evaluation of the position precision are improved.

Description

Technical Field

[0001] The present application relates to the field of intelligent transportation information technology, and in particular, to a method, device, equipment, vehicle, storage medium and product for evaluating position accuracy. Background Art

[0002] In recent years, the development of in-vehicle intelligent systems and cloud computing technologies has enabled the rapid development of autonomous driving technology from the driving assistance stage to the driverless stage. Currently, autonomous driving technology mainly relies on high-precision maps to achieve long-distance perception and lane-level planning and guidance. However, the production of high-precision maps currently mainly relies on on-site measurement methods. Due to factors such as high labor costs, high equipment prices, and harsh working conditions, conducting on-site measurements on a larger scale is not the optimal solution. Therefore, crowd-sourced mapping is an alternative to on-site measurement methods. Compared with high-precision and high-cost professional collection methods, the mapping technology of crowd-sourced collection methods has become an ideal way for large-scale and low-cost mapping. With the development of artificial intelligence algorithms and sensors, crowd-sourced mapping can achieve a high degree of automation in the mapping process. However, due to the diverse sources and uneven quality of crowd-sourced data and the limitations of existing mapping methods, the quality problems of maps generated by crowd-sourced methods objectively exist. The crowd-sourced maps output by crowd-sourced mapping need to be strictly verified before they can be released and used. Therefore, a reasonable and effective method for evaluating the quality of crowd-sourced mapping is an important requirement in the field of autonomous driving. Summary of the Invention

[0003] The present application provides a method, device, equipment, vehicle, storage medium and product for evaluating position accuracy.

[0004] In a first aspect, the present application provides a method for evaluating position accuracy, the method comprising: Obtaining trajectory data and visual data collected by a vehicle during driving; Generating vector data corresponding to the driving environment of the vehicle based on the trajectory data and the visual data; the vector data includes at least one vector element; Evaluating the position accuracy of the vector element based on the trajectory data, the visual data, and the vector data to obtain a position accuracy evaluation result.

[0005] According to the above technical means, the present application realizes the internal evaluation of position accuracy according to the self-characteristics of the trajectory data, the visual data, and the vector data, and can evaluate the position accuracy without relying on a high-precision map, thereby improving the accuracy and efficiency of position accuracy evaluation.

[0006] Further, based on the trajectory data, visual data, and vector data, the position accuracy of vector elements is evaluated to obtain a position accuracy evaluation result, including: extracting features from the vector data, trajectory data, and visual data to obtain multiple feature information; performing feature mapping on the multiple feature information to predict the position accuracy corresponding to the vector elements and obtain the position accuracy evaluation result.

[0007] According to the above technical means, the position accuracy evaluation result of the position accuracy prediction of vector elements in this application can provide important reference information for fields such as autonomous driving and intelligent transportation systems, helping to improve the safety and reliability of the system.

[0008] Further, the multiple feature information includes trajectory feature information, positioning feature information, visual feature information, and geometric and semantic feature information of vector elements; extracting features from the vector data, trajectory data, and visual data to obtain multiple feature information, including: extracting geometric and semantic feature information from the vector data and trajectory data; extracting trajectory feature information and positioning feature information from the trajectory data; extracting visual feature information from the visual data.

[0009] According to the above technical means, this application extracts multiple feature information from the visual data, trajectory data, and vector data, and then maps the vector elements' position accuracy based on the multiple feature information, with relatively high accuracy.

[0010] Further, the semantic feature information includes the field of view distance; the field of view distance is the distance between the center point of the vector element and the vehicle's driving trajectory.

[0011] According to the above technical means, this application derives the field of view distance based on the vector data and trajectory data. Based on the field of view distance, the accuracy in the visual data can be better corrected, and further the accuracy of the position accuracy prediction based on the visual data can be improved.

[0012] Further, the position accuracy evaluation result includes the accuracy evaluation result of the vector element and the position deviation; the method further includes: in response to the accuracy evaluation result of the vector element being inaccurate, based on the position deviation of the vector element, performing position adjustment on the vector element to obtain adjusted vector data; based on the adjusted vector data, constructing a map corresponding to the driving environment.

[0013] According to the above technical means, since the position of the vector element in the adjusted vector data is adjusted, the accuracy is relatively high, and thus the accuracy of the scene map constructed based on the adjusted vector data is also relatively high.

[0014] Further, before performing feature mapping on multiple pieces of feature information to predict the position accuracy corresponding to the vector element and obtaining the position accuracy evaluation result, the method further includes: identifying the scene of the driving environment based on the trajectory feature information among the multiple pieces of feature information to obtain the scene category; and performing feature mapping on the multiple pieces of feature information based on the scene category to predict the position accuracy corresponding to the vector element and obtaining the position accuracy evaluation result.

[0015] According to the above technical means, the position accuracy of the vector element based on the scene category in the present application is more intelligent and the prediction accuracy is also relatively high.

[0016] Further, the trajectory feature information includes at least one trajectory point, and each trajectory point corresponds to a number of satellites in view and a positioning status; identifying the scene of the driving environment based on the trajectory feature information among the multiple pieces of feature information to obtain the scene category includes: determining the average number of satellites in view of at least one number of satellites in view, and statistically analyzing the discrete distribution of at least one positioning status; in response to the average number of satellites in view being greater than the first threshold and the discrete distribution of at least one positioning status being within the first preset distribution set, determining that the driving environment is in an open scene; and in response to the average number of satellites in view being less than the second threshold and the ratio of the discrete distribution of at least one positioning status being within the second preset distribution set being greater than the preset ratio, determining that the driving environment is in a satellite rejection scene.

[0017] According to the above technical means, the present application distinguishes the driving scene of the current vehicle based on the number of satellites in view and the positioning status in the trajectory feature information, and then predicts the position accuracy of the vector element in different scenes, with relatively high accuracy.

[0018] Further, evaluating the position accuracy of the vector element based on the trajectory data, visual data, and vector data to obtain the position accuracy evaluation result includes: using a preset position accuracy evaluation model to evaluate the position accuracy of the vector element based on the trajectory data, visual data, and vector data to obtain the position accuracy evaluation result.

[0019] According to the above technical means, based on the trained preset position accuracy evaluation model, the present application can directly evaluate the position accuracy of the vector element based on the trajectory data, visual data, and vector data, and can achieve the evaluation of the position accuracy without relying on the ground truth elements in the high-precision map, thereby improving the efficiency of the position accuracy evaluation.

[0020] Further, the preset position accuracy evaluation model has corresponding model parameters for different scenarios: based on trajectory data, visual data, and vector data, the position accuracy of vector elements is evaluated to obtain a position accuracy evaluation result, including: determining the scenario category of the vehicle's driving environment; using the preset position accuracy evaluation model, calling the model parameters matching the scenario category, and based on the trajectory data, visual data, and vector data, evaluating the position accuracy of vector elements to obtain a position accuracy evaluation result.

[0021] According to the above technical means, the model parameters of the model for evaluating the position accuracy of vector elements in this application vary according to different scenario categories, can better adapt to the changes in scenarios, and the accuracy of position accuracy evaluation is relatively high.

[0022] Further, before using the preset position accuracy evaluation model to evaluate the position accuracy of vector elements based on trajectory data, visual data, and vector data to obtain a position accuracy evaluation result, the method further includes: obtaining sample data; the sample data includes trajectory sample data, visual sample data, and vector sample data matching the trajectory sample data and visual sample data; using the position accuracy evaluation model to be trained, based on the sample data, predicting the sample position accuracy of the sample vector elements to obtain a sample position prediction result; based on the sample prediction result and the target prediction result corresponding to the sample data, adjusting the model parameters of the position accuracy evaluation model to be trained to obtain the preset position accuracy evaluation model.

