A high-precision map construction method based on positioning perception fusion and related equipment

Abstract

This invention provides a method and related equipment for constructing high-precision maps based on location-aware fusion, relating to the field of high-precision map construction technology. The method includes the following steps: evaluating the reliability of location information and sensing features to obtain evaluation results; when the evaluation results are lower than a preset evaluation threshold, selecting a preset specific fusion mode; under the specific fusion mode, reducing the mutual correction strength between location information and sensing features, and determining the degree of inconsistency between location information and sensing features; when the degree of inconsistency exceeds a preset degree threshold, identifying the main error sources based on the evaluation results, and obtaining the fusion result by suppressing the influence of the main error sources; and generating high-precision map elements based on the fusion result. The method of this invention aims to solve the problem of error accumulation when both location information and sensing features are unreliable, improving the accuracy and robustness of high-precision map construction.

Description

Technical Field

[0001] This invention relates to the field of high-precision map construction technology, and more specifically, to a high-precision map construction method and related equipment based on positioning and perception fusion. Background Technology

[0002] The construction of high-precision maps relies heavily on the vehicle system's accurate determination of the vehicle's own position and its accurate perception of the surrounding environment. In routine operation, map-collecting vehicles are equipped with a combined navigation system consisting of a satellite positioning receiver, inertial measurement unit, and wheel speedometers to acquire real-time positioning information such as the vehicle's position and attitude. Simultaneously, sensing devices such as LiDAR and cameras continuously scan the road environment, identifying static elements such as lane lines, streetlights, traffic signs, guardrails, and building outlines. The backend processing system correlates the environmental features extracted by the sensing devices with the positioning information provided by the combined navigation system, accurately mapping the geometric shape and spatial position of each element to a geographic coordinate system. Through multiple rounds of data acquisition and fusion processing, a high-precision 3D map containing rich semantic information is generated. The core of this process lies in the tight coupling of positioning information and sensing features, using an information fusion framework to achieve data alignment and cross-validation, ensuring that the absolute positional accuracy of map elements reaches the centimeter level.

[0003] However, this process faces significant challenges in complex urban environments. When vehicles enter densely built-up "urban canyons" or the lower levels of multi-level overpasses, the severe obstruction of satellite signals by buildings prevents the positioning receiver from receiving a stable and sufficient amount of satellite signal. This leads to frequent interruptions or significant jumps in positioning results, with accuracy plummeting from centimeter-level to meter-level or even worse. In this situation, the integrated navigation system is forced to rely excessively on inertial measurement units and wheel speedometers for dead reckoning. The reckoning error accumulates over time and distance, causing the vehicle's actual position to gradually deviate from the system's calculated position. Simultaneously, highly reflective surfaces in the environment, such as the metal side panels of vans or the glass curtain walls of buildings, can cause specular reflection of the laser beam emitted by the lidar. Some laser beams are only captured by the receiver after one or more reflections, resulting in numerous "ghost" points in the point cloud data. For example, the guardrail of the opposite lane may be incorrectly mapped to the side of the current lane due to reflection, or the ground lane lines may appear as floating, false images due to reflection from glass curtain walls. The system simultaneously suffers from severe drift in positioning information and false distortion of perceived features. In traditional fusion mechanisms, the system attempts to correct positioning errors using perceived features. However, when the perceived features themselves contain errors—for example, mistaking a false lane line generated by reflections for a real feature—the system may incorrectly correct the positioning information towards that false feature. This correction not only fails to correct the original positioning error but also pushes the location information in an even more inaccurate direction. Subsequently, the incorrectly corrected positioning information is used as a benchmark for subsequent registration of environmental features, creating a vicious cycle of error accumulation: positioning errors lead to misjudgments of environmental features, and misjudged features further exacerbate positioning errors. Ultimately, when the vehicle leaves the area, the generated high-precision map may exhibit structural distortions, such as the creation of non-existent road boundaries or deviations of real feature positions from actual locations by tens of centimeters to several meters. Once such errors are embedded in the map data, they will pose serious safety hazards to autonomous vehicles that rely on that map.

[0004] There is currently no effective technical solution to the above problems. Summary of the Invention

[0005] The purpose of this invention is to provide a method and related equipment for constructing high-precision maps based on positioning and sensing fusion, which aims to solve the problem of error accumulation when both positioning information and sensing features are unreliable, thereby improving the accuracy and robustness of high-precision map construction.

[0006] In a first aspect, the present invention provides a method for constructing a high-precision map based on positioning and sensing fusion, comprising the following steps:

[0007] S1. Obtain location information and assess the reliability of the location information to obtain the first assessment result;

[0008] S2. Acquire the perceptual features and evaluate the reliability of the perceptual features to obtain the second evaluation result;

[0009] S3. When both the first evaluation result and the second evaluation result are lower than the corresponding preset evaluation threshold, select the preset specific fusion mode;

[0010] S4. In a specific fusion mode, reduce the mutual correction strength between positioning information and sensing features, and determine the degree of inconsistency between positioning information and sensing features. When the degree of inconsistency exceeds a preset threshold, determine the main source of error based on the first evaluation result and the second evaluation result. When fusing positioning information and sensing features based on mutual correction strength, obtain the fusion result by suppressing the influence of the main source of error.

[0011] S5. Generate high-precision map elements based on the fusion results.

[0012] The high-precision map construction method based on positioning and perception fusion provided by this invention effectively avoids the accumulation of errors caused by positioning drift and perception distortion by dynamically evaluating the reliability of positioning information and perception features, intelligently selecting fusion mode and suppressing error sources, thus improving the accuracy and robustness of high-precision map construction.

[0013] Furthermore, in step S1, the step of evaluating the reliability of the positioning information and obtaining the first evaluation result includes:

[0014] The reliability of location information is determined by assessing the degree to which it is affected by environmental factors, thus obtaining the first assessment result.

[0015] Furthermore, in step S2, the step of evaluating the reliability of the perceived features to obtain the second evaluation result includes:

[0016] The reliability of the sensing features is determined by evaluating their geometric properties, topological relationships, and the degree to which they are affected by environmental reflections, thus obtaining a second evaluation result.

[0017] Furthermore, before performing step S3, if the evaluation results of the geometric characteristics or topological relationships of the perceived feature indicate that the reliability of the perceived feature is questionable, the following steps are performed:

[0018] A1. Identify reflective surfaces in the environment and determine their geometry;

[0019] A2. Trace the origin and direction of the laser beams that correspond to the questionable sensing features;

[0020] A3. Determine the reflection path of the laser beam based on the origin, direction, and geometry of the emission.

[0021] A4. Obtain the original physical source of the perceived features;

[0022] A5. Verify the physical rationality of the reflection path and the authenticity of the original physical source, and obtain the verification results;

[0023] A6. Based on the verification results, the reliability assessment results of the perceived features are revised to obtain the revised second assessment results.

[0024] Furthermore, in step S4, the step of determining the degree of inconsistency between the location information and the perceived features includes:

[0025] B1. Calculate the instantaneous geometric deviation between the current location information and the perceived features;

[0026] B2. Monitor the dynamic changes of instantaneous geometric deviation within a preset time window to obtain information on the dynamic changes of instantaneous geometric deviation;

[0027] B3. Based on the dynamic changes in instantaneous geometric deviation, assess the fluctuation range of instantaneous geometric deviation;

[0028] B4. Determine the degree of inconsistency based on instantaneous geometric deviation and fluctuation amplitude.

