Multi-model collaborative fusion positioning method and device, equipment and storage medium

Abstract

The invention belongs to the technical field of space positioning, and discloses a multi-model collaborative fusion positioning method and device, equipment and a storage medium. The method comprises the steps that firstly, according to a real-time driving scene, a plurality of positioning models with complementary advantages are intelligently selected from a model library for parallel calculation, a positioning result and the uncertain ellipse area output by each model in the current round are output, and a confidence coefficient score of the positioning result is obtained based on the area; performing outlier detection according to a plurality of positioning results, and determining an outlier model in combination with a confidence score; and finally, dynamically selecting a fusion strategy according to an outlier detection result, and carrying out weighted voting fusion on the output of the credible model to generate a final positioning result. According to the method, the optimal model combination can be called in any driving environment through scene self-adaptive model selection, the limitation of a single model is overcome, and the stability and reliability of the system in a complex and severe environment are remarkably improved by combining an outlier detection and weighted voting fusion mechanism.

Description

Technical Field

[0001] This application relates to the field of spatial positioning technology, and in particular to multi-model collaborative fusion positioning methods, devices, equipment and storage media. Background Technology

[0002] In today's era of rapid development in intelligent driving, the Internet of Things, and precise location services, high-precision and high-reliability positioning technology is the cornerstone supporting their realization. However, existing mainstream positioning technologies all have their inherent advantages and limitations. No single technology can independently meet the high-precision positioning needs of all-weather, all-scenario environments in complex and dynamic settings.

[0003] To overcome the limitations of single technologies, solutions integrating multiple technologies have become an inevitable trend. Technological advancements have spawned numerous different positioning algorithms and models, built upon various principles (such as geometric calculation, inertial estimation, and environmental feature matching), forming multiple heterogeneous model libraries based on different principles. However, traditional fusion methods, such as filtering or weighted averaging, rely on stable assumptions about the model error characteristics. The significant differences between heterogeneous models and their dynamic performance fluctuations make designing a unified, static, and optimal fusion rule extremely difficult. Therefore, the adaptability of the fusion strategy itself has become a bottleneck restricting the performance improvement of positioning systems.

[0004] Therefore, how to design a fusion positioning method that can intelligently coordinate multiple heterogeneous positioning models and has high fault tolerance for abnormal results is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] The main purpose of this application is to provide a multi-model collaborative fusion positioning method, device, equipment and storage medium, which aims to solve the technical problem in the prior art of how to design a fusion positioning method that can intelligently coordinate multiple heterogeneous positioning models and has high fault tolerance for abnormal results.

[0006] To achieve the above objectives, this application proposes a multi-model collaborative fusion localization method, the method comprising: Based on the real-time driving scenario of the vehicle, multiple cooperative positioning models are determined from a pre-set positioning model library; Real-time environmental perception data is input into the collaborative localization model to obtain the model localization output result and the model localization confidence score corresponding to the model localization output result. Outlier detection is performed on the model localization output to obtain the model localization outlier detection result; Based on the outlier detection results of the model, determine the localization result fusion strategy; Based on the location result fusion strategy and the model location reliability score, the location outputs of the multiple models are fused to obtain a fused location result.

[0007] Optionally, the step of determining multiple cooperative positioning models from a preset positioning model library based on the real-time driving scenario of the vehicle includes: Based on a preset model positioning capability query table, obtain the multi-dimensional positioning capability distribution results of each positioning model in the positioning model library under the real-time driving scenario; The multi-dimensional positioning capability distribution results of each positioning model are complementary and paired to obtain multiple candidate model pairings, and the candidate model pairings include at least three positioning models. The resource-outcome cost-effectiveness of the candidate model pairings is evaluated, and the target pairings are determined based on the evaluation results. The multiple localization models of the target pairing results are used as the cooperative localization model.

[0008] Optionally, the step of inputting real-time environmental perception data into the cooperative localization model to obtain the model localization output includes: Based on the data modalities required by the cooperative localization model, the real-time environmental perception data is preprocessed to generate a standardized data stream that conforms to the input format of each model; Each standardized data stream is input in parallel into the corresponding cooperative localization model to obtain the original localization output results of each cooperative localization model; The original positioning output results are formatted to generate a standard-format model positioning output result, which includes longitude positioning results, latitude positioning results, altitude positioning results, and positioning covariance matrix.

[0009] Optionally, the step of inputting real-time environmental perception data into the collaborative localization model to obtain model positioning reliability scores that correspond one-to-one with the model positioning output includes: Based on the positioning output, the positioning covariance matrix generated synchronously with the current positioning result is obtained; The positioning covariance matrix is ​​as follows: , The variance of the longitude error. The variance of latitude error. This represents the covariance between longitude and latitude errors; Based on the location covariance matrix, a characteristic equation is constructed and solved to obtain the solution result of the characteristic equation. Based on the solution of the characteristic equation, the area of ​​the uncertainty ellipse of the positioning output result is obtained; The weighting coefficients are determined based on the real-time driving scenario of the vehicle and the changing trend of the area of ​​the uncertain ellipse within a preset time period. Based on the weighting coefficients, the area of ​​the uncertainty ellipse, and the historical positioning error parameters, the model positioning reliability score corresponding to the model positioning output is obtained.

[0010] Optionally, the step of performing outlier detection on the model localization output to obtain model localization outlier detection results includes: Based on the model positioning output results of each cooperative positioning model, the spatial distribution variance of the positioning results is obtained; Based on the real-time driving scenario of the vehicle, determine the spatial distribution variance threshold under the current scenario; If the spatial distribution variance is greater than the spatial distribution variance threshold, then it is determined that there are outliers in the model localization results; If the spatial distribution variance is less than or equal to the spatial distribution variance threshold, then it is determined that there are no outliers in the model localization results.

[0011] Optionally, after determining that there are outliers in the model localization result if the spatial distribution variance is greater than the spatial distribution variance threshold, the method further includes: Based on the model positioning output results of each cooperative positioning model, the spatial mean point of the positioning result is obtained; Calculate the spatial distance between the localization output of each model and the spatial mean point; If the spatial distance is greater than the spatial distribution variance of a preset multiple, then the collaborative localization model corresponding to the model localization output result is marked as a potential outlier model. Based on the real-time driving scenario of the vehicle, determine the model positioning confidence threshold for the current scenario; If the model location reliability score corresponding to the potential outlier model is less than the model location reliability threshold, then the potential outlier model is marked as an outlier model.

[0012] Optionally, the step of fusing the localization outputs of the multiple models according to the localization result fusion strategy and the model localization confidence score to obtain a fused localization result includes: If outliers exist in the model localization results, the outlier models in the collaborative localization model are removed. Based on the real-time driving scenario of the vehicle, the remaining cooperative positioning models are assigned positioning voting weights. The fused positioning result is obtained by fusing the positioning voting weight allocation result, the model positioning output result, and the model positioning reliability score corresponding to the model positioning output result.

[0013] Furthermore, to achieve the above objectives, this application also proposes a multi-model collaborative fusion positioning device, which includes: The model scheduling module is used to determine multiple cooperative positioning models from a preset positioning model library based on the real-time driving scenario of the vehicle. The parallel positioning module is used to input real-time environmental perception data into the collaborative positioning model to obtain the model positioning output result and the model positioning confidence score corresponding one-to-one with the model positioning output result. The outlier detection module is used to perform outlier detection on the model localization output results to obtain the model localization outlier detection results; The localization fusion module is used to determine the localization result fusion strategy based on the localization outlier detection results of the model. The positioning fusion module is further configured to fuse the positioning outputs of the multiple models according to the positioning result fusion strategy and the model positioning confidence score to obtain a fused positioning result.

[0014] In addition, to achieve the above objectives, this application also proposes a multi-model collaborative fusion positioning device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the multi-model collaborative fusion positioning method as described above.