[0023] According to the above technical means, the preset neural network model in this application is obtained by batch training the neural network using sample data, and the model output result is the position accuracy evaluation result, which can be directly used later, improving the efficiency of position accuracy evaluation.

[0024] In a second aspect, this application provides a position accuracy evaluation device, which includes: An acquisition module, configured to acquire trajectory data and visual data collected during the vehicle's driving process; A generation module, configured to generate vector data corresponding to the vehicle's driving environment based on the trajectory data and visual data; the vector data contains at least one vector element; An evaluation module, configured to evaluate the position accuracy of vector elements based on the trajectory data, visual data, and vector data to obtain a position accuracy evaluation result.

[0025] In a third aspect, this application provides an electronic device, which includes: a processor, a memory, and a communication bus; when the processor executes the running program stored in the memory, the above position accuracy evaluation method is implemented.

[0026] Fourthly, the present application provides a vehicle, including the above-mentioned electronic device.

[0027] Fifthly, the present application provides a computer-readable storage medium, characterized in that a computer program is stored thereon, and when the computer program is executed by a processor, the above-mentioned position accuracy evaluation method is implemented.

[0028] Sixthly, the present application provides a computer program product, including a computer program or instruction, characterized in that when the computer program or instruction is executed by a processor, the above-mentioned position accuracy evaluation method is implemented.

[0029] Advantages of the present application: (1) The present application realizes the internal evaluation of position accuracy according to the characteristics of trajectory data, visual data, and vector data. Without relying on a high-precision map, the evaluation of position accuracy can be realized, improving the accuracy and efficiency of position accuracy evaluation.

[0030] (2) The position accuracy evaluation result of the position accuracy prediction of vector elements in the present application can provide important reference information for fields such as autonomous driving and intelligent transportation systems, helping to improve the safety and reliability of the system.

[0031] (3) The present application extracts features from visual data, trajectory data, and vector data to obtain multiple feature information, and then maps the vector elements' position accuracy based on the multiple feature information, with relatively high accuracy.

[0032] (4) Since the positions of vector elements in the adjusted vector data in the present application are adjusted, with relatively high accuracy, the scene map constructed based on the adjusted vector data also has relatively high accuracy.

[0033] (5) The position accuracy of vector elements based on scene categories in the present application is more intelligent and the prediction accuracy is also relatively high.

[0034] (6) Based on the trained preset position accuracy evaluation model, the present application can directly evaluate the position accuracy of vector elements based on trajectory data, visual data, and vector data. Without relying on the ground truth elements in the high-precision map, the evaluation of position accuracy can be realized, improving the efficiency of position accuracy evaluation.

[0035] (7) The model parameters of the model for evaluating the position accuracy of vector elements in the present application vary according to different scene categories, can better adapt to scene changes, and the accuracy of position accuracy evaluation is relatively high. Description of the Drawings

[0036] Figure 1 It is a schematic flowchart of a position accuracy evaluation method provided by an embodiment of the present application; Figure 2 A schematic flowchart of an exemplary method for obtaining a position accuracy evaluation result provided by an embodiment of the present application Figure 1 ; Figure 3 A schematic flowchart of an exemplary method for extracting feature information provided by an embodiment of the present application; Figure 4 A schematic diagram of an exemplary position accuracy evaluation result provided by an embodiment of the present application; Figure 5 A schematic flowchart of an exemplary method for constructing an environmental map provided by an embodiment of the present application Figure 1 ; Figure 6 A schematic flowchart of an exemplary method for obtaining a position accuracy evaluation result provided by an embodiment of the present application Figure 2 ; Figure 7 A schematic flowchart of an exemplary method for scene recognition provided by an embodiment of the present application; Figure 8 A schematic flowchart of an exemplary method for obtaining a position accuracy evaluation result provided by an embodiment of the present application Figure 3 ; Figure 9 A schematic flowchart of an exemplary model training method provided by an embodiment of the present application Figure 1 ; Figure 10 A schematic flowchart of an exemplary model training method provided by an embodiment of the present application Figure 2 ; Figure 11 A schematic flowchart of an exemplary method for constructing an environmental map provided by an embodiment of the present application Figure 2 ; Figure 12 A schematic flowchart of an exemplary position accuracy evaluation method provided by an embodiment of the present application; Figure 13 A schematic diagram of the structure of a position accuracy evaluation device provided by an embodiment of the present application; Figure 14 A schematic diagram of the structure of an electronic device provided by an embodiment of the present application. Detailed implementation manners

[0037] The embodiments of the present application will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present application from the content disclosed in this specification. The present application can also be implemented or applied through other different specific embodiments. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present application. It should be understood that the preferred embodiments are only for explaining the present application and not for limiting the protection scope of the present application.

[0038] In recent years, the development of in-vehicle intelligent systems and cloud computing technologies has enabled the rapid development of autonomous driving technology from the driving assistance stage to the driverless stage. Currently, autonomous driving technology mainly relies on high-precision maps to achieve long-distance perception and lane-level planning and guidance. The production of high-precision maps currently mainly uses on-site measurement methods. The high-precision maps obtained by measuring with high-precision sensors have high accuracy and are usually used as reference true value data for algorithm research and development or map production (see a crowdsourced map quality evaluation method, system, and storage medium disclosed in Patent CN116124178A). By using real-time kinematic (RTK), global navigation satellite system (GNSS) static measurement, and unmanned aerial vehicle mapping means, crowdsourced map data within a certain range can be obtained, and its accuracy can reach the centimeter level. However, due to factors such as high labor costs, high equipment prices, and harsh working conditions, conducting on-site measurements on a larger scale is not the optimal solution. Therefore, crowdsourced mapping is an alternative to on-site measurement methods. Compared with the high-precision and high-cost professional acquisition methods, the mapping technology of the crowdsourced acquisition method has become an ideal way for large-scale and low-cost mapping. With the development of artificial intelligence algorithms and sensors, crowdsourced mapping can achieve a high degree of automation in the mapping process. However, due to the diverse sources and uneven quality of crowdsourced data and the limitations of existing mapping methods, the quality problems of maps generated in a crowdsourced manner objectively exist, and the crowdsourced maps output in a crowdsourced manner need to be strictly verified before they can be released and used. Therefore, a reasonable and effective crowdsourced mapping quality evaluation method is an important requirement in the field of autonomous driving.

[0039] Currently, in response to the above requirements, a crowdsourced map quality evaluation method, system, and storage medium disclosed in Patent CN116124178A can complete quality evaluation by comparing elements based on the crowdsourced map information sent from the cloud and the road information collected by in-vehicle devices. It is necessary to implement communication between cloud data and vehicle data, but the scope of test evaluation is limited to areas with ground truth references. The evaluation process involves the matching process of vehicle data and cloud data, resulting in a certain amount of computational overhead; a vehicle positioning method and system based on in-vehicle cameras and maps disclosed in Patent CN117930297A can calculate the first lateral distance and the second lateral distance in different ways, and use the difference between the two lateral distances as an actual residual to match with the prediction error of the Long Short-Term Memory (LSTM) model as an indicator of whether the GNSS signal deviates. The main effect is to improve the accuracy of vehicle positioning. However, for data collected through crowdsourcing, there is still a lack of a method for evaluating the position accuracy in the absence of a high-precision base map reference, and relevant position accuracy evaluation and position deviation correction methods have not been fully explored in scenarios without a base map; in the paper "Positional Accuracy EvaluationIntegrated with EgoSensors for Crowdsourced HD Map", the quality of map elements is directly evaluated based on empirical rules, and specific rules are adopted for quality evaluation of map data in different scenarios. The design of this rule mainly relies on subjective human experience judgment, and its generalization ability needs to be improved.