[0029] Furthermore, in step S4, when the degree of inconsistency exceeds a preset threshold, the step of determining the main source of error based on the first evaluation result and the second evaluation result includes:

[0030] C1. Based on the current location information, perception features, first evaluation result, and second evaluation result, generate multiple scenario explanation hypotheses and assign an initial confidence level to each scenario explanation hypothesis;

[0031] C2. Continuously receive new sensor data and independently update the internal state of each scenario interpretation hypothesis based on the new sensor data; wherein, each scenario interpretation hypothesis independently processes the positioning information and perception features according to its own interpretation of the positioning information and perception features; the sensor data is real-time monitoring data of sensors pre-deployed on the vehicle;

[0032] C3. For each new frame of sensor data, evaluate the consistency between the new sensor data and the state predicted by each scenario interpretation hypothesis, and dynamically adjust and accumulate the confidence scores of each scenario interpretation hypothesis based on the explanatory power of each scenario interpretation hypothesis under the current environmental conditions.

[0033] C4. Continuously monitor the confidence scores of the explanatory hypotheses for each scenario. When the confidence score of an explanatory hypothesis for a scenario is significantly higher than that of all other explanatory hypotheses for the same scenario, select that explanatory hypothesis for the scenario as the final explanation. If no explanatory hypothesis for a scenario reaches the convergence condition after a preset time or distance, mark the current region as an uncertain region.

[0034] C5. Based on the selected final interpretation, determine the main sources of information that cause the inconsistency between the current location information and the perceived features, thereby identifying the main source of error.

[0035] Furthermore, in step S4, the step of obtaining the fusion result by suppressing the influence of the main error sources when fusing localization information and perceptual features based on mutual correction intensity includes:

[0036] D1. Obtain the identification results of the main error sources;

[0037] D2. Based on the identification results of the main error sources, dynamically adjust the weights of the main error sources in the fusion process of localization information and perception features, or adjust the covariance matrix of the information corresponding to the main error sources to obtain the fusion result.

[0038] Secondly, the present invention provides a high-precision map construction device based on positioning and perception fusion, comprising:

[0039] The first evaluation module is used to acquire location information, evaluate the reliability of the location information, and obtain the first evaluation result.

[0040] The second evaluation module is used to acquire perceptual features and evaluate the reliability of the perceptual features to obtain the second evaluation result.

[0041] The judgment and selection module is used to select a preset specific fusion mode when both the first evaluation result and the second evaluation result are lower than the corresponding preset evaluation threshold.

[0042] The fusion module is used to reduce the mutual correction intensity between positioning information and sensing features under a specific fusion mode, and to determine the degree of inconsistency between positioning information and sensing features. When the degree of inconsistency exceeds a preset threshold, the main source of error is determined based on the first evaluation result and the second evaluation result. When fusing positioning information and sensing features based on mutual correction intensity, the fusion result is obtained by suppressing the influence of the main source of error.

[0043] The generation module is used to generate high-precision map features based on the fusion results.

[0044] The high-precision map construction device based on positioning and perception fusion provided by this invention effectively blocks the mutual reinforcement cycle of erroneous information when the reliability of positioning information and perception features is reduced due to environmental interference, thus preventing the distortion of high-precision map elements and ensuring the accuracy and reliability of map construction results.

[0045] Thirdly, the present invention provides an electronic device including a processor and a memory, the memory storing computer-readable instructions, which, when executed by the processor, perform the steps of the high-precision map construction method based on positioning-aware fusion provided in the first aspect above.

[0046] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the steps of the high-precision map construction method based on location-aware fusion provided in the first aspect above.

[0047] As can be seen from the above, the high-precision map construction method based on positioning and perception fusion provided by this invention effectively solves the problems of interruption and jump in positioning information and "ghosting" of perception features faced by high-precision map construction in complex urban environments (such as "urban canyons" or the lower level of multi-level overpasses). Existing methods may lead to mutual correction of erroneous information when both positioning and perception information are problematic, forming a vicious cycle and ultimately generating severely distorted map data. This application avoids blindly correcting each other when both are unreliable by independently evaluating the reliability of positioning information and perception features and dynamically adjusting the fusion strategy based on the evaluation results. More importantly, this application introduces a judgment mechanism for the main source of error, which can accurately identify the root cause of inconsistency and ensure the accuracy and robustness of the fusion result by suppressing the influence of this error source. Accordingly, this application can significantly improve the construction accuracy and reliability of high-precision maps, effectively avoid map data distortion, and thus provide a safer and more reliable map foundation for autonomous vehicles, overcoming the safety risks existing in the prior art.

[0048] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing embodiments of the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings. Attached Figure Description

[0049] Figure 1 This is a flowchart of a high-precision map construction method based on positioning and perception fusion provided in an embodiment of the present invention.

[0050] Figure 2 This is a schematic diagram of a high-precision map construction device based on positioning and perception fusion provided in an embodiment of the present invention.

[0051] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0052] Label Explanation:

[0053] 100. First evaluation module; 200. Second evaluation module; 300. Judgment and selection module; 400. Fusion module; 500. Generation module; 13. Electronic device; 1301. Processor; 1302. Memory; 1303. Communication bus. Detailed Implementation

[0054] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0055] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0056] Please refer to Figure 1 , Figure 1 This is a flowchart of a high-precision map construction method based on location-aware fusion. The method includes the following steps:

[0057] S1. Obtain location information and assess the reliability of the location information to obtain a first assessment result. The specific steps include: determining the reliability of the location information by assessing the degree to which the location information is affected by environmental factors, and obtaining a first assessment result.

[0058] S2. Acquire sensing features and evaluate the reliability of the sensing features to obtain a second evaluation result. The specific steps include: determining the reliability of the sensing features by evaluating the geometric characteristics, topological relationships, and degree of influence of environmental reflections of the sensing features, and obtaining a second evaluation result.

[0059] S3. Based on the first evaluation result and the second evaluation result, determine the fusion mode of the positioning information and the perception feature, and select the preset specific fusion mode when both the first evaluation result and the second evaluation result are lower than the corresponding preset evaluation threshold;

[0060] S4. Under a specific fusion mode, reduce the mutual correction strength between location information and sensing features (mutual correction strength refers to the weight or influence of one information source on another during information fusion. This strength is not fixed but dynamically changes, and it is usually closely related to factors such as the quality, confidence, uncertainty, or error covariance of the information source itself. For example, when location information has high accuracy and low uncertainty, its correction strength on sensing features may be higher; conversely, when sensing features provide very clear and highly confident environmental cues, their correction strength on location information will also increase accordingly), and determine the degree of inconsistency between location information and sensing features. When the degree of inconsistency exceeds a preset threshold, determine the main source of error based on the first and second evaluation results, and obtain the fusion result by suppressing the influence of the main source of error when fusing location information and sensing features based on mutual correction strength.

[0061] S5. Generate high-precision map elements based on the fusion results.

