A map information updating method, device, apparatus, and storage medium

Abstract

The present disclosure provides a map information updating method and device, equipment and storage medium, relates to the technical field of computers, in particular to the technical field of data processing, multi-modal large model, map, automatic driving, intelligent transportation and the like. The specific implementation scheme is as follows: a target driving track is determined; according to the target driving track, a track feature is determined, and based on the track feature, a target construction section is determined; based on the road network information and the real scene image of the target construction section, a multi-modal large model is used to determine the construction information of the target construction section; and based on the construction information, the map information of the target construction section is updated.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and in particular to the fields of data processing, multimodal large models, maps, autonomous driving, and intelligent transportation. Background Technology

[0002] Road construction is a major cause of dynamic changes in road network topology and traffic capacity. The dynamic updating of road construction status is crucial to the operational efficiency of intelligent transportation systems and the quality of map navigation services. In complex urban road networks, construction often leads to lane reductions, road closures, or temporary changes in traffic flow, causing static map data to quickly become invalid and unable to meet real-time navigation needs. Therefore, there is an urgent need for a method that can accurately and promptly detect road construction events and dynamically update map information accordingly to improve the reliability of navigation guidance. Summary of the Invention

[0003] This disclosure provides a method, apparatus, device, and storage medium for updating map information.

[0004] According to one aspect of this disclosure, a method for updating map information is provided, comprising: Determine the target driving trajectory; Based on the target driving trajectory, determine the trajectory characteristics, and based on the trajectory characteristics, determine the target construction section; Based on the road network information and real-scene images of the target construction section, the construction information of the target construction section is determined using a multimodal large model. Based on the construction information, the map information of the target construction section is updated.

[0005] According to another aspect of this disclosure, a map information updating apparatus is provided, comprising: The trajectory determination module is used to determine the target driving trajectory; The road segment determination module is used to determine the trajectory characteristics based on the target driving trajectory, and to determine the target construction road segment based on the trajectory characteristics; The construction information determination module is used to determine the construction information of the target construction section based on the road network information and real-scene images of the target construction section, using a multimodal large model. The map update module is used to update the map information of the target construction section based on construction information.

[0006] According to another aspect of this disclosure, an electronic device is provided, comprising: At least one processor; and The memory is communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform any of the methods described in the present disclosure.

[0007] According to another aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are used to cause the computer to perform any of the methods according to embodiments of this disclosure.

[0008] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements any of the methods according to embodiments of this disclosure.

[0009] This disclosure utilizes the trajectory features of the target driving trajectory to determine the target construction section, filtering out normal traffic areas, reducing the cost of calling subsequent high-computing models and data processing latency, and improving the efficiency of road construction perception. Furthermore, based on the road network information and real-world images of the target construction section, a multimodal large model is used to determine the construction information of the target construction section, enabling a deep understanding of the semantic logic of the construction scene and achieving a refined expression of construction information, thereby improving the accuracy and efficiency of map information updates. In addition, this refined expression of construction information can reduce navigation misjudgments caused by information ambiguity, thus improving the reliability of navigation guidance in road construction scenarios.

[0010] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0011] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein: Figure 1 This is a schematic diagram of a construction site using a passageway; Figure 2 This is a schematic diagram of the road diversion construction scenario; Figure 3 This is a schematic diagram illustrating an application scenario according to an embodiment of this disclosure; Figure 4 This is a flowchart illustrating the implementation of a map information updating method according to an embodiment of the present disclosure; Figure 5 This is a flowchart illustrating a method for updating map information according to an embodiment of the present disclosure; Figure 6 This is a schematic diagram of preprocessing an initial driving trajectory according to an embodiment of the present disclosure. Figure 1 ; Figure 7This is a schematic diagram of preprocessing an initial driving trajectory according to an embodiment of the present disclosure. Figure 2 ; Figure 8 This is a schematic diagram of preprocessing an initial driving trajectory according to an embodiment of the present disclosure. Figure 3 ; Figure 9 This is a schematic diagram of the structure of a map information updating device 900 according to an embodiment of the present disclosure; Figure 10 A schematic block diagram of an example electronic device 1000 that can be used to implement embodiments of the present disclosure is shown. Detailed Implementation