[0040] In view of the above technical problems, the present application provides a position accuracy evaluation method, which is implemented by an electronic device. Figure 1 The following is a schematic flowchart of a position accuracy evaluation method provided by an embodiment of the present application, as Figure 1 shown, including the following steps S101 to S103: Step S101: Obtain the trajectory data and visual data collected by the vehicle during driving.

[0041] In an embodiment of the present application, the electronic device can obtain the corresponding trajectory data and visual data of the vehicle during driving. Exemplarily, the electronic device can obtain the trajectory data and visual data through in-vehicle sensors. Among them, the trajectory data can be obtained by a GNSS sensor and an Inertial Measurement Unit (IMU) sensor, and the visual data can be obtained based on an in-vehicle visual sensor.

[0042] Exemplarily, the trajectory data is from an inertial navigation sensor (inertial measurement unit sensor) and includes important features such as the absolute position, speed, driving direction of the vehicle, and the number of satellites captured by the GNSS sensor. These features reflect the physical environment in which the vehicle is located during driving and the satellite signal quality of the road section. If there is occlusion in the physical environment or the satellite signal quality is poor, it may cause the offset and detection loss of the mapping elements caused by the vehicle on this road section.

[0043] Exemplarily, the electronic device can preprocess the trajectory data, such as denoising, smoothing, etc., to improve the accuracy and reliability of the data.

[0044] Exemplarily, the visual data is collected by an on-vehicle camera and includes environmental information such as road signs, lane lines, traffic lights, etc., aiming to supplement the information missing in the trajectory data.

[0045] Exemplarily, the electronic device can preprocess the image data (visual data), such as grayscale conversion, binarization, denoising, etc., to facilitate subsequent feature extraction and recognition.

[0046] In the embodiments of the present application, the electronic device is a device with a position accuracy evaluation function, which can be a tablet computer, a notebook computer, a palm computer, a personal digital assistant (Personal Digital Assistant, PDA), a desktop computer, etc. The exemplary electronic device is not limited here.

[0047] Step S102: Generate vector data corresponding to the driving environment of the vehicle based on the trajectory data and the visual data; the vector data contains at least one vector element.

[0048] In the embodiments of the present application, the electronic device can use the Simultaneous Localization and Mapping (SLAM) algorithm to generate vector data corresponding to the driving environment of the vehicle based on the trajectory data and the visual data.

[0049] In the embodiments of the present application, the vector elements at least include arrows, lane boundary lines, road signs, and traffic facilities.

[0050] Exemplarily, the implementation of generating vector data can also be as follows: Extract key feature points from the trajectory data, such as turning points and parking points; Use computer vision technology to extract key visual features such as road signs, traffic lights, and obstacles from the image data; Generation of road vector data: Combine the key feature points of the trajectory and the key visual features, and use geometric modeling and interpolation algorithms to generate vector data such as the center line and boundary line of the road. Smooth and optimize the generated vector data to improve the accuracy and usability of the data; Generation of traffic element vector data: Generate corresponding vector data according to the position and attribute information of traffic elements such as recognized traffic lights and intersections; Integrate and associate the traffic element vector data with the road vector data to construct a complete driving environment vector map.

[0051] Exemplarily, after generating the vector data, post-processing can also be performed on the generated vector data, such as removing redundant data, repairing broken line segments, etc., and formatting and standardizing the data for subsequent applications.

[0052] Step S103: Based on the trajectory data, visual data, and vector data, evaluate the position accuracy of the vector elements to obtain a position accuracy evaluation result.

[0053] In the embodiments of the present application, when the electronic device obtains the trajectory data, visual data, and vector data, it will evaluate the position accuracy of the vector elements based on their respective data characteristics to obtain a position accuracy evaluation result.

[0054] Exemplarily, the electronic device can use a neural network model to construct a position accuracy evaluation model, and then use the position accuracy evaluation model for evaluation to obtain a position accuracy evaluation result.

[0055] Exemplarily, the electronic device can also match the trajectory data with the vector data to determine which vector elements the vehicle interacts with during driving (such as driving on the road, approaching traffic signs, etc.); Record the position information at the time of matching, including Global Positioning System (GPS) coordinates, timestamps, and the corresponding vector element types, and use the visual data to verify the matching result; Judge the position accuracy of the vector elements by comparing the position relationship between the image captured by the camera and the vector elements. Of course, technologies such as image recognition and object detection can also be used to assist the verification process.

[0056] Compared with the technical solutions that need to rely on high-precision maps in the related art, the present application realizes the internal evaluation of position accuracy according to the inherent characteristics of the trajectory data, visual data, and vector data, and can evaluate the position accuracy without relying on high-precision maps, improving the accuracy and efficiency of position accuracy evaluation.

[0057] In some embodiments, such as Figure 2 shown, when the electronic device executes the above step S103, it may further include the following steps S201 and S202: Step S201: Extract features from the vector data, trajectory data, and visual data to obtain multiple feature information.

[0058] In the embodiments of the present application, the electronic device will extract features from the vector data, trajectory data, and visual data to obtain multiple feature information.

[0059] Exemplarily, the electronic device may perform feature extraction based on a neural network of deep learning, or may perform feature extraction based on an algorithm of machine learning. Of course, it may also perform feature extraction based on statistics. The way of feature extraction can be set according to actual needs and application scenarios, and the present application does not limit this.

[0060] In the embodiments of the present application, the electronic device may perform feature extraction after fusing the vector data, trajectory data, and visual data to obtain multiple feature information, or may perform corresponding feature extraction based on a single data. For example, perform feature extraction on the visual data to extract visual feature information.

[0061] Step S202: Perform feature mapping on the multiple feature information to predict the position accuracy corresponding to the vector element, and obtain a position accuracy evaluation result.

[0062] In the embodiments of the present application, the electronic device may map the multiple feature information into a feature space to form a feature vector, and may use appropriate feature scaling and normalization methods to ensure the consistency and comparability of the feature vector, and select a suitable position accuracy evaluation model for prediction to obtain a position accuracy evaluation result.

[0063] In this way, for the position accuracy of the vector element, the predicted position accuracy evaluation result can provide important reference information for fields such as autonomous driving and intelligent transportation systems, and help improve the safety and reliability of the system.

[0064] In some embodiments, the multiple feature information includes trajectory feature information, positioning feature information, visual feature information, and geometric feature information and semantic feature information of the vector element; such as Figure 3 shown, when the electronic device executes the above step S201, it may execute the following steps S301 to S303: Step S301: Extract features from the vector data and trajectory data to obtain geometric feature information and semantic feature information.

[0065] In an embodiment of the present application, feature extraction is performed on the vector data corresponding to each trajectory point in the trajectory data to obtain geometric feature information and semantic feature information.