[0062] This application relates to the field of high-precision map construction technology, and addresses the problem of map element distortion caused by the decreased reliability of positioning information and perception features in complex environments. It provides a high-precision map construction method based on positioning and perception fusion. Specifically, the reliability of positioning information can be assessed by analyzing the consistency of positioning data over time, for example, by calculating the displacement change rate between consecutive positioning points; when the change rate exceeds a preset threshold, the reliability is considered reduced. Furthermore, this assessment process can also be achieved through cross-validation of multi-source positioning data, such as comparing the degree of difference between the position information output by satellite positioning receivers and inertial measurement units; the greater the difference, the lower the reliability. In particular, the reliability of perception features can be assessed by the frequency of recurrence of statistical features in multiple scans; for example, by continuously scanning lane line features, if the frequency of a specific feature in a preset number of scans is lower than a threshold, the reliability is considered reduced. As a preferred implementation, this assessment can also be achieved through matching degree analysis of perception features with other sensor data, for example, by correlating and verifying corresponding features in LiDAR point cloud data and camera images; the lower the matching degree, the lower the reliability. Therefore, when determining the fusion mode of positioning information and sensing features, if both the first and second evaluation results are lower than a preset evaluation threshold, a specific fusion mode is selected to avoid the conventional strong correction mechanism. Specifically, reducing the mutual correction strength between positioning information and sensing features can be achieved by scaling the correction amount using a pre-set attenuation coefficient, such as multiplying the correction amount by a fixed coefficient of 0.5. Furthermore, determining the degree of inconsistency between positioning information and sensing features can be achieved by calculating the Euclidean distance deviation between the two in the spatial coordinate system, for example, by measuring the straight-line distance between the vehicle position indicated by the positioning information and the lane center point identified by the sensing features. In particular, determining the main source of error can be achieved by directly comparing the numerical values ​​of the first and second evaluation results; if the first evaluation result is significantly lower than the second evaluation result, the positioning information is determined to be the main source of error. As a preferred implementation, this determination can also be based on trend analysis of historical reliability data; for example, if the reliability evaluation result of the positioning information continuously declines over a continuous period, it is determined to be the main source of error. Therefore, when fusing positioning information and sensing features based on mutual correction intensity, the influence of the main error source can be suppressed by dynamically adjusting data weights. For example, when positioning information is identified as the main error source, its weight in the fusion calculation is reduced to 30% of the original weight. Furthermore, this suppression process can also be achieved by modifying the data confidence parameter, for example, assigning a lower confidence value to the information corresponding to the main error source. This embodiment effectively blocks the mutual reinforcement path of error information when both positioning information and sensing features are affected by environmental interference by dynamically evaluating the reliability of the two sources and implementing an adaptive fusion strategy. This avoids the problem of map feature distortion caused by error correction, and is particularly suitable for high-precision map construction scenarios in urban canyons or strong reflection environments.Therefore, the high-precision map elements generated based on the fusion results can accurately reflect the characteristics of the real environment, providing a reliable data foundation for subsequent map applications.

[0063] This high-precision map construction method based on positioning and sensing fusion achieves accurate generation of high-precision map elements through a systematic process. Positioning information is acquired, and its reliability is determined by assessing the degree to which it is affected by environmental factors, yielding the first assessment result. Specifically, firstly, it is necessary to identify and quantify the main environmental factors affecting positioning accuracy. These factors include, but are not limited to: satellite signal obstruction (such as in urban canyons with tall buildings or dense forests), multipath effects (signals are reflected by buildings or other objects, resulting in signals received from multiple paths), atmospheric delay (the influence of the ionosphere and troposphere on signal propagation speed), and potential signal interference (such as radio frequency interference or spoofing).

[0064] To assess the impact of these environmental factors, various metrics provided by the positioning receiver can be utilized. For example, by monitoring the carrier-to-noise ratio (C / N0) or signal-to-noise ratio (SNR) of satellite signals, the quality and strength of the signal can be determined; a lower C / N0 usually indicates signal obstruction or interference. Simultaneously, geometrical accuracy factors (DOPs, such as PDOP, HDOP, VDOP, etc.) are analyzed. These factors reflect the impact of satellite geometry on positioning accuracy; a high DOP value indicates poor satellite distribution and potentially larger positioning errors. Furthermore, the magnitude of errors caused by environmental factors can be directly assessed by analyzing the volatility of positioning results or deviations from a known reference point (if available). In some cases, data from the receiver's internal inertial measurement unit (IMU) or other sensors can also aid in determining environmental changes; for example, positioning accuracy may be affected when the equipment is in rapid movement or experiencing severe vibration.

[0065] After quantifying the impact of environmental factors, the next step is to determine the reliability of the location information based on preset reliability standards or thresholds. These standards can be set based on historical data, experimental results, or specific application requirements. For example, a minimum C / N0 threshold can be set, below which signals are considered unreliable; or a maximum permissible DOP value can be set, exceeding which positioning results are considered insufficiently accurate. By comparing the currently evaluated indicators with these thresholds, it can be determined whether the current location information meets the reliability requirements.

[0066] Finally, through the comprehensive evaluation process described above, a first evaluation result can be obtained. This result is typically a preliminary judgment on the reliability of the location information. It can be a qualitative description (such as "high reliability," "medium reliability," or "low reliability") or a quantitative indicator (such as a reliability score or confidence percentage). This first evaluation result provides important reference for subsequent location data processing, fusion, or application, helping the system decide whether to adopt the current location information, or whether it is necessary to switch to other location modes or perform error correction.

[0067] Sensing features are acquired, and their reliability is determined by evaluating their geometric characteristics, topological relationships, and the degree to which they are affected by environmental reflections, resulting in a second evaluation result. Specifically, sensing features typically refer to key information points, lines, surfaces, or regions extracted from sensor data to describe the environment or target. To evaluate the reliability of these features, in-depth analysis can be conducted from the following aspects:

[0068] First, assess the geometric properties of the perceived features. This includes analyzing attributes such as the feature's shape, size, location, and orientation. For example, a reliable feature should maintain relatively stable and consistent geometric properties when perceived from different viewpoints or at different times. If a feature's geometry changes frequently, its size fluctuates significantly, or its location drifts, its reliability decreases. By calculating statistics (such as mean and variance) of these geometric properties or performing geometric consistency tests, their stability can be quantified, thus providing a preliminary assessment of the feature's reliability.

[0069] Second, assess the topological relationships of perceived features. Topological relationships describe non-metric properties such as spatial connections, adjacencies, and inclusions between features. For example, if a feature is always adjacent to another specific feature, or always contained within a larger feature, and this relationship remains unchanged under different perceptual conditions, then the topological relationships of these features are stable. By analyzing the connection patterns of nodes and edges in the feature graph, and the evolution of these patterns over time, the structural stability of the feature group can be determined. If topological relationships are frequently broken or reconstructed, it indicates low reliability of the relevant features, and there may be false positives or false negatives.

[0070] Third, assess the degree to which perceived features are affected by environmental reflections. Environmental reflections, especially specular or multipath reflections, can produce false features or distort the geometry and intensity information of true features in sensor data. For example, in lidar data, smooth surfaces may produce strong reflection points, but these points may not represent the actual object surface; in image data, reflections from water or glass may cause ghosting or positional shifts in objects. The impact of reflections on features can be identified and quantified by analyzing the intensity distribution and texture information of features or by cross-validating using multi-sensor fusion techniques. If the appearance or disappearance of a feature is highly correlated with reflection phenomena, or if its properties exhibit anomalies within the reflection area, its reliability will be significantly reduced.

[0071] Based on the above assessments, by quantitatively analyzing the consistency of geometric characteristics, the stability of topological relationships, and the degree of influence from environmental reflections, a reliability metric or confidence level can be assigned to each perceived feature or feature set. This metric is the "second evaluation result," reflecting the reliability of the feature under the current environmental and perception conditions. This result can then be used to guide subsequent decision-making processes, such as filtering low-reliability features in target tracking, removing unstable feature points in map building, or assigning higher weights to high-reliability features in behavior prediction, thereby improving the robustness and accuracy of the entire perception system.