[0015] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the multi-model collaborative fusion localization method as described above.

[0016] The technical solution of this application includes: determining multiple cooperative positioning models from a preset positioning model library based on the real-time driving scenario of the vehicle; inputting real-time environmental perception data into the cooperative positioning model to obtain model positioning output results and model positioning reliability scores corresponding one-to-one with the model positioning output results; performing outlier detection on the model positioning output results to obtain model positioning outlier detection results; determining a positioning result fusion strategy based on the model positioning outlier detection results; and fusing the multiple model positioning output results according to the positioning result fusion strategy and the model positioning reliability scores to obtain a fused positioning result.

[0017] The one or more technical solutions proposed in this application have at least the following technical effects: First, by dynamically selecting a cooperative positioning model based on the real-time driving scenario of the vehicle, this application enables the system to flexibly respond to drastic changes in signal or environmental characteristics such as urban canyons, under overpasses, and tunnel entrances and exits. It consistently calls upon multiple complementary positioning models for parallel computation in the current scenario, thus ensuring the quality of the positioning results from the outset. Second, by generating confidence scores synchronously with the model positioning results, the fusion process can accurately reflect the credibility of each model at the current moment. It can assign higher weights to more reliable models based on real-time data, thereby significantly improving the accuracy of the fusion results. Combined with outlier detection, it can effectively identify and eliminate abnormal positioning results, ensuring the accuracy of the final positioning result generated during the positioning result fusion stage. In summary, this application systematically solves the key challenges in heterogeneous model fusion through scene-adaptive model selection, confidence-guided dynamic fusion, and an outlier detection fault-tolerance mechanism, ultimately achieving a synergistic leap in positioning accuracy, adaptability, and robustness. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating an embodiment of the multi-model collaborative fusion localization method of this application. Figure 2 This is a flowchart illustrating Embodiment 2 of the multi-model collaborative fusion localization method of this application. Figure 3 This is a flowchart illustrating Embodiment 3 of the multi-model collaborative fusion localization method of this application; Figure 4 This is a flowchart illustrating Embodiment 4 of the multi-model collaborative fusion localization method of this application; Figure 5 This is a schematic diagram of the module structure of the multi-model collaborative fusion positioning device according to an embodiment of this application; Figure 6 This is a schematic diagram of the device structure of the hardware operating environment involved in the multi-model collaborative fusion localization method in the embodiments of this application.

[0021] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0022] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0023] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0024] In this embodiment, for ease of description, the following description uses a multi-model collaborative fusion positioning device as the execution subject.

[0025] In complex environments and with varying user distributions, the system lacks adaptability to channel structure characteristics, resulting in significant quantization errors. Simply increasing the codebook size further exacerbates the overhead. Thirdly, it lacks efficient feature extraction. The main solution in this application's embodiment is: based on the real-time driving scenario of the vehicle, multiple cooperative positioning models are determined from a pre-set positioning model library; real-time environmental perception data is input into the cooperative positioning models to obtain model positioning output results and model positioning reliability scores corresponding one-to-one with the model positioning output results; outlier detection is performed on the model positioning output results to obtain model positioning outlier detection results; a positioning result fusion strategy is determined based on the model positioning outlier detection results; and the multiple model positioning output results are fused according to the positioning result fusion strategy and the model positioning reliability scores to obtain a fused positioning result.

[0026] Currently, all existing mainstream positioning technologies have their inherent advantages and limitations. For example, while Global Navigation Satellite Systems (GNSS) can provide absolute positioning information with global coverage, their signals are easily blocked and interfered with by multipath reflections in non-line-of-sight environments such as urban canyons and tunnels, leading to a sharp drop in accuracy or even failure. Inertial Navigation Systems (INS) can provide continuous autonomous positioning for short periods, but their errors accumulate and drift over time, making them unsuitable for long-term independent operation. Positioning technologies based on wireless signals such as Wi-Fi, Bluetooth, and UWB are effective in indoor scenarios, but their coverage is limited and easily affected by environmental changes. Visual / LiDAR and other environmental feature matching positioning technologies offer high accuracy, but heavily rely on lighting conditions, weather, and the completeness of prior maps. No single technology can independently meet the high-precision positioning requirements of all-weather, all-scenario environments in complex dynamic settings.

[0027] Traditional multi-source fusion localization methods cannot fundamentally overcome the aforementioned problems due to inherent limitations: First, model selection is disconnected from application scenarios, typically employing fixed, pre-defined model combinations and fusion weights, making it difficult to adapt to dynamically changing driving environments and potentially failing to achieve optimal model performance during scenario switching. Second, the fusion strategy lacks effective perception and fault tolerance mechanisms for instantaneous performance fluctuations and abnormal outputs of the models. Temporary failures or abnormal output values ​​from one or more sources, or unreasonable fusion weights, can severely drag down overall localization accuracy and reliability. Therefore, designing a fusion localization method that can intelligently coordinate multiple heterogeneous localization models and possesses high fault tolerance for abnormal results is a pressing technical problem in this field.

[0028] This application first intelligently selects multiple complementary positioning models from a model library based on the real-time driving scenario, performs parallel computation, and outputs the positioning results and the area of ​​the uncertainty ellipse of each model's output in this round. A confidence score for the positioning result is obtained based on this area. Then, outlier detection is performed on multiple positioning results, and outlier models are identified by combining the confidence scores. Finally, a fusion strategy is dynamically selected based on the outlier detection results, and the outputs of the reliable models are weighted and fused to generate the final positioning result. This method, through scene-adaptive model selection, ensures that the most suitable model combination can be used in any driving environment, overcoming the limitations of a single model. Combined with outlier detection and weighted voting fusion mechanisms, it significantly improves the stability and reliability of the system in complex and harsh environments.

[0029] This application provides a solution aimed at addressing the technical problem of the lack of a global resource scheduling mechanism in the prior art, which can dynamically perceive task characteristics, coordinate computing and communication resources, and thus achieve overall performance optimization.

[0030] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or a multi-model collaborative fusion positioning device capable of achieving the above functions. The following description uses a multi-model collaborative fusion positioning device as the executing entity to illustrate this embodiment and the subsequent embodiments.

[0031] Based on this, the embodiments of this application provide a multi-model collaborative fusion localization method, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the multi-model collaborative fusion localization method of this application.

[0032] In this embodiment, the multi-model collaborative fusion localization method includes steps S10 to S50: Step S10: Based on the real-time driving scenario of the vehicle, determine multiple cooperative positioning models from the preset positioning model library.

[0033] It should be noted that the core purpose of this step is to select a series of complementary positioning models or positioning sources from a pre-set positioning model library based on the real-time driving scenario, and to fuse the positioning results generated by these selected objects in order to improve the accuracy of the final fusion result.

[0034] Understandably, due to the limitations of the positioning principles and data foundations of various positioning models, no single positioning model or sensor can maintain optimal performance in all environments. Each model has its "advantageous scenarios" and "disadvantageous scenarios." For example, the Global Navigation Satellite System (GNSS) can provide absolute positioning information with global coverage, but in non-line-of-sight environments such as urban canyons and tunnels, its signal is easily blocked and interfered with by multipath reflections, resulting in a sharp drop in accuracy or even failure. Inertial navigation systems (INS) can provide continuous autonomous positioning for a short period of time, but their errors accumulate and drift over time, making them unsuitable for long-term independent operation. Positioning technologies based on wireless signals such as Wi-Fi, Bluetooth, and UWB are effective in indoor scenarios, but their coverage is limited and easily affected by environmental changes. Visual / LiDAR and other environmental feature matching positioning technologies offer high accuracy, but heavily rely on lighting conditions, weather, and the completeness of prior maps. Therefore, the first key concept of this application is to elevate the "scenario" to the primary variable driving the entire fusion process.