[0066] In an embodiment of the present application, the geometric feature information may include shape features, position features, and topological features; the shape features may be the boundary shapes of vector elements (such as roads, buildings, traffic signs, etc.), and calculate their perimeter, area, aspect ratio, etc.; for linear elements (such as roads), their length, curvature, direction, etc. can be calculated; determine the position of the vector element in the geographic coordinate system, including longitude, latitude, altitude, etc.; calculate the relative position relationship between the vector element and other elements (such as adjacent roads, intersections); analyze the connection relationship between vector elements, such as the intersection points of roads, the adjacent relationship of buildings, etc.; extract network structure features, such as the connectivity of the road network, the degree of nodes, etc.

[0067] In an embodiment of the present application, the semantic feature information may include category information, function information, and attribute information; among them, according to the attributes and metadata of the vector element, determine its category (such as road type, building use, etc.); use classification labels or codes to represent this category information; function information: analyze the function or role of the vector element, such as the indication meaning of traffic signs.

[0068] Step S302: Perform feature extraction on the trajectory data to obtain trajectory feature information and positioning feature information.

[0069] In an embodiment of the present application, the electronic device can perform feature extraction on the trajectory data to obtain trajectory feature information and positioning feature information. The trajectory feature information may include trajectory points, timestamps, and category information; the positioning feature information may include the number of satellites acquired corresponding to each trajectory point.

[0070] Step S303: Perform feature extraction on the visual data to obtain visual feature information.

[0071] In an embodiment of the present application, the electronic device can perform feature extraction on the visual data to obtain visual feature information. Exemplarily, the extraction method of image features can use traditional methods: Scale-Invariant Feature Transform (SIFT), Speeded-Up Robust Features (SURF), or use deep learning methods: such as using a pre-trained Convolutional Neural Network (CNN) model (such as VGG16, ResNet, Inception, etc.) for feature extraction.

[0072] In this way, feature extraction is performed on visual data, trajectory data, and vector data to obtain multiple feature information, and then the position accuracy of vector elements is mapped based on the multiple feature information, with relatively high accuracy.

[0073] In some embodiments, the semantic feature information includes the field of view distance; the field of view distance is the distance between the center point of the vector element and the driving trajectory of the vehicle.

[0074] In the embodiments of the present application, the semantic feature information further includes the field of view distance, which is a derivative feature information based on vector data and trajectory data, and the field of view distance is the shortest Euclidean distance between the center point of the vector element and the driving trajectory of the vehicle.

[0075] As Figure 4 shown, the arrow is the vehicle-end mapping element 41, the dashed rectangle is the vehicle-end mapping lane line element 42, the dashed line is the trajectory 43, and the shortest Euclidean distance from the center point of the vehicle-end mapping element 41 to the trajectory 43 is the field of view distance 44; the shortest Euclidean distance from the center point of the vehicle-end mapping lane line element 42 to the trajectory 43 is the field of view distance 44. Among them, the vehicle-end mapping element and the vehicle-end mapping lane line element are vector elements.

[0076] Exemplarily, the field of view distance is the distance from the vehicle-mounted sensor to the landmark, and the positioning point of the moving vehicle is regarded as the position of the vehicle-mounted sensor. Exemplarily, the field of view distance can be the shortest Euclidean distance between the center point of the vector element and the vehicle driving trajectory line.

[0077] In the embodiments of the present application, the field of view distance is derived based on vector data and trajectory data. Based on the field of view distance, the accuracy in visual data can be better corrected, and thus the accuracy of predicting the position accuracy based on visual data can be improved.

[0078] In some embodiments, the position accuracy evaluation result includes the accuracy evaluation result of the vector element and the position deviation; as Figure 5 shown, the electronic device may further include the following steps S501 and S502: Step S501, in response to the accuracy evaluation result of the vector element being inaccurate, based on the position deviation of the vector element, perform position adjustment on the vector element to obtain the adjusted vector data.

[0079] In the embodiments of the present application, if the electronic device responds to the accuracy evaluation result of the vector element being inaccurate, it can adjust the position of the vector element based on the position deviation of the vector element. Since when the electronic device predicts the position accuracy of the vector element to obtain the position accuracy evaluation result, the position deviation of the vector element will also be attached. As Figure 4As shown, the true value element 45 and the true value lane line element 46 are element true values, and the absolute position error 47, that is, the position deviation between the position of the predicted vector element and the element true value. The electronic device can adjust the position of the vector element based on the position deviation to obtain the adjusted vector data.

[0080] Step S502: Based on the adjusted vector data, construct a map corresponding to the driving environment.

[0081] In an embodiment of the present application, after the electronic device obtains the adjusted vector data, it can construct a map corresponding to the driving environment based on the adjusted vector data, and then perform assisted driving based on the constructed map.

[0082] In this way, since the position of the vector element in the adjusted vector data is adjusted and has high accuracy, the scene map constructed based on the adjusted vector data also has high accuracy.

[0083] In some embodiments, before the electronic device executes step S302, as Figure 6 shown, the following steps S601 and S602 can also be executed: Step S601: Based on the trajectory feature information in the multiple feature information, identify the scene of the driving environment to obtain the scene category.

[0084] In an embodiment of the present application, the electronic device identifies the driving scene from the trajectory feature information in the multiple feature information to obtain the scene category. The trajectory feature information may include the number of satellites in view, the positioning status, the direction, the speed, etc. To identify the scene of the driving environment, the electronic device can perform scene identification based on some or all of this information to obtain the scene category. Exemplarily, the scene category may be an open scene (non-tunnel scene) and a satellite denial scene (tunnel scene).

[0085] Step S602: Based on the scene category, perform feature mapping on the multiple feature information, and predict the position accuracy corresponding to the vector element to obtain the position accuracy evaluation result.

[0086] In an embodiment of the present application, for different scene categories, the electronic device can use different prediction methods. Considering that in a tunnel scene, the data obtained may be biased or have low accuracy, and the quality is poor compared with the data obtained in a non-tunnel scene. At this time, some methods can be used to enhance the quality of the data, or perform refined processing on the data with low quality, and then perform the prediction work on the position accuracy of the vector element.

[0087] In this way, the position accuracy of the vector element based on the scene category will be more intelligent and the prediction accuracy will also be higher.

[0088] In some embodiments, the trajectory feature information includes at least one trajectory point, and each trajectory point corresponds to a number of satellite acquisitions and a positioning status; when the electronic device executes step S601, as Figure 7 shown, it may include the following steps S701 to S703: Step S701: Determine the average value of the number of satellite acquisitions of at least one number of satellite acquisitions, and statistically analyze the discrete distribution of at least one positioning status.

[0089] In the embodiments of the present application, the electronic device can obtain the number of satellite acquisitions and the positioning status of each trajectory point in at least one trajectory point.

[0090] Exemplarily, the positioning status may include different types, such as two-dimensional positioning, three-dimensional positioning, single-point positioning, differential positioning, etc. The type of positioning status can be defined based on actual requirements and application scenarios, and the present application does not make any limitations in this regard.

[0091] In the embodiments of the present application, after the electronic device obtains at least one number of satellite acquisitions, it determines the average value of at least one number of satellite acquisitions to obtain the average number of satellite acquisitions, and statistically analyzes the discrete distribution of at least one positioning status. Among them, the discrete distribution of the positioning status can understand the probability distribution of the current positioning status and can more accurately determine the position.

[0092] Step S702: In response to the average number of satellite acquisitions being greater than the first threshold and the discrete distribution of at least one positioning status being within the first preset distribution set, determine that the driving environment is an open scenario.