[0072] Specifically, when both the first and second evaluation results are below their corresponding preset evaluation thresholds, a specific fusion mode is selected. Further, under this specific fusion mode, the mutual correction strength between positioning information and perceived features is reduced, and the instantaneous geometric deviation between the two and its dynamic changes within a preset time window are calculated to determine the degree of inconsistency. When the degree of inconsistency exceeds a preset threshold, the main source of error is identified, and the fusion result is obtained by suppressing the influence of the main source of error during the fusion process. Based on this fusion result, high-precision map elements are generated. This method effectively blocks the mutual reinforcement cycle of erroneous information, prevents distortion of high-precision map elements, and ensures the accuracy and reliability of the map construction results, even when the reliability of both positioning information and perceived features is reduced due to environmental interference.

[0073] In some embodiments, before performing step S3, if the evaluation results of the geometric characteristics or topological relationships of the sensed feature indicate that the reliability of the sensed feature is questionable, the following steps are performed:

[0074] A1. Identify reflective surfaces in the environment and determine their geometry;

[0075] A2. Trace the origin and direction of the laser beams that correspond to the questionable sensing features;

[0076] A3. Determine the reflection path of the laser beam based on the origin, direction, and geometry of the emission.

[0077] A4. Obtain the original physical source of the perceived features;

[0078] A5. Verify the physical rationality of the reflection path and the authenticity of the original physical source, and obtain the verification results;

[0079] A6. Based on the verification results, the reliability assessment results of the perceived features are revised to obtain the revised second assessment results.

[0080] Specifically, identifying reflective surfaces in the environment refers to detecting surfaces that may cause specular reflection. This can be achieved using deep learning-based image segmentation algorithms or point cloud curvature analysis techniques, aiming to provide key geometric parameters for reflection path calculation. Tracing the origin and direction of the laser beam's emission refers to determining the specific spatial location and initial propagation direction of the laser emitted by the lidar. This can be achieved through the fusion calculation of sensor calibration parameters and real-time vehicle pose information, aiming to accurately locate the emission reference point of the perception system. In practical applications, determining the laser beam's reflection path refers to simulating the laser's propagation trajectory in complex environments. This can be achieved using ray tracing algorithms combined with a parameterized model of the reflective surface geometry, aiming to verify whether questionable features might be generated by reflection. Specifically, obtaining the original physical source of the perceived features is... Tracing the real object entity corresponding to the tracing feature can be achieved using multi-frame point cloud data association or cross-sensor data matching technology, with the aim of distinguishing between directly perceived features and false features generated by reflection. Among them, the physical rationality test of the reflection path refers to verifying whether the path conforms to the optical reflection law, which can be achieved based on the geometric constraints of the incident angle and the reflection angle. The authenticity test of the original physical source refers to confirming the actual existence of the source object in the environment, which can be achieved by fusing semantic information from camera images for cross-verification, with the aim of establishing a dual verification mechanism to ensure the reliability of the analysis. In practical applications, correcting the reliability assessment result of the perceived feature refers to dynamically adjusting the assessment value, which can be achieved using a Bayesian probability model or a confidence weighting method, with the aim of making the assessment result more accurately reflect the true reliability of the perceived feature.

[0081] Specifically, when the evaluation results of the geometric characteristics or topological relationships of a perceived feature indicate that its reliability is questionable, the system first identifies reflective surfaces in the environment and determines their geometry, providing basic parameters for reflection path calculation. Then, it traces the origin and direction of the laser beam emission corresponding to the questionable perceived feature to precisely locate the emission position of the sensing system. Based on the emission origin, direction, and the geometry of the reflective surface, the system determines the reflection path of the laser beam and simulates the actual propagation trajectory to verify whether the questionable feature could be generated by reflection. Next, it obtains the original physical source of the perceived feature, tracing its true object origin. Then, it verifies the physical rationality of the reflection path and the authenticity of the original physical source, ensuring the reliability of the analysis through a dual verification mechanism. Finally, it corrects the reliability evaluation results of the perceived feature based on the verification results, obtaining a corrected second evaluation result. This series of steps is executed sequentially, with information flowing between steps, forming a closed-loop verification mechanism. This ensures that when the reliability of a perceived feature is questionable, false features caused by environmental reflection can be accurately identified and corrected, avoiding deviations in subsequent fusion processes due to erroneous evaluations.

[0082] As a specific implementation method, considering a map-collecting vehicle traveling in an urban canyon section, when the reliability of the perceived feature assessment is questionable, the system identifies a large glass curtain wall on one side of the road as a reflective surface and determines its planar geometry. It traces the origin of the laser beam corresponding to the questionable perceived feature to the location of the vehicle-mounted lidar, with the direction being horizontal and forward. Based on the origin, direction, and planar shape of the glass curtain wall, the system determines the path of the laser beam after one reflection to reach the receiver. It obtains the original physical source of the perceived feature, discovering that it actually originates from the guardrail of the opposite lane. The physical rationality of the reflection path is checked, confirming that the path conforms to the law of reflection, and the authenticity of the original physical source is verified through camera data. Based on the verification results, the system corrects the reliability assessment result of the perceived feature, increasing the originally low reliability assessment value and avoiding misjudging the reflected feature as a real element.

[0083] Through the above scheme, this application can accurately identify and process false perception features caused by environmental reflections, correct the reliability assessment results, and thus provide a reliable basis for the subsequent fusion of positioning information and perception features. It effectively avoids position deviation and feature distortion caused by erroneous assessment, and improves the accuracy and reliability of high-precision map construction.

[0084] In some embodiments, step S4, determining the degree of inconsistency between the location information and the perceived features, includes:

[0085] B1. Calculate the instantaneous geometric deviation between the current location information and the perceived features;

[0086] B2. Monitor the dynamic changes of instantaneous geometric deviation within a preset time window to obtain information on the dynamic changes of instantaneous geometric deviation;

[0087] B3. Based on the dynamic changes in instantaneous geometric deviation, assess the fluctuation range of instantaneous geometric deviation;

[0088] B4. Determine the degree of inconsistency based on instantaneous geometric deviation and fluctuation amplitude.

[0089] Instantaneous geometric deviation refers to the instantaneous difference between the current positioning information and the perceived features in spatial location. It can be achieved using spatial measurement methods such as Euclidean distance, Manhattan distance, or Chebyshev distance. Its purpose is to capture the instantaneous mismatch between positioning and perception in spatial location. The preset time window refers to the time period used to analyze the dynamic changes of instantaneous geometric deviation. It can be a fixed duration or a time window that is dynamically adjusted according to environmental conditions (for example, when the environment fluctuates drastically, a rapid response is required, or high data accuracy is required, the time window may be shortened to increase the sampling frequency or processing real-time performance; while when the environment is stable, resources are limited, or energy consumption needs to be saved, the time window may be extended to reduce the processing burden or extend battery life). Its purpose is to limit the observation range of dynamic changes and avoid computational redundancy caused by excessively long historical data. Dynamic change information refers to the evolution trajectory of instantaneous geometric deviation within a preset time window. It can be represented by time series data or the rate of change of deviation, aiming to reflect the continuous trend of deviation change. Fluctuation amplitude refers to the stability index of instantaneous geometric deviation within the time window. It can be assessed using statistical measures such as standard deviation, variance, or range, aiming to quantify the degree of deviation fluctuation and distinguish between random noise and systematic errors. Inconsistency degree refers to the reliability index derived from a combination of instantaneous geometric deviation and fluctuation amplitude. It can be determined using weighted combination or fuzzy logic methods, aiming to provide a comprehensive quantitative value of inconsistency for subsequent error source identification. Specifically, regarding the weighted combination method, the basic idea is to assign different weights to instantaneous geometric deviation and fluctuation amplitude to reflect their relative importance in determining the degree of inconsistency. The specific operation steps are as follows:

[0090] 1. First, it is necessary to obtain the specific values ​​of instantaneous geometric deviation and fluctuation amplitude. Since these two indicators may have different dimensions or numerical ranges, in order to ensure their comparability when combined, they usually need to be standardized, for example, normalized to the range of 0 to 1.