[0012] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0013] The term "and / or" in this disclosure indicates that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. The term "at least one" in this document means any combination of at least two of a plurality of options, such as including at least one of A, B, and C, which can mean including any one or more elements selected from the set of A, B, and C. The terms "first" and "second" in this document refer to and distinguish multiple similar technical terms, and do not imply a specific order or a limitation to only two. For example, "first feature" and "second feature" refer to two types / two features; the first feature can be one or more, and the second feature can also be one or more.

[0014] Road construction, as a core element affecting real-time traffic capacity, is crucial to the operational efficiency of intelligent transportation systems and the quality of map navigation services. For consumers (C-end users), accurate road construction information provides a strong sense of safety, ensuring a smooth travel experience while serving as a safety warning and alleviating driving anxiety. For business users (B-end users), road construction data is related to the safety of intelligent transportation; accurate road construction data not only improves the safety of autonomous driving but also provides key assistance and safety assurance for vehicle decision-making. Therefore, accurately detecting road construction events and updating map information accordingly is crucial for building intelligent transportation systems and affects the effectiveness of map navigation services in guiding user behavior.

[0015] Based on road construction elements, road construction events can be categorized into types such as lane borrowing, detour construction, and road closure. In complex construction scenarios involving lane borrowing (such as temporarily using the opposite lane), users may avoid the relevant road sections due to safety concerns or lack of information. To ensure users' travel experience, safety, and efficiency, there is an urgent need to enhance the detailed representation of road construction and dynamically update map information accordingly.

[0016] To enhance real-time awareness of road construction events, it is necessary to standardize construction semantic expressions, optimize navigation guidance strategies, and implement differentiated route planning. For example, personalized guidance can be provided based on user profiles or historical behavior: for users unwilling to detour, dynamic toll penalties can be imposed on detour sections of construction roads, thus providing users with alternative routes; for users willing to detour, the detour planning scheme can be retained, while enhancing voice notifications and visual guidance to improve traffic safety and efficiency.

[0017] In summary, current products demand more refined representations of road construction information, especially in complex scenarios such as lane borrowing and detour construction. Improving the granularity and real-time nature of road construction data can provide high-precision, interpretable traffic decision support for autonomous driving and intelligent navigation, thereby enhancing the accuracy of path planning in high-risk, high-uncertainty, and complex scenarios. In one example, a complex road construction scenario could include... Figure 1 The construction scene shown is a road borrowing project and Figure 2 The diversion construction scenario is shown. For example... Figure 1 As shown, there is a two-way four-lane road. Due to construction on the two right lanes (such as the two southbound lanes), the direction of travel in the second lane on the left is reversed (changing from northbound to southbound) to allow vehicles to pass. Figure 2 As shown, the original lane on the left needs to be constructed and vehicles cannot pass. At this time, vehicles can be guided to the recommended lane on the right, thereby realizing the diversion construction of the left road.

[0018] In existing technologies, the discovery of road construction events and the updating of map information mainly rely on image recognition technology or massive amounts of vehicle trajectory data. The typical process is as follows: suspected road sections are identified by detecting trajectory features, which are then manually verified and confirmed to generate road construction data, and finally the data is published and the map information is updated.

[0019] In one example, the method of updating map information based on image recognition technology includes: using real-scene images collected by the vehicle to extract key construction elements such as barriers, cones, and water-filled barriers using a recognition model; then using image feature difference technology to detect anomalies to identify road sections suspected of being under construction; after manual verification and confirmation, the construction results data is finally generated and published, and the map information is updated accordingly.

[0020] In another example, the method of updating map information based on vehicle trajectory data includes: using vehicle trajectory features (such as traffic surges (i.e., vehicle trajectories suddenly decrease or disappear in a certain road segment) and route detours (i.e., vehicles concentrate on abnormal turning, U-turns, or deviations from normal paths when approaching construction points), etc.), using machine learning models or preset rules to discover road segment traffic anomalies (such as construction, temporary traffic control closures, etc.), which are then manually verified and confirmed, and finally the construction results data are generated and published, and the map information is updated accordingly.