[0035] It should be understood that the key to this step lies in the dynamic mapping logic of "determining the model based on the scenario," which is based on pre-established and fully validated scenario-model matching rules. For example, when the system determines that the vehicle is traveling on a structured urban road with clear lane lines and rich visual features, the system will prioritize including the "visual-inertial odometry model" in the collaborative model set; while when the vehicle enters an open highway with good GNSS signals, the "high-precision GNSS differential positioning model" will be the primary candidate. This intelligent selection based on rules or machine learning algorithms ensures that the model being called is more likely to output high-precision, high-reliability positioning results in the current location environment.

[0036] It is important to note that this step identifies multiple cooperative positioning models. This means that the decision-making process is not a simple either-or switch between models, but rather the formation of an "optimal subset of models" for the current scenario. This design fully leverages the complementarity of different models in various positioning capabilities. For example, in the transitional area of ​​a tunnel entrance / exit where signal strength changes drastically, the system may simultaneously select an "inertial navigation model" (to compensate for momentary GNSS failures) and a "pre-stored map-based matching model" (to provide absolute position constraints). Their collaborative work smooths the transition and suppresses abrupt changes or drift of a single model. Therefore, the series of positioning models selected in this step is a scenario-optimized combination with complementary potential, providing accurate input for subsequent parallel computation, confidence assessment, and intelligent fusion.

[0037] Step S20: Input the real-time environmental perception data into the collaborative localization model to obtain the model localization output result and the model localization confidence score corresponding to the model localization output result.

[0038] It should be noted that this step follows the previously selected "cooperative localization model subset" and is the execution phase for parallel data processing. This step includes two key outputs: the first is the "model localization output result," which is the vehicle localization result independently calculated by each activated model based on the sensor data at the current moment (i.e., "real-time environmental perception data"), typically including information such as position, speed, and attitude. The second is the corresponding "positioning reliability score," which characterizes the reliability of the model's output result at the current moment.

[0039] Understandably, the core purpose of this step is to provide both "data" and "reliability" as the basis for subsequent fusion decisions. Obtaining the positioning results from multiple models is only the first step; more importantly, it's about objectively evaluating the reliability of each result. This evaluation is dynamic and real-time, reflecting the instantaneous performance of the model in a specific environment. For example, when a vehicle briefly passes through a shady area causing a temporary loss of visual model feature tracking, its confidence score will immediately and significantly decrease; conversely, when the vehicle exits the tunnel, GNSS signals recover, and the satellite geometry is good, the confidence of the GNSS model will rapidly increase. This dynamically changing confidence is the key difference from traditional fixed-weight fusion methods, enabling the fusion system to perceive the reliability of each information source.

[0040] Optionally, inputting real-time environmental perception data into the cooperative positioning model to obtain model positioning output results includes: preprocessing the real-time environmental perception data according to the data modalities required by the cooperative positioning model to generate standardized data streams that conform to the input format of each model; inputting each standardized data stream in parallel into the corresponding cooperative positioning model to obtain the original positioning output results of each cooperative positioning model; and performing format unification processing on the original positioning output results to generate model positioning output results in a standard format, wherein the model positioning output results include longitude positioning results, latitude positioning results, altitude positioning results, and positioning covariance matrix.

[0041] Understandably, by deploying models based on GPU multi-threading or distributed computing frameworks, ensuring that three or more models simultaneously receive real-time sensing data (such as GNSS, IMU, camera, 5G positioning signals), each model can process the data and output positioning results in parallel. The output format can be uniformly set as: latitude, longitude, altitude, and uncertain ellipse parameters.

[0042] Step S30: Perform outlier detection on the model localization output to obtain the model localization outlier detection result.

[0043] It's important to note that after obtaining the localization results and their confidence scores for each model, outlier detection is necessary. If a localization result has a high confidence score, but its output value significantly deviates from the consensus of most models or the abnormal results of physical motion constraints, then this result may be an outlier. For example, when a vehicle is traveling at a constant speed in a straight line, if one of the models outputs a point that severely deviates from the driving trajectory due to instantaneous interference, this result is considered an "outlier." Detection algorithms can be based on statistical methods (such as Isolation Forest, Z-score) or physical consistency (such as a significant discrepancy with the displacement calculated by the inertial measurement unit).

[0044] Optionally, the step of performing outlier detection on the model localization output to obtain model localization outlier detection results includes: obtaining the spatial distribution variance of the localization results based on the model localization output results of each cooperative localization model; determining a spatial distribution variance threshold in the current scenario based on the real-time driving scenario of the vehicle; if the spatial distribution variance is greater than the spatial distribution variance threshold, it is determined that there are outliers in the model localization results; if the spatial distribution variance is less than or equal to the spatial distribution variance threshold, it is determined that there are no outliers in the model localization results.

[0045] It's important to note that spatial distribution variance refers to the spatial dispersion of the output results from all participating collaborative localization models. This variance reflects the consensus among the localization results of each model; the larger the variance, the greater the discrepancy between models, and the higher the likelihood of outliers. The spatial distribution variance threshold is dynamically adjusted based on the real-time driving scenario. For example, in open scenarios such as highways, where high consistency in localization is required, the threshold can be set relatively low to quickly capture subtle anomalies. However, in signal-complex areas such as multi-story parking lots, significant discrepancies between models are normal. In such cases, the threshold is automatically increased to avoid overly sensitively misjudging reasonable differences as outliers.

[0046] Understandably, this step first assesses the likelihood of anomalies from a macroscopic perspective. If the variance does not exceed a threshold, all results are considered generally reliable, requiring no further processing and improving efficiency. If outliers are identified, they provide trigger signals for subsequent steps (such as checking each result individually or applying weighted penalties). For example, when the system detects a surge in variance, a more sophisticated algorithm can be activated to identify and eliminate specific faulty models, thereby ensuring the robustness of the fusion system.

[0047] Step S40: Determine the localization result fusion strategy based on the outlier detection results of the model.

[0048] It should be noted that when the system determines that there are no outliers, it means that the output results of each model are highly consistent and reliable. In this case, the optimal choice is to use dynamic weighted average fusion based on confidence level. It can smoothly integrate all information and allow the high-confidence model to dominate, thereby obtaining accurate and stable output.

[0049] Understandably, when outliers are detected, it indicates that there are potentially erroneous localization results in the dataset that deviate from the truth. In this case, the primary goal of the fusion strategy is robustness rather than smoothness. Robust estimation algorithms such as RANSAC / PROSAC can be used in this step. Essentially, they use iterative sampling and consistency verification to select the largest set of "interior points" (i.e., model results that are close to each other and likely correct) from all candidate results, and decisively eliminate inconsistent "outside points" (i.e., outlier models). For example, if a GPS signal provides a deviated point due to multipath errors caused by urban canyons, while the visual and LiDAR model results are consistent, the robust algorithm can effectively identify and eliminate GPS outliers.

[0050] It should be understood that after removing outliers, the weights can be redistributed again according to the confidence scores in the previous steps. The remaining inlier models can be weighted averaged, or the coordinate interval with the most occurrences can be selected by "voting". If multiple models are concentrated within a 2-meter range, the center of that interval can be taken as the output with the highest confidence score.

[0051] Step S50: Based on the localization result fusion strategy and the model localization confidence score, fuse the localization outputs of the multiple models to obtain the fused localization result.

[0052] It should be noted that in this step, the remaining valid model localization outputs are mathematically synthesized by combining the confidence scores of each model, ultimately generating an optimal, single fused localization result. The weights are determined by normalizing the confidence scores of each model, thus transforming the decision logic into actual location coordinates.

[0053] Understandably, by implementing the fusion strategy selected in the aforementioned steps, this step can effectively achieve two effects: in the ideal case where there are no outliers, the overall accuracy is improved by finely balancing the contributions of each model through weighted averaging; while in the case of outliers, it first filters and cleans up through robust algorithms to ensure that the final result is determined only by reliable "interior" models, thereby significantly enhancing the system's ability to resist interference and individual model failures.