[0091] 2. Determine the weights (w1 and w2) of instantaneous geometric deviation and fluctuation amplitude based on system characteristics, historical data analysis, expert experience, or specific application scenarios. These weights reflect the degree of influence of each factor on the final degree of inconsistency. For example, if instantaneous geometric deviation is considered to have a greater impact on reliability, its weight w1 will be higher than the weight w2 of fluctuation amplitude, and it is generally required that the sum of all weights is 1 (i.e., w1 + w2 = 1).

[0092] 3. Multiply the standardized instantaneous geometric deviation value by its corresponding weight w1, and multiply the standardized volatility value by its corresponding weight w2. Then add these two products together to obtain the final degree of inconsistency. The calculation formula can be expressed as: Degree of inconsistency = (w1 × standardized instantaneous geometric deviation) + (w2 × standardized volatility). This method provides a single numerical value that intuitively reflects the overall level of inconsistency.

[0093] The fuzzy logic method offers an effective approach to handling uncertainty and fuzzy information, particularly suitable for situations where the relationship between instantaneous geometric deviations and fluctuation amplitudes and the degree of inconsistency is difficult to describe with precise mathematical models. Its process for determining the degree of inconsistency typically includes the following three main stages:

[0094] 1. Convert the precise numerical values ​​of instantaneous geometric deviations and fluctuation amplitudes (i.e., "crystal" inputs) into fuzzy sets (i.e., "fuzzy" inputs). This requires defining a series of linguistic variables (such as "small," "medium," and "large") and their corresponding membership functions for each input variable. The membership function describes the degree to which an input value belongs to a certain fuzzy set. For example, a numerical value of an instantaneous geometric deviation might belong to the "medium deviation" set with a membership degree of 0.8, while belonging to the "large deviation" set with a membership degree of 0.2.

[0095] 2. Based on a predefined fuzzy rule base, inference is performed on the fuzzy input. Fuzzy rules are typically expressed in the form of "IF-THEN", for example: "If the instantaneous geometric deviation is 'large' and the fluctuation amplitude is 'large', then the degree of inconsistency is 'very high'". These rules are usually established by domain experts based on their empirical knowledge. The inference mechanism activates the corresponding rules according to the membership degree of the input value and calculates the activation strength of each rule, thereby generating a fuzzy output.

[0096] 3. Convert the fuzzy output (a fuzzy set representing a fuzzy concept indicating the degree of inconsistency) obtained from fuzzy inference back into a clear numerical value. Commonly used defuzzification methods include the centroid method, the median method, and the maximum membership degree average method. Through defuzzification, a specific numerical value is finally obtained, representing the degree of inconsistency after considering both instantaneous geometric deviation and fluctuation amplitude.

[0097] Specifically, the proposed solution first calculates the instantaneous geometric deviation at the current moment through step B1 to obtain the immediate difference between positioning and sensing in spatial location; then, step B2 monitors the dynamic changes of this deviation within a preset time window to obtain continuous evolution information of the deviation; next, step B3 evaluates the fluctuation amplitude based on the dynamic change information to quantify the stability characteristics of the deviation; finally, step B4 combines the magnitude of the instantaneous geometric deviation and its fluctuation amplitude to determine the degree of inconsistency. This design, by introducing a dynamic evaluation mechanism with a time dimension, can effectively distinguish between genuine inconsistency and random noise, avoiding misjudging instantaneous interference as systematic errors, and thus accurately identifying false consistency when both positioning and sensing have related errors.

[0098] As a specific implementation method, when a map-collecting vehicle travels through an urban canyon area filled with high-rise buildings, satellite positioning signals may be obstructed, causing positioning information drift. Simultaneously, glass curtain walls can cause LiDAR reflections, generating false sensing features. In this scenario, the system calculates the instantaneous geometric deviation between the current positioning point and the sensing feature point; monitors the continuous trend of this deviation within a preset time window; assesses the fluctuation amplitude of the deviation, for example, by calculating the standard deviation to quantify stability; finally, based on the magnitude of the instantaneous geometric deviation and its fluctuation amplitude, it determines the degree of inconsistency, thereby identifying false consistency caused by environmental reflections and preventing the system from misclassifying erroneous features as reliable information.

[0099] Through the above scheme, this application can accurately quantify the degree of inconsistency when both positioning information and perception features are erroneous and exhibit false consistency, effectively preventing the system from misjudging information as reliable and making wrong decisions, and ensuring accurate identification and suppression of major error sources during the construction of high-precision maps.

[0100] In some embodiments, step S4, when the degree of inconsistency exceeds a preset threshold, involves determining the primary source of error based on the first evaluation result and the second evaluation result, including:

[0101] C1. Based on the current location information, perception features, first evaluation result, and second evaluation result, generate multiple scenario explanation hypotheses and assign an initial confidence level to each scenario explanation hypothesis;

[0102] C2. Continuously receive new sensor data and independently update the internal state of each scenario interpretation hypothesis based on the new sensor data; wherein, each scenario interpretation hypothesis independently processes the positioning information and perception features according to its own interpretation of the positioning information and perception features; the sensor data is real-time monitoring data of sensors (such as LiDAR, millimeter-wave radar, vision sensors and other sensors) pre-deployed on the vehicle (such as point cloud data continuously output by LiDAR, and image sequences continuously output by vision sensors).

[0103] C3. For each new frame of sensor data, evaluate the consistency between the new sensor data and the state predicted by each scenario interpretation hypothesis, and dynamically adjust and accumulate the confidence scores of each scenario interpretation hypothesis based on the explanatory power of each scenario interpretation hypothesis under the current environmental conditions.

[0104] C4. Continuously monitor the confidence scores of the explanatory hypotheses for each scenario. When the confidence score of an explanatory hypothesis for a scenario is significantly higher than that of all other scenario explanatory hypotheses, select that scenario explanatory hypothesis as the final explanation (capable of identifying and establishing an explanatory framework that best fits the current data or target context from a series of preset or dynamically generated scenarios and their associated hypotheses); if no scenario explanatory hypothesis reaches the convergence condition after a preset time or distance, mark the current region as an uncertain region (firstly, when a region is marked as uncertain, it usually means that the system's current understanding of that region is vague, contradictory, or lacks sufficient information. If these uncertainties are not specially handled and are directly included in the subsequent routine processing flow, This could potentially lead to the propagation of erroneous information, thereby reducing the accuracy and reliability of the final output of the entire system. For example, in image recognition or target tracking, if the features of a certain region are not obvious or are occluded, marking it as an uncertain region can prevent the system from making incorrect judgments based on insufficient information, thus providing a clear target for subsequent refined processing. Secondly, marking uncertain regions can guide the system or operator to allocate more computational resources or adopt more complex processing strategies for these regions in subsequent steps. For regions with high certainty, standard and efficient algorithms can be used for rapid processing; while for uncertain regions, processing such as re-evaluation and verification, manual intervention, or information supplementation may be required.

[0105] C5. Based on the selected final interpretation, determine the main sources of information that cause the inconsistency between the current location information and the perceived features, thereby identifying the main source of error.

[0106] Specifically, in step C1, generating multiple scenario explanation hypotheses involves constructing a series of potential cause models that may lead to the current inconsistency, based on the available location information, perception features, and their respective reliability assessment results (i.e., the first assessment result and the second assessment result). For example, these hypotheses may include "incorrect location information," "incorrect perception features," or "environmental anomalies causing deviations in both." An initial confidence level is assigned to each scenario explanation hypothesis, which can be based on prior knowledge or a uniform distribution, representing the probability that each hypothesis is true before any new evidence is available.