[0093] In the embodiments of the present application, if the average number of satellite acquisitions is greater than the first threshold and the discrete distribution of at least one positioning status is within the first preset distribution set, it is determined that the driving environment of the vehicle is an open scenario. The first threshold may be 10, but of course it may also be other values, which can be set based on actual requirements and application scenarios, and the present application does not make any limitations in this regard; the first preset distribution set may be , but of course it may also be other sets, which can be set based on actual requirements and application scenarios, and the present application does not make any limitations in this regard.

[0094] Step S703: In response to the average number of satellite acquisitions being less than the second threshold and the ratio of the discrete distribution of at least one positioning status being within the second preset distribution set being greater than the preset ratio, determine that the driving environment is a satellite denial scenario.

[0095] In an embodiment of the present application, if the average number of satellites acquired is less than a second threshold, and the ratio of the discrete distribution of at least one positioning state located in a second preset distribution set is greater than a preset ratio, it is determined that the driving environment is in a satellite rejection scenario. The second threshold may be 5 or other values. The second threshold may be dependent on the first threshold or may exist independently, and can be set based on actual requirements and application scenarios. The present application does not limit this; the second preset distribution set may be {1, 6}, or of course other sets. There is no intersection between the first preset distribution set and the second preset distribution set. The positioning states aggregated in the first preset distribution set are characterized as good signals, and the positioning states aggregated in the second preset distribution set are characterized as bad signals. Of course, the relationship between the first preset distribution set and the second preset distribution set can be set based on actual requirements and application scenarios. The present application does not limit this; the preset ratio may be one-half, one-third, or other ratios. Exemplarily, the ratio can be set based on actual requirements and application scenarios. The present application does not limit this.

[0096] In this way, based on the number of satellites acquired and the positioning state in the trajectory feature information, the driving scenario of the current vehicle is distinguished, and then the position accuracy prediction of vector elements is performed according to different scenarios, with relatively high accuracy.

[0097] In some embodiments, when the electronic device executes the above step S103, it may include the following steps: using a preset position accuracy evaluation model, based on trajectory data, visual data, and vector data, evaluate the position accuracy of vector elements to obtain a position accuracy evaluation result.

[0098] In an embodiment of the present application, the electronic device may use a preset position accuracy evaluation model to evaluate the position accuracy of vector elements based on trajectory data, visual data, and vector data. The preset position accuracy evaluation model is a model pre-trained based on a neural network and having a function of evaluating the position accuracy of vector elements.

[0099] In this way, based on the trained preset position accuracy evaluation model, it is possible to directly evaluate the position accuracy of vector elements based on trajectory data, visual data, and vector data, and the position accuracy evaluation can be achieved without relying on the true value elements in the high-precision map, improving the efficiency of position accuracy evaluation.

[0100] In some embodiments, the preset position accuracy evaluation model is provided with corresponding model parameters for different scenarios; when the electronic device executes the above step S103, as Figure 8 shown, it may include the following step S801 and step S802: Step S801, determine the scenario category of the driving environment of the vehicle.

[0101] In an embodiment of the present application, since the quality of the trajectory data, visual data, and vector data obtained in different scenarios is different, the electronic device can determine the scenario category of the driving environment of the vehicle, and then perform subsequent processing based on the scenario category, making the processing more accurate.

[0102] Step S802: Use a preset position accuracy evaluation model, call the model parameters matching the scenario category, and evaluate the position accuracy of the vector elements based on the trajectory data, visual data, and vector data to obtain a position accuracy evaluation result.

[0103] In an embodiment of the present application, after the electronic device obtains the corresponding scenario category, considering different quality data, for the scenario category of the driving environment of the vehicle, different model parameters can be set for the preset position accuracy evaluation model. In this way, use the preset position accuracy evaluation model, call the model parameters matching the scenario category, and evaluate the position accuracy of the vector elements based on the trajectory data, visual data, and vector data to obtain a position accuracy evaluation result.

[0104] In this way, the model parameters of the model for evaluating the position accuracy of the vector elements are different according to different scenario categories, which can better adapt to the changes of the scenario, and the accuracy of the position accuracy evaluation is relatively high.

[0105] In some embodiments, before the electronic device executes "using a preset position accuracy evaluation model to evaluate the position accuracy of vector elements based on trajectory data, visual data, and vector data to obtain a position accuracy evaluation result", as Figure 9 shown, the following steps S901 to S903 may further be included: Step S901: Obtain sample data; the sample data includes trajectory sample data, visual sample data, and vector sample data matching the trajectory sample data and visual sample data.

[0106] Exemplarily, the preparation of the training set (sample data) includes collecting samples containing trajectory data, visual data, vector data, and field-of-view distance derivative features from crowdsourced data, and ensuring that these samples cover different road scenarios and conditions. Each sample data is labeled with the true value of its position accuracy; perform a feature selection step, analyze the initially extracted features, use statistical methods to screen out the features that have a significant impact on the position accuracy evaluation, and further reduce the feature dimension through dimensionality reduction techniques, retaining the most important features for the model prediction effect.

[0107] Step S902: Use the position accuracy evaluation model to be trained, and based on the sample data, predict the sample position accuracy of the sample vector elements to obtain a sample position prediction result.

[0108] In an embodiment of the present application, an electronic device may construct an architecture of a neural network model (a position accuracy evaluation model to be trained), including an input layer, multiple hidden layers, and an output layer, and select an appropriate loss function and optimization algorithm; then, input sample data into the position accuracy evaluation model to be trained, predict the sample position accuracy of the sample vector elements, and obtain a sample position prediction result.

[0109] Step S903: Based on the sample prediction result and the target prediction result corresponding to the sample data, adjust the model parameters of the position accuracy evaluation model to be trained to obtain a preset position accuracy evaluation model.

[0110] Exemplarily, the electronic device inputs the training set data into the neural network model, calculates the output result (sample position prediction result) through forward propagation, and calculates the error (loss information) between the model output result and the true value.

[0111] Exemplarily, based on the error between the model output result and the true value, update the model parameters through the backpropagation algorithm, and verify and evaluate the model through the validation set to identify and solve the overfitting or underfitting problem, and finally obtain a preset position accuracy evaluation model.

[0112] Among them, the preset neural network model (preset position accuracy evaluation model) is obtained by batch training the neural network using the training set (sample data), and the model output result is the position accuracy evaluation result.

[0113] As Figure 10 shown, an exemplary training method for a position accuracy evaluation model is provided. Exemplarily, the training method can be applied to a training device, and the training device may include a data acquisition unit, a data association unit, a data matching subunit, a data visualization subunit, a scene recognition unit, and a model feature extraction subunit, a model training subunit, a position accuracy prediction unit, an output subunit, an adjustment subunit, and a loss function subunit included in the position accuracy prediction unit.

[0114] Exemplarily, the training method may include the following steps S1001 to S1004, which will be discussed in conjunction with the training device here: Step S1001: Obtain a training set.

[0115] Here, the training device will use the data acquisition unit to specify the data time range and route; the data acquisition unit can obtain a complete set of supporting data collected by the same vehicle acquisition platform (in-vehicle sensor) within the specified time range: trajectory data, vector data, and image (visual) data.

[0116] The data association unit specifies the time length of a single sample. The data association unit can split the entire dataset according to the specified time length to create data samples (sample data) at a specified scale. Exemplarily, by inputting the time range and route of the data: start time, end time, route name, and mapping date parameters, the corresponding paths of the trajectory data, visual data, and vector data can be obtained. Then, based on the obtained corresponding paths of the trajectory data, visual data, and vector data, the trajectory data, visual data, and vector data are retrieved from the corresponding storage areas, and data extraction and packaging are performed in accordance with the standardized data packet organization form.