[0054] It should be understood that after outputting the fused localization results, the reasonableness of the results can be verified. This can be done by combining the prior urban canyon map (such as road boundaries and lane line coordinates) to verify whether the fused results are located within a reasonable area. If the deviation from the road boundary exceeds a certain distance, the fusion is deemed to have failed, triggering a backup mechanism, such as activating IMU short-term prediction. Furthermore, the fused localization results can be used for subsequent system algorithm optimization. For example, the output format can be converted to a standard format for fusion dimensions, fusion longitude, fusion altitude, fusion uncertainty ellipse area, and the model numbers participating in the fusion. Additionally, outlier model information (such as outlier time and cause) can be recorded. This information can also be used for subsequent overall system algorithm optimization.

[0055] This embodiment dynamically selects the optimal combination of cooperative localization models in real-time vehicle driving scenarios, calculates the localization confidence scores of each model in real time, performs outlier detection based on spatial distribution variance, and finally adaptively selects dynamic weight averaging or robust estimation algorithms for fusion based on the detection results, thereby realizing intelligent filtering and optimal synthesis of multi-source localization information.

[0056] In summary, this technical solution effectively addresses the reliability issues of single localization models in complex environments by organically combining scene perception, confidence assessment, outlier detection, and an adaptive fusion strategy. First, scene-based dynamic model selection ensures that the source data participating in the fusion possesses high accuracy potential in the current location environment. Second, real-time confidence assessment provides dynamic and refined weighting criteria for fusion. Third, the outlier detection mechanism effectively identifies and isolates transient abnormal data, preventing it from contaminating the final result. Finally, the adaptive fusion strategy guarantees optimal output results under varying data quality conditions. Therefore, this solution significantly improves the accuracy, robustness, and overall reliability of the fusion localization system in various driving scenarios, providing more reliable location perception assurance for high-level autonomous driving.

[0057] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in the first embodiment described above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 2 Step S10 in the multi-model collaborative fusion localization method includes steps S101 to S104: Step S101: Based on the preset model positioning capability query table, obtain the multi-dimensional positioning capability distribution results of each positioning model in the positioning model library under the real-time driving scenario.

[0058] It's important to note that the core of this step lies in utilizing a pre-built "model positioning capability lookup table" to quantitatively evaluate the overall performance of each model in the positioning model library under the current specific driving scenario. This lookup table is a structured database, where rows typically correspond to different positioning models (such as GNSS, visual odometry, and laser SLAM), and columns correspond to different scene dimensions (such as weather conditions, road type, and signal strength) and performance indicators (such as accuracy, availability, and refresh rate). By inputting real-time scene parameters, the system can query the "multi-dimensional positioning capability distribution results" for each model, i.e., a series of quantified capability scores.

[0059] Understandably, the core value of this step lies in transforming qualitative scene perception into quantitative model capability assessment, providing an objective, data-driven decision-making basis for subsequent model selection. For example, when the system detects a current scene of "nighttime, rain, urban elevated road," the lookup table will show that: high-precision GNSS models may have lower "accuracy" and "availability" scores due to weather and obstruction; visual odometry models may experience a sharp drop in "reliability" scores due to insufficient nighttime lighting and slippery glass reflections; while lidar SLAM models are less affected by weather and perform well on structured roads, resulting in relatively higher scores across the board. This multi-dimensional quantitative distribution effectively improves the synergistic effect of model combinations.

[0060] It should be understood that positioning-related evaluation metrics mainly include accuracy metrics, reliability metrics, timeliness metrics, and robustness metrics. Accuracy metrics primarily include: absolute accuracy, which is the error distance between the location coordinates (such as latitude and longitude) output by the positioning system and the actual value; and relative accuracy, which refers to the error between the relative change in location output by the positioning system and the actual relative change over a period of time. Reliability metrics mainly include: availability, which refers to the probability that the system can provide positioning services in a specific area and time. For example, GPS availability is close to 100% in open skies, but 0% in tunnels. It reflects the system's coverage and robustness; and continuity, which refers to the system's ability to continuously provide services that meet accuracy and integrity requirements without interruption during a specific task. Timeliness metrics mainly include: refresh rate, which is the number of times the model outputs positioning results per second (Hz). A high refresh rate (such as 100Hz) is crucial for the real-time control of high-speed vehicles; and positioning latency, which refers to the time difference between sensor data acquisition and the final positioning result output. Robustness is a comprehensive indicator, not a single value. It refers to the ability of a positioning system to maintain its accuracy, availability, and integrity in complex environments (such as urban canyons, severe weather, and partial sensor failure).

[0061] Step S102: Complementary pairing of the multi-dimensional positioning capability distribution results of each positioning model to obtain multiple candidate model pairing combinations.

[0062] It's important to note that this step aims to intelligently combine and analyze the capability distribution results of the models obtained in the preceding steps. The core of this process is to identify and construct "complementary pairs." Here, "complementarity" does not refer to simply combining any two models, but rather to a combination where their positioning capabilities complement each other, allowing them to leverage each other's strengths and compensate for each other's weaknesses. Specifically, the system analyzes the strengths and weaknesses of different models across various dimensions. For example, one model might have high absolute accuracy in open scenes but a low update rate, while another model might be unaffected by signal obstruction and have a high update rate but suffer from cumulative errors. Pairing them allows the former to provide accurate absolute position anchors, while the latter ensures a smooth and continuous trajectory, thus forming a relationship where strengths complement each other and weaknesses cancel each other out.

[0063] Understandably, candidate model pairings require each combination to include at least three positioning models. This is to build a more robust and redundant fusion positioning system. If only two models are paired, and one fails due to sudden environmental changes, the entire system will degenerate into a single positioning source, posing a high risk. Introducing a third or more models allows for multiple backups and cross-validation. For example, a typical candidate combination might include GNSS (providing absolute coordinates), laser SLAM (providing high-precision relative motion and resistance to weather interference), and visual odometry (providing rich features to assist in correction). When GNSS signals are lost under an overpass, the system can still maintain high-precision positioning through the collaboration of laser SLAM and visual odometry, ensuring the system's continued reliability.

[0064] Step S103: Perform a resource-outcome cost-effectiveness evaluation on the candidate model pairings, and determine the target pairings based on the evaluation results.

[0065] It should be noted that the purpose of this step is to further screen and optimize the candidate combinations obtained in the previous steps. Here, "resources" is a comprehensive concept, encompassing not only computing resources (such as CPU / GPU utilization and memory consumption) and communication resources (such as data transmission bandwidth), but also the vehicle's power consumption and the hardware costs of the sensors themselves. "Results" refers to the positioning performance ultimately provided by this model combination, including indicators such as accuracy, reliability, and latency. Cost-effectiveness evaluation uses a predefined utility function to quantify the resource costs required to achieve a specific improvement in positioning performance.

[0066] Understandably, in principle, the more models involved in the localization process, the more accurate the final fusion localization result will be. However, this approach ignores the negative impacts of the sharp increase in system resource consumption, real-time requirements, and decision complexity.

[0067] It should be understood that the core of this step lies in evaluating the "cost-effectiveness" of each candidate combination, that is, the additional computational, power consumption, and time costs required to achieve a unit improvement in positioning performance. This enables the system to intelligently select a "target pairing combination" that consumes the most resources and ensures the most sustainable system operation while meeting the positioning requirements of the current scenario, thereby achieving performance optimization under limited resource constraints, rather than blindly pursuing theoretically ultimate accuracy.

[0068] Step S104: Use the multiple localization models of the target pairing results as the cooperative localization model.