[0107] In step C2, continuously receiving new sensor data means that the system continuously acquires real-time data streams from positioning sensors (such as GNSS, IMU) and perception sensors (such as LiDAR, cameras) on the vehicle or other platforms. Independently updating the internal state of each scenario interpretation hypothesis means that each hypothesis model independently processes and integrates this new sensor data based on its specific interpretation of positioning information and perception features to predict its system state under the current hypothesis. For example, if a hypothesis considers the positioning information to be incorrect, it may attempt to use perception features to correct the positioning information and predict the vehicle's attitude and position; conversely, if a hypothesis considers the perception features to be incorrect, it may rely more heavily on positioning information to predict the expected performance of the perception features.

[0108] In step C3, for each new frame of sensor data, the consistency between the new sensor data and the predicted state of each scene explanation hypothesis is evaluated. This involves comparing the actually received sensor data with the data predicted by each hypothesis model based on its internal state, quantifying the degree of matching between the two. The confidence scores of each scene explanation hypothesis are dynamically adjusted and accumulated based on its explanatory power under the current environmental conditions. This means that, based on the consistency evaluation results and the adaptability and explanatory power of the hypothesis model to the current environment (e.g., whether there is occlusion, reflection, etc.), the confidence scores of each hypothesis are updated in real time using Bayesian inference, particle filtering, or other probabilistic update methods. Hypotheses that better match the actual data and provide a more reasonable explanation of the environment will have their confidence scores increased; conversely, their confidence scores will decrease.

[0109] In step C4, continuously monitoring the confidence scores of each scenario explanation hypothesis means that the system continuously tracks the trend of confidence changes for each hypothesis. When the confidence score of a scenario explanation hypothesis is significantly higher than that of all other scenario explanation hypotheses—for example, when its confidence score exceeds a certain preset threshold and there is a significant difference in confidence scores compared to other hypotheses—this hypothesis is selected as the final explanation for the current scenario. If, after a preset time or distance, the confidence scores of all hypotheses still have not reached the convergence condition (i.e., no hypothesis can clearly dominate), it indicates that there is high uncertainty in the current region. The system will mark this region as an uncertain region, which may require a more conservative strategy or manual intervention.

[0110] In step C5, based on the selected final interpretation, the primary source of information causing the inconsistency between the current location information and the perceived features is determined. This means that once a scenario interpretation hypothesis is confirmed as the final interpretation, the system explicitly indicates whether the inconsistency is caused by location information, perceived features, or other factors, based on the error type represented by that hypothesis. For example, if the final interpretation is "location information error," then the primary source of error is determined to be the location information.

[0111] Specifically, the proposed solution first generates multiple scenario interpretation hypotheses based on positioning information, sensing features, and reliability assessment results, assigning initial confidence levels to establish a multi-faceted explanation basis for inconsistencies. Then, it continuously receives new sensor data, allowing each scenario interpretation hypothesis to independently update its internal state, ensuring each hypothesis evolves autonomously based on its own interpretation logic. Next, for each new data frame, it evaluates its consistency with the predicted states of each hypothesis and dynamically adjusts the confidence score based on the interpretability under current environmental conditions, achieving refined accumulation of hypothesis credibility. Based on this, it continuously monitors changes in confidence scores, selecting a hypothesis that significantly leads as the final explanation; otherwise, it marks uncertain areas. Finally, it identifies the main source of error based on the selected explanation. This process, through a multi-hypothesis parallel verification mechanism, uses the reliability assessment results as initial weights and continuously verifies the physical rationality of each hypothesis using new sensor data, avoiding misjudgments caused by relying on static assessments. Especially in scenarios where satellite signal interruptions and reflection interference coexist, it effectively distinguishes between positioning drift and sensing distortion, preventing the system from falling into a vicious cycle of correction.

[0112] As a specific implementation method, the solution of this application is implemented as follows: When the vehicle travels to a commercial street with dense glass curtain walls, the system detects that the inconsistency between the positioning information and the perceived features exceeds a preset threshold, and then generates three scenario explanation hypotheses, including "cumulative positioning system error," "glass curtain wall reflection interference," and "double error superposition." As the vehicle continues to travel, the system continuously receives LiDAR point cloud and camera image data, and each hypothesis updates its state independently: for example, the "glass curtain wall reflection interference" hypothesis traces the laser beam reflection path based on the curtain wall geometry to verify the physical source of the point cloud features; the "cumulative positioning system error" hypothesis calculates the position drift trend based on inertial measurement unit data. For each newly acquired frame of sensor data, the system evaluates its consistency with the predicted state of each hypothesis. For example, when a floating lane line feature matching the curtain wall reflection path appears in the point cloud, the confidence score of the "glass curtain wall reflection interference" hypothesis is significantly improved due to its superior explanatory power. After a preset time window of data accumulation, if the confidence score of this hypothesis continues to lead, the system determines that the perceived features are the main source of error and marks the area as a high-reflection risk zone.

[0113] Through the above technical solution, this application can accurately identify the main sources of error in complex scenarios where the reliability of both positioning information and perception features is in the low trust range and there are significant inconsistencies. This is achieved through a multi-hypothesis dynamic verification mechanism, which avoids the acceptance or isolation of erroneous information sources due to misjudgment. This effectively suppresses the vicious cycle of correction caused by positioning drift and perception distortion, and ensures the accuracy of high-precision map element generation.

[0114] In some embodiments, step S4, when fusing localization information and sensing features based on mutual correction intensity, includes obtaining the fusion result by suppressing the influence of major error sources.

[0115] D1. Obtain the identification results of the main error sources;

[0116] D2. Based on the identification results of the main error sources, dynamically adjust the weight of the main error sources in the fusion process of localization information and perception features, or adjust the covariance matrix of the information corresponding to the main error sources, thereby reducing the contribution of the main error sources to the fusion result and obtaining the fusion result.

[0117] Specifically, identifying the primary source of error means determining which aspect of the location information or perceived features is the main source of inconsistency. This can be achieved using a confidence-based assessment mechanism, such as analyzing historical reliability records of sensor data or environmental interference indicators. Dynamically adjusting weights can be understood as modifying the influence of the primary source of error in the fusion calculation in real time. This can be achieved using weight coefficient update strategies in adaptive filtering algorithms, such as dynamically scaling the weight coefficients based on the confidence level of the error source. Adjusting the covariance matrix specifically refers to modifying the covariance parameter representing the uncertainty of the primary source of error data. This can be achieved using covariance scaling methods within the Kalman filtering framework, such as increasing the diagonal elements of the covariance matrix to quantify the uncertainty of the data source.

[0118] Specifically, the proposed solution first identifies the main sources of error, transforming abstract error judgments into actionable targets, thus providing a clear operational basis for the suppression mechanism. Based on this, weights or the covariance matrix are dynamically adjusted according to the identification results, weakening the impact of the main error sources in real time. Dynamically adjusting the weights directly reduces the numerical contribution of erroneous data in the fusion calculation, allowing reliable information to dominate the decision-making process; while adjusting the covariance matrix quantifies the uncertainty of the error sources, naturally weakening their contribution within the filtering algorithm framework. These two mechanisms complement each other, ensuring a high degree of matching between suppression measures and error types, while maintaining the system's full utilization of effective information, effectively blocking the transmission and amplification of erroneous information in the fusion process, and ultimately achieving high-precision fusion results.