[0117] The data matching subunit assigns different timestamps to the trajectory data, visual data, and vector data, and develops corresponding string processing and matching logics to complete the data matching. In this way, multiple data samples of the same time length can be obtained. Among them, the assignment methods of the timestamps of the trajectory data, visual data, and vector data are different, and different data can be distinguished based on the strings of the timestamps.

[0118] The data visualization subunit can obtain the complete set of trajectory data, vector data, and visual data collected by the same vehicle acquisition platform (in-vehicle sensors) within a specified time range by the data acquisition unit, and can implement the visualization function through the interface interaction method, and at the same time realize the visualization of the vector data, trajectory data, and positioning data (visual data).

[0119] Step S1002, scene recognition.

[0120] Here, the scene recognition unit can extract the number of satellites in view ( ), positioning status ( ) in the trajectory data to construct a feature vector . For this feature vector, the variation of the feature vector within the window is judged in the form of a sliding window. Based on the variation of the feature vector within the sliding window (the time length of a single sample), it is identified whether the current trajectory is in a tunnel, that is, a satellite denial environment. Exemplarily, the determination method can include: S1. Construct a time series, construct a target GNSS data feature sequence through a sliding window, and statistically analyze the statistical features of the sliding window (see the above-mentioned step matching subunit); S2. Perform feature statistics, calculate the mean value of the number of satellites in view, and statistically analyze the discrete distribution of the positioning status, and judge the open scene and the satellite denial scene according to the established rules. Exemplarily, the rules can be: The mean value is greater than the threshold 10, and distributed in then it is determined as an open scene; The mean value is less than 5, and distributed in If the proportion is greater than 50%, it is determined as the satellite rejection link; S3. Perform sliding window update and dynamic judgment. The sliding window moves at a fixed step size and repeats the above judgment, recording the recognition results of each window. In this way, the corresponding scene recognition results can be obtained for the above-mentioned multiple data samples of the same time length.

[0121] Step S1003. Position accuracy prediction.

[0122] Here, the model feature extraction sub-unit is used to extract multi-dimensional feature spaces (trajectory features, vector features, visual features, and derivative features (geometric features and semantic features)) from trajectory data, visual data, and vector data (crowdsourced map data), and serialize these feature data in tensor format and input them into a preset neural network model (the position accuracy prediction model to be trained). The trajectory features are obtained by processing the trajectory data through a specific neural network feature extraction sub-unit (for example, a one-dimensional convolutional network), the visual features are processed by a corresponding neural network (for example, a two-dimensional convolutional network) for the visual data collected by the vehicle-mounted camera, the vector features are generated by the simultaneous localization and mapping (SLAM) technology, and the derivative feature (semantic feature) mainly refers to the field of view distance, which is the distance from the vehicle-mounted sensor to the landmark, and the positioning point of the moving vehicle is regarded as the position of the vehicle-mounted sensor.

[0123] In this way, a position accuracy evaluation model is constructed based on the neural network, fully considering the temporal characteristics of the data and the mutual relationship between multiple types of data, and improving the effect of position accuracy evaluation.

[0124] It should be noted that when performing position accuracy prediction, the scene recognition result will be input as a parameter into the position accuracy prediction unit, and a special position accuracy classifier (position accuracy prediction unit) with corresponding parameters is set for different scene labels. The special position accuracy classifier is pre-trained based on the corresponding category samples. The position accuracy of each element is evaluated according to the characteristics of each scene. Exemplarily, an "accurate" or "inaccurate" label will be assigned to each individual vehicle-end vector element.

[0125] An output subunit is used to decode the trajectory features, visual features, and vector features obtained by the model feature extraction subunit, perform position accuracy classification prediction, obtain a position accuracy classification result, write it into the output data according to a fixed data type, and store it in a medium. Among them, the model feature extraction subunit extracts appropriate features for evaluating the construction of the model, using different types of features such as statistical features, frequency domain features, and time domain features. At the same time, a feature selection method is adopted to screen out the features that have a significant impact on quality evaluation. Then, an appropriate evaluation model (the position accuracy evaluation model to be trained) is selected for construction and training, and the model is optimized according to the validation set to obtain the best performance, and then combined with the vehicle motion model, sensor fusion technology, etc. for fusion evaluation. Among them, through image processing technology, the visual features can calculate the geometric transformation between the image and the road coordinate system, and then obtain the vehicle's lane position, and compare it with the trajectory data, so as to detect the sections with trajectory drift and deviation, which can be used to assist in correcting the difference between the trajectory and the vector mapping (vector elements).

[0126] Step S1004, model training.

[0127] Here, a model training subunit is used to train a preset neural network model (the position accuracy prediction model to be trained) based on training sample data and true value labels, compare the prediction result with the true result to obtain an error, and backpropagate the error to each network layer of the model to adjust the weights and parameters of the network layer, so as to obtain a preset position accuracy prediction model. Exemplarily, the training sample data can refer to the position information and semantic label information of the crowdsourced map data collected at the vehicle end and the corresponding high-precision map true value elements. The semantic label information includes the classification information in the map ground markings. If the type of the ground marking is a lane line, it should include the classification of its solid line and dashed line. Based on the sample crowdsourced map data and the corresponding high-precision map true value elements, the preset neural network model is trained.

[0128] Among them, an adjustment subunit is used to adjust the feature weights mentioned in the model feature extraction subunit of the preset neural network model based on the difference between the position of the sample crowdsourced map data and the position of the corresponding true value data, and the difference between the attributes of the sample crowdsourced map data and the true value semantic labels (adjust the model parameters of the position accuracy evaluation model to be trained based on the loss information to obtain a preset position accuracy evaluation model).

[0129] A loss function subunit is used to calculate the cross entropy between the position of the crowdsourced map data collected at the vehicle end in the sample data and the high-precision map position label to obtain a loss value (loss information).

[0130] The adjustment subunit is used to adjust the parameters of the preset neural network model through backpropagation based on the loss value.

[0131] In this way, a trained preset neural network model (preset position accuracy evaluation model) can be obtained.

[0132] As Figure 11 shown, an exemplary map generation method is provided, which may include the following steps S1101 and S1102: Step S1101: Input the trajectory data, visual data, and vector data into the preset neural network model, and output the position accuracy based on the crowd-sourced map data.

[0133] Here, the trajectory data and visual data will be obtained first, and then the vector data will be obtained based on the trajectory data and visual data through the simultaneous localization and mapping algorithm.

[0134] In this way, without relying on remote sensing images and field measurements, the position accuracy of road markings such as arrows and lane boundary lines in the map to be evaluated can be evaluated on the road sections in different scenarios, and corresponding labels can be assigned to each vector element (road sign) according to the position accuracy prediction result.

[0135] Step S1102: Assist in the production of crowd-sourced maps according to the position accuracy based on the crowd-sourced map data output by the model.

[0136] Here, the production of crowd-sourced maps is assisted according to the position accuracy output by the model. Among them, on the basis of outputting the position accuracy, the model will also output the position deviation.

[0137] In this way, internal evaluation is realized according to the characteristics of the crowd-sourced map data itself, the dependence on high-precision maps in the crowd-sourced map production process is reduced, and the production efficiency of the crowd-sourced maps is improved; it also gets rid of the dependence on the ground truth in the high-precision maps for the quality evaluation of the crowd-sourced map data, and evaluates the data quality based on the internal characteristics of the data, saving the map production cost.

[0138] As Figure 12 shown, an implementation manner of an exemplary method for evaluating the position accuracy of crowd-sourced maps is provided, including steps S1201 to S1205: Step S1201: Data acquisition.