[0069] It is understandable that the cooperative positioning model determined here is not static. It is an optimal snapshot based on the current environment and system state. When the vehicle driving scenario changes significantly (such as entering a tunnel from an open area with good GNSS), or when the system detects performance degradation or failure of a certain cooperative model, a new round of evaluation will be triggered, thereby dynamically updating the "target pairing combination" and the final "cooperative positioning model", achieving closed-loop optimization and continuous adaptation throughout the entire process.

[0070] In this embodiment, a multi-dimensional capability distribution of each model in the positioning model library under the current driving scenario is evaluated in real time based on a preset model positioning capability lookup table. Then, according to the principle of capability complementarity, multiple candidate pairings containing at least three positioning models are generated. Next, by evaluating the resource-result cost-effectiveness of each candidate pairing, the target pairing with the optimal resource consumption while meeting the current positioning requirements is selected. Finally, this target pairing is confirmed as the collaborative positioning model set to be actually deployed. This scheme achieves the best balance between positioning performance and system overhead through scene awareness and dynamic optimization configuration.

[0071] In summary, this technical solution, through scene awareness and quantitative capability assessment, enables the system to accurately match the current environment's requirements for the positioning model, thereby fundamentally improving the overall adaptability and environmental robustness of the fusion positioning system. Secondly, by employing capability complementarity pairing rules, a fusion core with inherent redundancy and cross-validation capabilities is constructed. Even if a single model temporarily fails, the system can still maintain reliable positioning by relying on other models, enhancing the system's continuous service reliability. Finally, the introduction of a resource-outcome cost-effectiveness evaluation mechanism allows the system to balance improved positioning accuracy with resource consumption such as computational load. In conclusion, this method, through a series of interconnected decision-making processes, ultimately achieves a triple benefit: improved positioning performance, ensured system reliability, and optimized resource utilization.

[0072] Based on the first embodiment of this application, in the third embodiment of this application, the content that is the same as or similar to that in the first embodiment described above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 3 Step S20 in the multi-model collaborative fusion localization method includes steps S201 to S205: Step S201: Based on the positioning output result, obtain the positioning covariance matrix generated synchronously with the current positioning result.

[0073] It should be noted that the covariance matrix is: ,in This represents the variance of the longitude error, used to measure the degree of dispersion of the longitude error. The larger the value, the more unstable the longitude positioning. The variance of latitude error is used to measure the degree of dispersion of latitude error. The larger the value, the more unstable the latitude positioning. The covariance is the covariance of longitude and latitude errors, reflecting their correlation. The covariance matrix can be directly output from the model, or calculated statistically from a large amount of sample data to obtain the required variance and covariance: variance covariance Where N is the number of samples, and These are the longitude and latitude of the i-th sample, respectively. and This is the mean.

[0074] Understandably, the value of the positioning covariance matrix directly reflects the reliability of the current positioning result. For example, in open areas, the diagonal value of the covariance matrix provided by a Global Navigation Satellite System (GNSS) receiver is usually small, indicating high positioning accuracy and low uncertainty. Conversely, when a vehicle enters an urban canyon, due to factors such as multipath effects, the diagonal value of the GNSS variance matrix will increase significantly, indicating a decrease in the reliability of the current position estimate. Furthermore, for a visual odometry model, the covariance of its translational estimation may be small, but the covariance of its rotational estimation (especially the yaw angle around the vertical axis) may be relatively large.

[0075] Step S202: Construct a characteristic equation based on the positioning covariance matrix and solve it to obtain the solution result of the characteristic equation.

[0076] It should be noted that the eigenvalue decomposition of the localization covariance matrix yields its characteristic equation det(Σ-λI)=0, where I is the identity matrix and det represents the determinant. For the aforementioned covariance matrix, the characteristic equation is: (Var x -λ)(Var y -λ)-Cov xy Solving this quadratic equation, we can obtain two eigenvalues, λ1 and λ2, where λ = 0.

[0077] Understandably, this equation theoretically must have a solution because the aforementioned covariance matrix is ​​a real symmetric matrix, and its characteristic equation always has real solutions; an N-order real symmetric matrix has N real solutions. Furthermore, the two eigenvalues ​​λ1 and λ2 describe the degree of variation of the data in two orthogonal directions, and the square roots of λ1 and λ2 correspond to the lengths of the major and minor axes of the uncertainty ellipse, respectively.

[0078] Step S203: Based on the solution result of the characteristic equation, obtain the area of ​​the uncertainty ellipse of the positioning output result.

[0079] It should be noted that the eigenvalues ​​λ1 and λ2 of the covariance matrix correspond to the major and minor semi-axes of the uncertainty ellipse, respectively, and the square roots of these eigenvalues ​​correspond to the major and minor semi-axes of the uncertainty ellipse, respectively. Therefore, based on these two solutions, the area of ​​the uncertainty ellipse... ,in and The area S is an eigenvalue of the covariance matrix. As a scalar, this area value S can comprehensively and intuitively reflect the overall uncertainty level of the current positioning results. The larger the area, the more diffuse the positioning information area is, and the lower the reliability.

[0080] Step S204: Determine the weighting coefficient based on the real-time driving scenario of the vehicle and the changing trend of the area of ​​the uncertain ellipse within a preset time period.

[0081] It should be noted that since each positioning model outputs the area of ​​the uncertainty ellipse for this round of positioning simultaneously, the system can essentially know the confidence level of the model's output for this instance. The system can also analyze the time series of the uncertainty ellipse area formed by the model within a preset time period (e.g., the last 5 seconds) and its changing trend. By analyzing this time series, the system can determine the dynamic behavior of positioning reliability. For example, if the area is at a stable low level, it indicates that the current model is still highly reliable; if it continuously increases, it indicates that the reliability of the model's output is decreasing; if it exhibits drastic fluctuations, it may indicate that the received raw sensing data has been interfered with.

[0082] Step S205: Based on the weighting coefficients, the area of ​​the uncertainty ellipse, and the historical positioning error parameters, obtain the model positioning reliability score corresponding to the model positioning output result.

[0083] It should be understood that confidence scores The following is given by combining the area of ​​the uncertain ellipse and historical error:

[0084] in, and Let be the weight, and satisfy... , To avoid a minimum value where the denominator is 0, A higher value indicates that the current output of the model is more reliable. That is, the smaller the area of ​​the uncertainty ellipse (…). Smaller), smaller historical error ( (Smaller values ​​indicate a lower confidence score).

[0085] It should be noted that dynamically adjusting the weight parameter values ​​can improve the "scenario adaptability" of the confidence score and avoid being misled by outdated information from historical errors: in the event of a sudden change in the scenario, by... Increase to make real-time The dominant confidence score quickly identifies model failures; simultaneously, it retains the "stable reference" value of historical errors: when the scene is stable, it... Increase The long-term statistical properties enhance the reliability of confidence scores. For example, when a model suddenly enters a tunnel in an urban canyon (GNSS signal loss). The instantaneous increase can be achieved through a surge in Sᵢ. The value drops below 0.2, weakening its effectiveness in fusion; and after exiting the tunnel, The rebound can quickly restore the model's reasonable weights. The essence of dynamic adjustment is to ensure that the calculation of confidence scores can always maintain an accurate measure of model reliability in complex and ever-changing urban canyon scenarios.