[0119] As a specific implementation method, the solution of this application is implemented as follows: When the system detects that the inconsistency between the positioning information and the perceived features exceeds a preset threshold, it first identifies the main source of error as the perceived features (e.g., false lane line features caused by reflection from glass curtain walls) through a scene interpretation hypothesis mechanism. Subsequently, during the fusion process, the weight of the perceived features is dynamically adjusted from the default value to a lower level, or the diagonal element values ​​of its covariance matrix are increased, thereby significantly reducing the influence of the perceived features on the fusion result, allowing the positioning information to dominate the fusion process, and avoiding the incorrect correction of the location information.

[0120] Through the above scheme, this application can accurately identify the main sources of error and dynamically suppress their impact, effectively preventing the vicious cycle of location information being incorrectly corrected by false features, thereby ensuring the accuracy of the generated high-precision map elements in terms of geometric location and semantic description, and avoiding structural distortion of map data.

[0121] Please refer to Figure 2 , Figure 2 This invention provides a high-precision map construction device based on location-aware fusion in some embodiments. The device is integrated into a back-end control device in the form of a computer program, and includes:

[0122] The first evaluation module 100 is used to acquire location information, evaluate the reliability of the location information, and obtain a first evaluation result.

[0123] The second evaluation module 200 is used to acquire perceptual features and evaluate the reliability of the perceptual features to obtain a second evaluation result.

[0124] The judgment and selection module 300 is used to select a preset specific fusion mode when both the first evaluation result and the second evaluation result are lower than the corresponding preset evaluation threshold.

[0125] The fusion module 400 is used to reduce the mutual correction intensity between positioning information and sensing features in a specific fusion mode, and to determine the degree of inconsistency between positioning information and sensing features. When the degree of inconsistency exceeds a preset threshold, the main source of error is determined based on the first evaluation result and the second evaluation result. When fusing positioning information and sensing features based on mutual correction intensity, the fusion result is obtained by suppressing the influence of the main source of error.

[0126] The generation module 500 is used to generate high-precision map features based on the fusion results.

[0127] In some embodiments, the first evaluation module 100 performs the following when acquiring location information, evaluating the reliability of the location information, and obtaining a first evaluation result:

[0128] The reliability of location information is determined by assessing the degree to which it is affected by environmental factors, thus obtaining the first assessment result.

[0129] In some embodiments, the second evaluation module 200 is executed when acquiring perceived features, evaluating the reliability of the perceived features, and obtaining a second evaluation result:

[0130] The reliability of the sensing features is determined by evaluating their geometric properties, topological relationships, and the degree to which they are affected by environmental reflections, thus obtaining a second evaluation result.

[0131] In some embodiments, a correction module is also included, which, before the selection module 300 operates, performs the following steps when the evaluation results of the geometric characteristics or topological relationships of the sensed feature indicate that the reliability of the sensed feature is questionable:

[0132] A1. Identify reflective surfaces in the environment and determine their geometry;

[0133] A2. Trace the origin and direction of the laser beams that correspond to the questionable sensing features;

[0134] A3. Determine the reflection path of the laser beam based on the origin, direction, and geometry of the emission.

[0135] A4. Obtain the original physical source of the perceived features;

[0136] A5. Verify the physical rationality of the reflection path and the authenticity of the original physical source, and obtain the verification results;

[0137] A6. Based on the verification results, the reliability assessment results of the perceived features are revised to obtain the revised second assessment results.

[0138] In some embodiments, the fusion module 400 performs the following when determining the degree of inconsistency between the location information and the perceived features:

[0139] B1. Calculate the instantaneous geometric deviation between the current location information and the perceived features;

[0140] B2. Monitor the dynamic changes of instantaneous geometric deviation within a preset time window to obtain information on the dynamic changes of instantaneous geometric deviation;

[0141] B3. Based on the dynamic changes in instantaneous geometric deviation, assess the fluctuation range of instantaneous geometric deviation;

[0142] B4. Determine the degree of inconsistency based on instantaneous geometric deviation and fluctuation amplitude.

[0143] In some embodiments, the fusion module 400 performs the following when determining the primary source of error based on a first evaluation result and a second evaluation result when the inconsistency level exceeds a preset threshold:

[0144] C1. Based on the current location information, perception features, first evaluation result, and second evaluation result, generate multiple scenario explanation hypotheses and assign an initial confidence level to each scenario explanation hypothesis;

[0145] C2. Continuously receive new sensor data and independently update the internal state of each scenario interpretation hypothesis based on the new sensor data; wherein, each scenario interpretation hypothesis independently processes the positioning information and perception features according to its own interpretation of the positioning information and perception features; the sensor data is real-time monitoring data of sensors pre-deployed on the vehicle;

[0146] C3. For each new frame of sensor data, evaluate the consistency between the new sensor data and the state predicted by each scenario interpretation hypothesis, and dynamically adjust and accumulate the confidence scores of each scenario interpretation hypothesis based on the explanatory power of each scenario interpretation hypothesis under the current environmental conditions.

[0147] C4. Continuously monitor the confidence scores of the explanatory hypotheses for each scenario. When the confidence score of an explanatory hypothesis for a scenario is significantly higher than that of all other explanatory hypotheses for the same scenario, select that explanatory hypothesis for the scenario as the final explanation. If no explanatory hypothesis for a scenario reaches the convergence condition after a preset time or distance, mark the current region as an uncertain region.

[0148] C5. Based on the selected final interpretation, determine the main sources of information that cause the inconsistency between the current location information and the perceived features, thereby identifying the main source of error.

[0149] In some embodiments, the fusion module 400 performs the following when obtaining the fusion result by suppressing the influence of major error sources when fusing localization information and sensing features based on mutual correction intensity:

[0150] D1. Obtain the identification results of the main error sources;

[0151] D2. Based on the identification results of the main error sources, dynamically adjust the weights of the main error sources in the fusion process of localization information and perception features, or adjust the covariance matrix of the information corresponding to the main error sources to obtain the fusion result.

[0152] Please refer to Figure 3 , Figure 3This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. The present invention provides an electronic device 13, including: a processor 1301 and a memory 1302. The processor 1301 and the memory 1302 are interconnected and communicate with each other via a communication bus 1303 and / or other forms of connection mechanism (not shown). The memory 1302 stores computer-readable instructions executable by the processor 1301. When the electronic device is running, the processor 1301 executes these computer-readable instructions to perform the high-precision map construction method based on location-aware fusion in any optional implementation of the above embodiments, to achieve the following functions: acquiring location information and evaluating the reliability of the location information. The process involves obtaining a first evaluation result; acquiring perception features and evaluating their reliability to obtain a second evaluation result; when both the first and second evaluation results are below the corresponding preset evaluation thresholds, selecting a preset specific fusion mode; in the specific fusion mode, reducing the mutual correction intensity between positioning information and perception features, and determining the degree of inconsistency between positioning information and perception features; when the degree of inconsistency exceeds a preset threshold, determining the main source of error based on the first and second evaluation results, and obtaining a fusion result by suppressing the influence of the main source of error when fusing positioning information and perception features based on the mutual correction intensity; and generating high-precision map elements based on the fusion result.

[0153] This invention provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it executes a high-precision map construction method based on location-aware fusion in any optional implementation of the above embodiments to achieve the following functions: acquiring location information and evaluating the reliability of the location information to obtain a first evaluation result; acquiring perception features and evaluating the reliability of the perception features to obtain a second evaluation result; when both the first evaluation result and the second evaluation result are lower than the corresponding preset evaluation thresholds, selecting a preset specific fusion mode; in the specific fusion mode, reducing the mutual correction intensity between the location information and the perception features, and determining the degree of inconsistency between the location information and the perception features; when the degree of inconsistency exceeds a preset degree threshold, determining the main source of error based on the first evaluation result and the second evaluation result, and obtaining a fusion result by suppressing the influence of the main source of error when fusing the location information and perception features based on the mutual correction intensity; and generating high-precision map elements based on the fusion result.