[0139] Here, the original data 121 of the vehicle-mounted sensor can be obtained through the vehicle-mounted sensor of the vehicle. The original data includes the original trajectory data and the original visual data. Among them, the original trajectory data is obtained based on the GNSS sensor and the IMU sensor, and includes important features such as the absolute position, speed, driving direction of the vehicle, and the number of satellites captured by the GNSS sensor. The original visual data includes environmental information such as road signs, lane lines, and traffic lights, aiming to supplement the information missing in the trajectory data and provide the model with a richer environmental perception ability. The collected visual data is usually stored in image format and can assist in scene recognition.

[0140] Step S1202, data extraction.

[0141] Here, the required trajectory data and visual data (multi-source sensor data 122) can be extracted from the original trajectory data and original visual data. Since the original data may contain all data for all time periods, the required data needs to be extracted here. For example, the position, driving direction, and the number of satellites captured by the GNSS sensor included in the trajectory data within a specified time range can be extracted from the original trajectory data, and data such as speed may not be extracted. Exemplarily, trajectory data (absolute position and the number of satellites) and visual data (the data collected by the camera at a fixed frequency is all required, and extraction means the visual data that matches the trajectory data) are extracted from the original data. Which feature information to extract can be set based on actual requirements and application scenarios, and this application does not make a limitation on this.

[0142] Here, the simultaneous localization and mapping algorithm can also be used to determine the corresponding vector data (vehicle-end high-freshness data 126: vectorization result 127 and trajectory data 128) based on the trajectory data (positioning signal 123 and inertial navigation data 124) and visual data (visual information 125). The vector data includes road signs, lane lines, traffic facilities, etc. obtained from multi-source mapping, and includes geometric attributes such as the centroid, perimeter, and area of vector elements (road signs).

[0143] Step S1203, feature extraction.

[0144] Here, feature extraction is performed on the trajectory data, vector data, and visual data to obtain a feature space: geometric feature 129, semantic feature 1210, trajectory feature 1211, positioning feature 1212, and visual feature 1213.

[0145] Step S1204, scene classification.

[0146] Here, the corresponding scene category is determined based on the number of satellites searched and the positioning status in the trajectory feature data. Among them, the scene categories include: non-tunnel scene 1214 (open scene) and tunnel scene 1215 (satellite rejection scene). Step S1205, mapping features.

[0147] Here, based on the scene category, the position accuracy prediction model 1216 (preset position accuracy prediction model) with corresponding parameters is called to map features, and then the geometric feature 129, semantic feature 1210, trajectory feature 1211, positioning feature 1212, and visual feature 1213 are input for position accuracy prediction to obtain the position accuracy evaluation result of each vector element.

[0148] Step S1206, evaluation.

[0149] It should be noted that, in addition to the accuracy of each vector element output by the preset position accuracy prediction model, information such as confidence and position deviation can also be output, that is, the position accuracy evaluation result 1217 is obtained.

[0150] The present application provides a method for evaluating position accuracy, which acquires trajectory data and visual data collected during the driving of a vehicle; generates vector data corresponding to the driving environment of the vehicle based on the trajectory data and the visual data; the vector data includes at least one vector element; evaluates the position accuracy of the vector element based on the trajectory data, the visual data, and the vector data, and obtains a position accuracy evaluation result. The present application realizes the internal evaluation of position accuracy according to the own characteristics of the trajectory data, the visual data, and the vector data, and can realize the evaluation of position accuracy without relying on a high-precision map, improving the accuracy and efficiency of position accuracy evaluation.

[0151] An embodiment of the present application provides a position accuracy evaluation device 13, as Figure 13 shown, including: An acquisition module 131, configured to acquire trajectory data and visual data collected during the driving of a vehicle; A generation module 132, configured to generate vector data corresponding to the driving environment of the vehicle based on the trajectory data and the visual data; the vector data includes at least one vector element; An evaluation module 133, configured to evaluate the position accuracy of the vector element based on the trajectory data, the visual data, and the vector data, and obtain a position accuracy evaluation result.

[0152] In an embodiment of the present application, the evaluation module 133 is further configured to extract feature information from the vector data, the trajectory data, and the visual data to obtain a plurality of feature information; perform feature mapping on the plurality of feature information to predict the position accuracy corresponding to the vector element, and obtain a position accuracy evaluation result.

[0153] In an embodiment of the present application, the plurality of feature information includes trajectory feature information, positioning feature information, visual feature information, and geometric feature information and semantic feature information of the vector element; the evaluation module 133 is further configured to extract geometric feature information and semantic feature information from the vector data and the trajectory data; extract trajectory feature information and positioning feature information from the trajectory data; extract visual feature information from the visual data.

[0154] In an embodiment of the present application, the semantic feature information includes a field of view distance; the field of view distance is the distance between the center point of the vector element and the driving trajectory of the vehicle.

[0155] In an embodiment of the present application, the position accuracy evaluation result includes the accuracy evaluation result of the vector element and the position deviation; the evaluation module 133 is further configured to, in response to the accuracy evaluation result of the vector element being inaccurate, perform position adjustment on the vector element based on the position deviation of the vector element to obtain the adjusted vector data; and construct a map corresponding to the driving environment based on the adjusted vector data.

[0156] In an embodiment of the present application, the evaluation module 133 is further configured to identify the scene of the driving environment based on the trajectory feature information among the multiple feature information to obtain the scene category; and perform feature mapping on the multiple feature information based on the scene category to predict the position accuracy corresponding to the vector element, so as to obtain the position accuracy evaluation result.

[0157] In an embodiment of the present application, the trajectory feature information includes at least one trajectory point, and each trajectory point corresponds to a satellite search number and a positioning state; the evaluation module 133 is further configured to determine the average value of the satellite search numbers of the at least one satellite search number and count the discrete distribution of the at least one positioning state; in response to the average value of the satellite search numbers being greater than the first threshold and the discrete distribution of the at least one positioning state being located in the first preset distribution set, determine that the driving environment is in an open scene; in response to the average value of the satellite search numbers being less than the second threshold and the ratio of the discrete distribution of the at least one positioning state being located in the second preset distribution set being greater than the preset ratio, determine that the driving environment is in a satellite rejection scene.

[0158] In an embodiment of the present application, the evaluation module 133 is further configured to use a preset position accuracy evaluation model to evaluate the position accuracy of the vector element based on the trajectory data, visual data, and vector data, so as to obtain the position accuracy evaluation result.

[0159] In an embodiment of the present application, the preset position accuracy evaluation model is provided with corresponding model parameters for different scenes; the evaluation module 133 is further configured to determine the scene category of the driving environment where the vehicle is located; use the preset position accuracy evaluation model to call the model parameters matching the scene category, and evaluate the position accuracy of the vector element based on the trajectory data, visual data, and vector data, so as to obtain the position accuracy evaluation result.

[0160] In an embodiment of the present application, the evaluation module 133 is further configured to obtain sample data; the sample data includes trajectory sample data, visual sample data, and vector sample data matching the trajectory sample data and the visual sample data; use the position accuracy evaluation model to be trained to predict the sample position accuracy of the sample vector element based on the sample data to obtain the sample position prediction result; and adjust the model parameters of the position accuracy evaluation model to be trained based on the sample prediction result and the target prediction result corresponding to the sample data to obtain the preset position accuracy evaluation model.