[0086] Understandable, and The value is dynamically adjusted according to the scenario. When the real-time scenario is highly consistent with the historical statistical scenario, the historical error... Its reference value is higher and should be increased. (reduced) When the real-time scene differs significantly from the historical scene, the area of ​​the real-time uncertainty ellipse is... It better reflects the current error and should be increased. (reduced) For example, when the GNSS signal is weak or the multipath effect is significant ( Increase (Reduced), tall buildings causing GNSS signal jumps, at which point the uncertainty ellipse area in the model output increases. It can suddenly increase (e.g., from 10m) 2 Increased to 100m 2 However, the historical error σᵢ may be much smaller than the actual error, reducing its reference value. In this case, it is necessary to increase α (e.g., α=0.7, β=0.3) to make... Real-time changes dominate the confidence score, i.e. The larger, The lower the value, the more it reflects the unreliable reality of the current state of the model. For example, when visual SLAM enters a textureless region (such as a street with a glass curtain wall)... Increase (Reducing) the amount of textureless regions causes visual feature matching to fail, affecting the model's output. The error will increase sharply (reflecting real-time positioning failure), but historical errors... (Statistics based on textured scenes) cannot reflect the specificity of the scene; in this case, the size can be increased. (e.g., α=0.8, =0.2), rapidly lowered by the surge in Sᵢ To avoid unreliable results from being included in the fusion.

[0087] In this embodiment, the output results and covariance matrix of each positioning model are obtained, and the area of ​​the ellipse representing positioning uncertainty is obtained by performing eigenvalue decomposition on the covariance matrix. Then, the weight coefficient of the model is dynamically determined by combining the real-time driving scenario and the changing trend of the uncertainty ellipse area within a preset time period. Finally, the positioning confidence score of each model's positioning output result is calculated by combining the weight coefficient, the instantaneous uncertainty ellipse area, and the historical positioning error parameters, which serves as the basis for multi-source information fusion.

[0088] In summary, this technical solution introduces the area of ​​the uncertainty ellipse based on eigenvalue decomposition as a quantitative indicator of instantaneous reliability. This area is then combined with scene-adapted dynamic weighting coefficients and historical error parameters. This allows the confidence score calculation to not only reflect the long-term statistical performance of the positioning model but also to sensitively capture instantaneous performance fluctuations caused by signal obstruction and multipath effects in complex urban environments. Therefore, this invention ultimately achieves accurate and adaptive evaluation of the reliability of multi-source positioning information in complex positioning scenarios, significantly improving the robustness and accuracy of the fusion positioning system.

[0089] Based on the first embodiment of this application, in the fourth embodiment of this application, the content that is the same as or similar to that in the first embodiment described above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 4 Step S50 in the multi-model collaborative fusion localization method includes steps S501 to S503: Step S501: If there are outliers in the model localization results, then remove the outliers from the collaborative localization model.

[0090] It should be noted that outliers here refer to abnormal values ​​in which the positioning result of one of these cooperative positioning models differs significantly from the majority of other models. The criteria for judging outliers are usually based on preset statistical thresholds (such as exceeding N times the standard deviation of the fusion result) or spatial distance thresholds (such as deviating from the fusion center by more than M meters). The core of this step is to identify and eliminate these obviously unreliable positioning sources to prevent them from contaminating the final integrated navigation result.

[0091] Understandably, for scenarios without outliers, this indicates that all current location results are reliable, and the process of removing outliers can be skipped directly to proceed to the subsequent dynamic weighted fusion process.

[0092] In a feasible embodiment, the specific implementation steps of the outlier elimination model further include steps A10 to A50: Step A10: Based on the model positioning output results of each cooperative positioning model, obtain the spatial mean point of the positioning result.

[0093] Step A20: Calculate the spatial distance between the localization output of each model and the spatial mean point.

[0094] Step A30: If the spatial distance is greater than the spatial distribution variance of a preset multiple, then the collaborative localization model corresponding to the model localization output result is marked as a potential outlier model.

[0095] It should be noted that by calculating the spatial mean of the outputs of all available cooperative localization models, a reference position representing the current "collective consensus" is obtained. Subsequently, the Euclidean distance between each model's output and this spatial mean is calculated, quantifying the degree of deviation of each model's result from the "consensus." Finally, a discrimination threshold is set based on statistical principles, namely a preset fold (usually within 2 to 3 times) of the spatial distribution variance. When a model's distance significantly exceeds this range, it is marked as a potential outlier model, meaning that the result is highly likely to be invalid or severely disturbed.

[0096] Understandably, the effectiveness of this method relies on a fundamental assumption: under normal operating conditions, the outputs of most reliable positioning sources should exhibit a Gaussian distribution around the true location. For example, in open road scenarios, the positioning points of multiple GNSS receivers and visual odometry models will cluster closely together. If a model outputs a point far from the cluster due to an internal fault, its spatial distance from the mean will be abnormally large, thus accurately identifying it as an outlier. Conversely, at the entrance to a city canyon with complex signals, the positioning accuracy of all models may temporarily decrease, leading to an increase in the spatial distribution variance itself. This makes the discrimination threshold more adaptable, avoiding misclassification of models that have shifted overall due to common-cause errors as outliers.

[0097] Step A40: Determine the model positioning confidence threshold in the current scenario based on the real-time driving scenario of the vehicle.

[0098] It's important to note that the reliability of various positioning models (such as GNSS, vision, and laser SLAM) varies systematically across different scenarios. Establishing a model positioning confidence threshold is crucial for enabling the system to adapt to environmental challenges. For example, in an "open road" scenario, GNSS signal quality is high, and its positioning results are typically very reliable. In this case, a higher confidence threshold can be set, allowing only the outputs of highly accurate models to be fully trusted. Conversely, when the system detects that a vehicle has entered an "urban canyon" scenario, anticipating that GNSS signals may be severely degraded due to obstruction and multipath effects, the confidence requirement for the GNSS model will be appropriately lowered (i.e., the threshold will be lowered). Simultaneously, the confidence threshold for models unaffected by radio signals, such as inertial navigation or visual odometry, may be increased, thus guiding the fusion algorithm to rely more on backup positioning sources in signal-fragmented areas.

[0099] Step A50: If the model location reliability score corresponding to the potential outlier model is less than the model location reliability threshold, then the potential outlier model is marked as an outlier model.

[0100] Understandably, the purpose of this is to perform a secondary review of potential outlier models. The first step is to quickly filter out outliers based on external consistency (comparison with results from other models). The second step then turns to examining internal reliability (quality parameters output by the model itself). If a model is marked as a spatial outlier, but its internal quality indicators show that its data is still reliable, the system may keep it in an observation state rather than directly removing it, in order to avoid erroneously discarding valuable location information due to the limitations of a single discrimination criterion. This enhances the ability to handle marginal cases.

[0101] It should be understood that this design significantly improves the system's robustness in complex environments. It doesn't simply make binary, either-or judgments, but rather introduces a more refined, multi-evidence-supported decision-making process. For example, at the moment of entering a tunnel, all GNSS models may simultaneously exhibit significant deviations due to signal loss and be initially classified as outliers, but the inertial navigation model, due to its self-consistency, can still maintain short-term accuracy. By re-verifying its internal inertial data, the system can confirm that it still has reference value, thus using it as the core positioning source during GNSS failures to ensure the continuity of positioning services.

[0102] Step S502: Based on the real-time driving scenario of the vehicle, perform positioning voting weight allocation on the remaining cooperative positioning models.

[0103] It should be noted that the aforementioned steps have completed the screening of outlier models, while this step focuses on assigning appropriate voting weights to the remaining cooperative localization models that have been preliminarily identified as "reliable".

[0104] It is understandable that the voting weights in this embodiment are directly derived from the credibility of each model in the current scenario. For example, models that perform stably in the current scenario (such as when the GNSS signal is strong) are given high voting weights, while models that are susceptible to interference (such as when Wi-Fi is blocked by multiple walls) are given lower voting weights. Step S503: The location voting weights are fused together with the model location output and the model location confidence score corresponding to the model location output to obtain the fused location result.

[0105] It should be noted that the inputs for this step include three key elements: the voting weights for each model's positioning determined in the previous step, the positioning output results calculated by each model itself (such as latitude and longitude coordinates), and the model positioning reliability score generated synchronously with the positioning results. The essence of this step is to weightedly synthesize this heterogeneous information, fully leveraging the complementary advantages of different models while ensuring data quality, ultimately outputting an optimal fused positioning result that surpasses any single model.