[0154] The computer-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0155] In the embodiments provided by this invention, it should be understood that the disclosed apparatus and method can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0156] Furthermore, the units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0157] Furthermore, the functional modules in the various embodiments of the present invention can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0158] In this document, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, without necessarily requiring or implying any such actual relationship or order between these entities or operations.

[0159] The above description is merely an embodiment of the present invention and is not intended to limit the scope of protection of the present invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for constructing high-precision maps based on location-aware fusion, characterized in that, Includes the following steps: S1. Obtain location information and assess the reliability of the location information to obtain the first assessment result; S2. Acquire the perceptual features and evaluate the reliability of the perceptual features to obtain the second evaluation result; S3. When both the first evaluation result and the second evaluation result are lower than the corresponding preset evaluation threshold, select the preset specific fusion mode; S4. In a specific fusion mode, reduce the mutual correction strength between positioning information and sensing features, and determine the degree of inconsistency between positioning information and sensing features. When the degree of inconsistency exceeds a preset threshold, determine the main source of error based on the first evaluation result and the second evaluation result. When fusing positioning information and sensing features based on mutual correction strength, obtain the fusion result by suppressing the influence of the main source of error. The mutual correction strength is the amount of correction between the localization information and the perceived features; S5. Based on the fusion results, generate high-precision map elements; Step S2, which involves evaluating the reliability of the perceived features to obtain a second evaluation result, includes: The reliability of the sensing features is determined by evaluating their geometric properties, topological relationships, and the degree to which they are affected by environmental reflections, thus obtaining a second evaluation result. Before performing step S3, if the evaluation results of the geometric characteristics or topological relationships of the perceived feature indicate that the reliability of the perceived feature is questionable, perform the following steps: A1. Identify reflective surfaces in the environment and determine their geometry; A2. Trace the origin and direction of the laser beams that correspond to the questionable sensing features; A3. Determine the reflection path of the laser beam based on the origin, direction, and geometry of the emission. A4. Obtain the original physical source of the perceived features; A5. Verify the physical rationality of the reflection path and the authenticity of the original physical source, and obtain the verification results; A6. Based on the verification results, the reliability assessment results of the perceived features are revised to obtain the revised second assessment results.

2. The high-precision map construction method based on positioning-aware fusion according to claim 1, characterized in that, Step S1, which involves evaluating the reliability of the location information and obtaining the first evaluation result, includes: The reliability of location information is determined by assessing the degree to which it is affected by environmental factors, thus obtaining the first assessment result.

3. The high-precision map construction method based on positioning-aware fusion according to claim 1, characterized in that, Step S4, which involves determining the degree of inconsistency between the location information and the perceived features, includes: B1. Calculate the instantaneous geometric deviation between the current location information and the perceived features; B2. Monitor the dynamic changes of instantaneous geometric deviation within a preset time window to obtain information on the dynamic changes of instantaneous geometric deviation; B3. Based on the dynamic changes in instantaneous geometric deviation, assess the fluctuation range of instantaneous geometric deviation; B4. Determine the degree of inconsistency based on instantaneous geometric deviation and fluctuation amplitude.

4. The high-precision map construction method based on positioning-aware fusion according to claim 1, characterized in that, In step S4, when the degree of inconsistency exceeds a preset threshold, the step of determining the main source of error based on the first evaluation result and the second evaluation result includes: C1. Based on the current location information, perception features, first evaluation result, and second evaluation result, generate multiple scenario explanation hypotheses and assign an initial confidence level to each scenario explanation hypothesis; C2. Continuously receive new sensor data and independently update the internal state of each scenario interpretation hypothesis based on the new sensor data; wherein, each scenario interpretation hypothesis independently processes the positioning information and perception features according to its own interpretation of the positioning information and perception features; the sensor data is real-time monitoring data of sensors pre-deployed on the vehicle; C3. For each new frame of sensor data, evaluate the consistency between the new sensor data and the state predicted by each scenario interpretation hypothesis, and dynamically adjust and accumulate the confidence scores of each scenario interpretation hypothesis based on the explanatory power of each scenario interpretation hypothesis under the current environmental conditions. C4. Continuously monitor the confidence scores of the explanatory hypotheses for each scenario. When the confidence score of an explanatory hypothesis for a scenario is significantly higher than that of all other explanatory hypotheses for the same scenario, select that explanatory hypothesis for the scenario as the final explanation. If no explanatory hypothesis for a scenario reaches the convergence condition after a preset time or distance, mark the current region as an uncertain region. C5. Based on the selected final interpretation, determine the main sources of information that cause the inconsistency between the current location information and the perceived features, thereby identifying the main source of error.

5. The high-precision map construction method based on positioning-aware fusion according to claim 1, characterized in that, In step S4, the steps to obtain the fusion result by suppressing the influence of the main error sources when fusing localization information and perceptual features based on mutual correction intensity include: D1. Obtain the identification results of the main error sources; D2. Based on the identification results of the main error sources, dynamically adjust the weights of the main error sources in the fusion process of localization information and perception features, or adjust the covariance matrix of the information corresponding to the main error sources to obtain the fusion result.

6. A high-precision map construction device based on positioning and sensing fusion, characterized in that, include: The first evaluation module is used to acquire location information, evaluate the reliability of the location information, and obtain the first evaluation result. The second evaluation module is used to acquire perceptual features and evaluate the reliability of the perceptual features to obtain the second evaluation result. The judgment and selection module is used to select a preset specific fusion mode when both the first evaluation result and the second evaluation result are lower than the corresponding preset evaluation threshold. The fusion module is used to reduce the mutual correction intensity between positioning information and sensing features under a specific fusion mode, and to determine the degree of inconsistency between positioning information and sensing features. When the degree of inconsistency exceeds a preset threshold, the main source of error is determined based on the first evaluation result and the second evaluation result. When fusing positioning information and sensing features based on mutual correction intensity, the fusion result is obtained by suppressing the influence of the main source of error. The mutual correction strength is the amount of correction between the localization information and the perceived features; The generation module is used to generate high-precision map features based on the fusion results; The second evaluation module is executed when acquiring perceived features, evaluating the reliability of those features, and obtaining the second evaluation result: The reliability of the sensing features is determined by evaluating their geometric properties, topological relationships, and the degree to which they are affected by environmental reflections, thus obtaining a second evaluation result. It also includes a correction module, which, before the selection module runs, performs the following steps when the evaluation results of the geometric characteristics or topological relationships of the sensed features indicate that the reliability of the sensed features is questionable: A1. Identify reflective surfaces in the environment and determine their geometry; A2. Trace the origin and direction of the laser beams that correspond to the questionable sensing features; A3. Determine the reflection path of the laser beam based on the origin, direction, and geometry of the emission. A4. Obtain the original physical source of the perceived features; A5. Verify the physical rationality of the reflection path and the authenticity of the original physical source, and obtain the verification results; A6. Based on the verification results, the reliability assessment results of the perceived features are revised to obtain the revised second assessment results.

7. An electronic device, characterized in that, It includes a processor and a memory, the memory storing computer-readable instructions, which, when executed by the processor, perform the steps in the high-precision map construction method based on location-aware fusion as described in any one of claims 1-5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it performs the steps in the high-precision map construction method based on location-aware fusion as described in any one of claims 1-5.

Images