[0161] An embodiment of the present application provides a position accuracy evaluation device, which acquires trajectory data and visual data collected during the driving of a vehicle; generates vector data corresponding to the driving environment of the vehicle based on the trajectory data and the visual data; the vector data includes at least one vector element; evaluates the position accuracy of the vector element based on the trajectory data, the visual data, and the vector data to obtain a position accuracy evaluation result; adopting the above implementation solution, the present application realizes the internal evaluation of the position accuracy according to the characteristics of the trajectory data, the visual data, and the vector data itself, and can realize the evaluation of the position accuracy without relying on a high-precision map, improving the accuracy and efficiency of the position accuracy evaluation.

[0162] Figure 14 FIG. 4 is a schematic structural diagram of an electronic device 14 provided by an embodiment of the present application. In practical applications, based on the same inventive concept of the above embodiments, as Figure 14 shown, the electronic device 14 of this embodiment includes: a processor 141, a memory 142, and a communication bus 143. An application program code 143 is stored in the memory 142.

[0163] An embodiment of the present application provides a vehicle, including the Figure 14 aforementioned electronic device.

[0164] An embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. The above computer-readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors and are applied to a parking space positioning device in a parking scenario. The computer program implements the position accuracy evaluation method as described above.

[0165] An embodiment of the present application provides a computer program product, including a computer program or instruction. When the computer program or instruction is executed by the processor 141, it implements the position accuracy evaluation method as described above.

[0166] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a hardware embodiment, a software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program code.

[0167] This application is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, and the combination of flows and / or blocks in the flowchart and / or block diagram can also be implemented. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in one block or multiple blocks.

[0168] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in one block or multiple blocks.

[0169] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in one block or multiple blocks.

[0170] As described above, the above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed in the present application should be covered by the protection scope of the present application.

Claims

1. A method for evaluating position accuracy, characterized in that: The method comprises: Obtain trajectory data and visual data collected by the vehicle during driving; Generate vector data corresponding to the driving environment of the vehicle based on the trajectory data and the visual data; the vector data includes at least one vector element; Based on the trajectory data, the visual data, and the vector data, the position accuracy of the vector element is evaluated to obtain a position accuracy evaluation result.

2. The position accuracy evaluation method according to claim 1, characterized in that: The step of evaluating the position accuracy of the vector element based on the trajectory data, the visual data, and the vector data to obtain a position accuracy evaluation result includes: Performing feature extraction on the vector data, the trajectory data, and the visual data to obtain a plurality of feature information; Feature mapping is performed on the multiple feature information, and the position accuracy corresponding to the vector element is predicted to obtain the position accuracy evaluation result.

3. The position accuracy evaluation method according to claim 2, characterized in that: The plurality of feature information includes trajectory feature information, positioning feature information, visual feature information, and geometric feature information and semantic feature information of the vector elements; the feature extraction of the vector data, the trajectory data, and the visual data to obtain the plurality of feature information includes: Performing feature extraction on the vector data and the trajectory data to obtain the geometric feature information and the semantic feature information; Performing feature extraction on the trajectory data to obtain the trajectory feature information and the positioning feature information; Feature extraction is performed on the visual data to obtain the visual feature information.

4. The position accuracy evaluation method according to claim 3, characterized in that: The semantic feature information includes a field of view distance; the field of view distance is a distance between a center point of the vector element and a driving track of the vehicle.

5. The position accuracy evaluation method according to claim 1, characterized in that: The position accuracy evaluation result includes the accuracy evaluation result of the vector element and the position deviation; the method further includes: In response to the accuracy evaluation result of the vector element being inaccurate, adjusting the position of the vector element based on the position deviation of the vector element to obtain adjusted vector data; A map corresponding to the driving environment is constructed based on the adjusted vector data.

6. The method for evaluating position accuracy according to claim 3, characterized in that: Before performing feature mapping on the plurality of feature information, predicting the position accuracy corresponding to the vector elements, and obtaining the position accuracy evaluation result, the method further includes: Based on the trajectory feature information among the plurality of feature information, the scene of the driving environment is identified to obtain a scene category; Based on the scene category, feature mapping is performed on the multiple feature information, and the position accuracy corresponding to the vector element is predicted to obtain the position accuracy evaluation result.

7. The method for evaluating position accuracy according to claim 6, characterized in that: The track feature information includes at least one track point, each track point corresponds to a search number and a positioning state; the scene of the driving environment is identified based on the track feature information in the plurality of feature information to obtain a scene category, including: Determine a mean value of at least one of the star search numbers, and calculate a discrete distribution of at least one of the positioning states; In response to the average value of the number of satellite searches being greater than a first threshold value, and the discrete distribution of at least one of the positioning states being located in a first preset distribution set, determining that the driving environment is in an open scene; In response to the mean value of the number of satellite searches being less than a second threshold value, and a ratio of a discrete distribution of at least one of the positioning states being located in a second preset distribution set being greater than a preset ratio, it is determined that the driving environment is in a satellite denial scenario.

8. The method for evaluating position accuracy according to any one of claims 1 to 7, characterized in that: The step of evaluating the position accuracy of the vector element based on the trajectory data, the visual data, and the vector data to obtain a position accuracy evaluation result includes: The position accuracy of the vector element is evaluated by using a preset position accuracy evaluation model based on the trajectory data, the visual data, and the vector data to obtain the position accuracy evaluation result.

9. The method for evaluating position accuracy according to claim 8, characterized in that: The preset position accuracy evaluation model is provided with corresponding model parameters for different scenarios; the position accuracy of the vector element is evaluated based on the trajectory data, the visual data, and the vector data to obtain the position accuracy evaluation result, including: Determining a scene category of a driving environment of the vehicle; The preset position accuracy evaluation model is used to call the model parameters matching the scene category, and the position accuracy of the vector element is evaluated based on the trajectory data, the visual data, and the vector data to obtain the position accuracy evaluation result.

10. The method for evaluating position accuracy according to claim 8, characterized in that: Before evaluating the position accuracy of the vector element by using a preset position accuracy evaluation model based on the trajectory data, the visual data, and the vector data to obtain the position accuracy evaluation result, the method further includes: Acquire sample data; the sample data includes trajectory sample data, visual sample data, and vector sample data matching the trajectory sample data and the visual sample data; Using the position accuracy assessment model to be trained, based on the sample data, predict the sample position accuracy of the sample vector element to obtain a sample position prediction result; Based on the sample prediction results and the target prediction results corresponding to the sample data, the model parameters of the position accuracy assessment model to be trained are adjusted to obtain the preset position accuracy assessment model.

11. A position accuracy assessment device, characterized in that: The device comprises: An acquisition module, used to acquire trajectory data and visual data collected by the vehicle during driving; A generating module, configured to generate vector data corresponding to the driving environment of the vehicle based on the trajectory data and the visual data; the vector data includes at least one vector element; An evaluation module is used to evaluate the position accuracy of the vector element based on the trajectory data, the visual data, and the vector data to obtain a position accuracy evaluation result.

12. An electronic device, characterized in that: The electronic device comprises: a processor, a memory and a communication bus; when the processor executes the running program stored in the memory, the position accuracy assessment method as claimed in any one of claims 1 to 10 is implemented.

13. A vehicle, characterized in that: Comprising the electronic device as claimed in claim 12.

14. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor, the position accuracy assessment method according to any one of claims 1 to 10 is implemented.

15. A computer program product comprising a computer program or instructions, characterized in that When the computer program or instruction is executed by a processor, the position accuracy assessment method according to any one of claims 1 to 10 is implemented.

Images