[0106] It should be understood that this dual-weighting mechanism (scene weight + real-time confidence) greatly enhances the system's adaptability and fault tolerance in dynamically changing environments. It ensures that the final result considers not only the applicability of the macroscopic scene but also the actual operating state of each model at a microscopic moment. For example, when a vehicle passes through a sparse forest, the scene weights of the GNSS and visual models may be roughly equal. However, if the confidence of the visual system decreases due to momentary strong light interference, the fusion algorithm will immediately reduce its influence, thereby effectively suppressing instantaneous errors and outputting a smooth, continuous, and reliable final positioning coordinate.

[0107] In this embodiment, the distance between the positioning results of each model and the spatial mean point is calculated, and potential outlier models are identified based on a preset variance ratio. Then, a second verification of the potential outlier models is performed using a model positioning confidence threshold dynamically determined by the real-time driving scenario. Models with confidence scores below the threshold are identified as outliers and removed. Subsequently, positioning voting weights are assigned to the remaining reliable models according to the current scenario, and finally, a weighted fusion is performed based on the weights, positioning results, and their confidence scores to output the final positioning result.

[0108] In summary, this technical solution, through an outlier removal mechanism based on both spatial consistency and internal confidence criteria, effectively distinguishes between overall accuracy degradation caused by common-cause errors and genuine model failures. This avoids erroneous removal of usable models in complex environments, thereby improving the system's robustness and usability. Furthermore, by dynamically binding voting weights to real-time driving scenarios and coordinating them with the model's real-time confidence during the final fusion stage, the system can intelligently switch the dominant positioning source in different environments, suppressing errors introduced by transient interference. These combined measures ultimately enable the fusion positioning system to maintain high accuracy, high continuity, and high reliability in various dynamically changing driving scenarios.

[0109] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the multi-model collaborative fusion localization method of this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0110] This application also provides a multi-model collaborative fusion positioning device; please refer to [reference needed]. Figure 5 The multi-model collaborative fusion positioning device includes: The model scheduling module 10 is used to determine multiple cooperative positioning models from a preset positioning model library based on the real-time driving scenario of the vehicle. The parallel positioning module 20 is used to input real-time environmental perception data into the collaborative positioning model to obtain the model positioning output result and the model positioning confidence score corresponding one-to-one with the model positioning output result. Outlier detection module 30 is used to perform outlier detection on the model localization output result to obtain the model localization outlier detection result; The localization fusion module 40 is used to determine a localization result fusion strategy based on the localization outlier detection results of the model. The positioning fusion module 40 is further configured to fuse the positioning outputs of the multiple models according to the positioning result fusion strategy and the model positioning confidence score to obtain a fused positioning result.

[0111] In one embodiment, the model scheduling module 10 is further configured to: obtain the multi-dimensional positioning capability distribution results of each positioning model in the positioning model library under the real-time driving scenario based on a preset model positioning capability query table; perform complementary pairing of the multi-dimensional positioning capability distribution results of each positioning model to obtain multiple candidate model pairing combinations, wherein each candidate model pairing combination includes at least three positioning models; perform resource-result cost-effectiveness evaluation on the candidate model pairing combinations, and determine the target pairing combination based on the evaluation results; and use the multiple positioning models of the target pairing results as the cooperative positioning model.

[0112] In one embodiment, the parallel positioning module 20 is further configured to preprocess the real-time environmental perception data according to the data modalities required by the cooperative positioning model, and generate standardized data streams that conform to the input format of each model; input each standardized data stream in parallel into the corresponding cooperative positioning model to obtain the original positioning output results of each cooperative positioning model; perform format unification processing on the original positioning output results to generate model positioning output results in a standard format, wherein the model positioning output results include longitude positioning results, latitude positioning results, altitude positioning results, and positioning covariance matrix.

[0113] In one embodiment, the parallel positioning module 20 is further configured to: obtain a positioning covariance matrix generated synchronously with the current positioning result based on the positioning output result; construct a characteristic equation based on the positioning covariance matrix and solve it to obtain the characteristic equation solution result; obtain the area of ​​the uncertainty ellipse of the positioning output result based on the characteristic equation solution result; determine weighting coefficients based on the real-time driving scenario of the vehicle and the changing trend of the area of ​​the uncertainty ellipse within a preset time period; and obtain the model positioning reliability score corresponding to the model positioning output result based on the weighting coefficients, the area of ​​the uncertainty ellipse, and historical positioning error parameters.

[0114] In one embodiment, the outlier detection module 30 is further configured to obtain the spatial distribution variance of the positioning results based on the model positioning output results of each cooperative positioning model; determine the spatial distribution variance threshold under the current scenario based on the real-time driving scenario of the vehicle; if the spatial distribution variance is greater than the spatial distribution variance threshold, it is determined that there are outliers in the model positioning results; if the spatial distribution variance is less than or equal to the spatial distribution variance threshold, it is determined that there are no outliers in the model positioning results.

[0115] In one embodiment, the outlier detection module 30 is further configured to: obtain the spatial mean point of the positioning results based on the model positioning output results of each cooperative positioning model; calculate the spatial distance between the positioning output results of each model and the spatial mean point; if the spatial distance is greater than the spatial distribution variance of a preset multiple, then mark the cooperative positioning model corresponding to the model positioning output result as a potential outlier model; determine the model positioning reliability threshold in the current scenario based on the real-time driving scenario of the vehicle; if the model positioning reliability score corresponding to the potential outlier model is less than the model positioning reliability threshold, then mark the potential outlier model as an outlier model.

[0116] In one embodiment, the positioning fusion module 40 is further configured to: remove outlier models from the collaborative positioning model if outliers exist in the model positioning results; allocate positioning voting weights to the remaining collaborative positioning models according to the real-time driving scenario of the vehicle; and fuse the positioning voting weights, the model positioning output results, and the model positioning confidence scores corresponding to the model positioning output results to obtain a fused positioning result.

[0117] The multi-model collaborative fusion positioning device provided in this application, employing the multi-model collaborative fusion positioning method described in the above embodiments, can solve the technical problem in the prior art of how to design a fusion positioning method that can intelligently coordinate multiple heterogeneous positioning models and has high fault tolerance for abnormal results. Compared with the prior art, the beneficial effects of the multi-model collaborative fusion positioning device provided in this application are the same as those of the multi-model collaborative fusion positioning method provided in the above embodiments, and other technical features in the multi-model collaborative fusion positioning device are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0118] This application provides a multi-model collaborative fusion positioning device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the multi-model collaborative fusion positioning method in Embodiment 1 above.

[0119] The following is for reference. Figure 6 The diagram illustrates a structure suitable for implementing a multi-model cooperative fusion positioning device according to the embodiments of this application. The multi-model cooperative fusion positioning device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), and in-vehicle terminals (e.g., in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. Figure 6 The multi-model collaborative fusion positioning device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0120] like Figure 6As shown, the multi-model cooperative fusion positioning device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the multi-model cooperative fusion positioning device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touch screen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 1003 including, for example, magnetic tape, hard disk, etc.; and communication devices 1009. Communication device 1009 allows the multi-model cooperative fusion positioning device to communicate wirelessly or wiredly with other devices to exchange data. Although the figure shows a multi-model cooperative fusion positioning device with various systems, it should be understood that implementation or possession of all the systems shown is not required. More or fewer systems may be implemented alternatively.

[0121] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0122] The multi-model collaborative fusion positioning device provided in this application, employing the multi-model collaborative fusion positioning method described in the above embodiments, solves the technical problem in the prior art of how to design a fusion positioning method capable of intelligently coordinating multiple heterogeneous positioning models and possessing high fault tolerance for abnormal results. Compared with the prior art, the beneficial effects of the multi-model collaborative fusion positioning device provided in this application are the same as those of the multi-model collaborative fusion positioning method provided in the above embodiments, and other technical features in this multi-model collaborative fusion positioning device are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0123] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0124] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0125] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the multi-model collaborative fusion localization method described in the above embodiments.

[0126] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), or any suitable combination thereof.

[0127] The aforementioned computer-readable storage medium may be included in the multi-model collaborative fusion positioning device; or it may exist independently and not be assembled into the multi-model collaborative fusion positioning device.

[0128] The aforementioned computer-readable storage medium carries one or more programs. When these programs are executed by the multi-model collaborative fusion positioning device, the multi-model collaborative fusion positioning device: determines multiple collaborative positioning models from a preset positioning model library based on the real-time driving scenario of the vehicle; inputs real-time environmental perception data into the collaborative positioning models to obtain model positioning output results and model positioning reliability scores corresponding one-to-one with the model positioning output results; performs outlier detection on the model positioning output results to obtain model positioning outlier detection results; determines a positioning result fusion strategy based on the model positioning outlier detection results; and fuses the multiple model positioning output results according to the positioning result fusion strategy and the model positioning reliability scores to obtain a fused positioning result.

[0129] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0130] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0131] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0132] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described multi-model collaborative fusion localization method. This solves the technical problem in the prior art of how to design a fusion localization method capable of intelligently coordinating multiple heterogeneous localization models and possessing high fault tolerance for abnormal results. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the multi-model collaborative fusion localization method provided in the above embodiments, and will not be repeated here.

[0133] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the multi-model collaborative fusion localization method described above.

[0134] The computer program product provided in this application solves the technical problem in the prior art of how to design a fusion localization method that can intelligently coordinate multiple heterogeneous localization models and has high fault tolerance for abnormal results. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the multi-model collaborative fusion localization method provided in the above embodiments, and will not be repeated here.

[0135] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A multi-model cooperative fusion positioning method, characterized in that, The multi-model cooperative fusion positioning method comprises: According to the real-time driving scene of the vehicle, a plurality of cooperative positioning models are determined from a preset positioning model library; Real-time environmental perception data is input into the cooperative positioning model to obtain a model positioning output result and a model positioning confidence score corresponding to the model positioning output result; Outlier detection is performed on the model positioning output result to obtain a model positioning outlier detection result; According to the model positioning outlier detection result, a positioning result fusion strategy is determined; According to the positioning result fusion strategy and the model positioning confidence score, the plurality of model positioning output results are fused to obtain a fused positioning result.

2. The multi-model cooperative fusion positioning method according to claim 1, characterized in that, According to the real-time driving scene of the vehicle, a plurality of cooperative positioning models are determined from a preset positioning model library, comprising: Based on a preset model positioning capability query table, the multi-dimensional positioning capability distribution results of each positioning model in the positioning model library under the real-time driving scene are obtained; The multi-dimensional positioning capability distribution results of each positioning model are complementarily paired to obtain a plurality of candidate model pairing combinations, the candidate model pairing combination comprising at least three positioning models; Resource-result cost ratio evaluation is performed on the candidate model pairing combination, and a target pairing combination is determined according to the evaluation result; The plurality of positioning models of the target pairing result are used as the cooperative positioning models.

3. The multi-model cooperative fusion positioning method according to claim 1, characterized in that, The real-time environmental perception data is input into the cooperative positioning model to obtain a model positioning output result, comprising: According to the data modality required by the cooperative positioning model, the real-time environmental perception data is preprocessed to generate standardized data streams conforming to the input format of each model; Each standardized data stream is input into the corresponding cooperative positioning model in parallel to obtain the original positioning output result of each cooperative positioning model; The original positioning output result is subjected to format unification processing to generate a model positioning output result in a standard format, the model positioning output result comprising a longitude positioning result, a latitude positioning result, an altitude positioning result, and a positioning covariance matrix.

4. The multi-model cooperative fusion positioning method according to claim 1, characterized in that, The real-time environmental perception data is input into the cooperative positioning model to obtain a model positioning output result, comprising: According to the positioning output result, a positioning covariance matrix is generated synchronously with the current positioning result; wherein the positioning covariance matrix is: , is a variance of the longitude error, is a variance of the latitude error, is a covariance of the longitude error and the latitude error; A characteristic equation is constructed according to the positioning covariance matrix and is solved to obtain a characteristic equation solution result; According to the characteristic equation solution result, an uncertainty ellipse area of the positioning output result is obtained; According to the real-time driving scene of the vehicle and the change trend of the uncertainty ellipse area within a preset time period, a weight coefficient is determined; According to the weight coefficient, the uncertainty ellipse area, and historical positioning error parameters, a model positioning confidence score corresponding to the model positioning output result is obtained.

5. The multi-model cooperative fusion positioning method according to claim 1, characterized in that, According to the model positioning output result of each cooperative positioning model, a spatial distribution variance of the positioning result is obtained; ​ determining a spatial distribution variance threshold under a current scene according to the real-time driving scene of the vehicle; if the spatial distribution variance is greater than the spatial distribution variance threshold, determining that there is an outlier in the model positioning result; if the spatial distribution variance is less than or equal to the spatial distribution variance threshold, determining that there is no outlier in the model positioning result.

6. The multi-model cooperative fusion positioning method according to claim 5, characterized in that, The method further includes, after the step of determining that there is an outlier in the model positioning result if the spatial distribution variance is greater than the spatial distribution variance threshold: obtaining a spatial mean point of the positioning result according to the model positioning output results of the respective collaborative positioning models; calculating spatial distances between the model positioning output results and the spatial mean point; if the spatial distance is greater than a preset multiple of the spatial distribution variance, marking the collaborative positioning model corresponding to the model positioning output result as a potential outlier model; determining a model positioning confidence threshold under a current scene according to the real-time driving scene of the vehicle; if the model positioning confidence score of the potential outlier model is less than the model positioning confidence threshold, marking the potential outlier model as an outlier model.

7. The multi-model cooperative fusion positioning method according to claim 1, characterized in that, The method further includes, after the step of fusing the multiple model positioning output results according to the positioning result fusion strategy and the model positioning confidence score to obtain a fused positioning result: if there is an outlier in the model positioning result, eliminating the outlier model in the collaborative positioning model; distributing positioning voting weights to the remaining collaborative positioning models according to the real-time driving scene of the vehicle; fusing the positioning voting weight distribution result, the model positioning output result, and the model positioning confidence score corresponding to the model positioning output result to obtain a fused positioning result.

8. A multi-model cooperative fusion positioning device, characterized in that, The multi-model collaborative fusion positioning device includes: a model scheduling module configured to determine multiple collaborative positioning models from a preset positioning model library according to a real-time driving scene of a vehicle; a parallel positioning module configured to input real-time environmental perception data to the collaborative positioning models to obtain model positioning output results and model positioning confidence scores corresponding to the model positioning output results; an outlier detection module configured to perform outlier detection on the model positioning output results to obtain model positioning outlier detection results; a positioning fusion module configured to determine a positioning result fusion strategy according to the model positioning outlier detection results; The positioning fusion module is further configured to fuse the multiple model positioning output results according to the positioning result fusion strategy and the model positioning confidence scores to obtain a fused positioning result.

9. A multi-model cooperative fusion positioning device, characterized in that, The multi-model collaborative fusion positioning device includes a memory, a processor, and a multi-model collaborative fusion positioning program stored on the memory and executable on the processor, and the multi-model collaborative fusion positioning program is configured to implement the steps of the multi-model collaborative fusion positioning method according to any one of claims 1 to 7.

10. A storage medium, characterized by The storage medium stores a multi-model collaborative fusion positioning program, and the multi-model collaborative fusion positioning program is executed by the processor to implement the steps of the multi-model collaborative fusion positioning method according to any one of claims 1 to 7.

Images