[0021] The aforementioned prior art has the following main drawbacks: The method of updating map information based on image recognition technology is heavily dependent on the coverage of the data collection vehicle, and it is easy to misjudge construction elements outside the road area as valid events, thereby increasing the cost of manual verification.

[0022] The method of updating map information based on vehicle trajectory data is limited by low positioning accuracy and noise from non-motorized vehicle trajectories, making it difficult to support automated operations. This results in a reliance on large-scale manual verification, which is not only costly and slow in updating, but can also only depict the construction status at the road level (i.e., can only determine whether "road construction" exists), and cannot achieve a refined representation of road construction.

[0023] Because multi-source data has failed to be integrated and standardized, it is difficult to support high-precision capture of road construction events. To achieve a refined representation of construction data that accurately depicts the real world, higher demands are placed on manual verification. Manual verification requires comprehensive comparison of real-time maps, historical data, and external information sources to confirm the authenticity, type, and specific impact range of road construction. This process inevitably leads to a surge in workload and slow update speed, making it difficult to meet the need for timely updates on road construction across a large area.

[0024] To address the aforementioned problems, this disclosure proposes a map information updating method. Figure 3 This is a schematic diagram illustrating an application scenario according to an embodiment of this disclosure, such as... Figure 3As shown in the illustration, the application scenario diagram of this disclosure may include, but is not limited to, a road construction sensing device 310 and a map service device 320. The road construction sensing device 310 and the map service device 320 can communicate via any type of wired or wireless network. Specifically, the road construction sensing device 310 can be used to sense roads under construction in the real world and determine the road construction information based on the sensed information. The map service device 320 can receive the construction information and update the map information accordingly. This disclosure does not impose a specific limitation on the number of road construction sensing devices 310. For example, the application scenario diagram of this disclosure may include one or more road construction sensing devices 310.

[0025] Figure 4 This is a flowchart illustrating the implementation of a map information updating method according to an embodiment of the present disclosure, including: S410, Determine the target driving trajectory; S420. Based on the target driving trajectory, determine the trajectory characteristics, and based on the trajectory characteristics, determine the target construction section; S430. Based on the road network information and real-scene images of the target construction section, the construction information of the target construction section is determined using a multimodal large model. S440. Based on construction information, update the map information for the target construction section.

[0026] In this embodiment of the disclosure, the target driving trajectory can be obtained by preprocessing the acquired initial driving trajectory. The initial driving trajectory can be vehicle driving data obtained from in-vehicle devices, mobile terminal navigation devices, etc. The process of preprocessing the initial driving trajectory will be described in detail in the following content.

[0027] Furthermore, this disclosure can determine the trajectory characteristics of a target driving trajectory based on the target driving trajectory. Here, the trajectory characteristics can reflect the real-time changes in the traffic capacity of each road segment in the target driving trajectory (such as changes in vehicle speed, steering, etc.). In the embodiments of this disclosure, for each road segment in the target driving trajectory, a fusion weight can be pre-set, and then the trajectory characteristics can be fused using the fusion weight to obtain the health score of each road segment. This health score can be used to determine the target construction road segment. The method for determining the target construction road segment will be described in detail in the following sections.

[0028] In this embodiment, construction information for the target construction section can be determined using a multimodal large model based on road network information and real-world images of the target construction section. Here, the multimodal large model can be an intelligent decision-making core with visual perception and logical reasoning capabilities; essentially, it is a neural network reasoning model capable of simultaneously understanding unstructured visual data and structured text data. In one example, this disclosure can construct prompts based on road network information and real-world images of the target construction section, and the multimodal large model can reason about the construction information of the target construction section based on these prompts. For example, during the reasoning phase, the multimodal large model can utilize a cross-attention mechanism to align or compare visual evidence in the real-world images with the road network information to determine the construction information of the target construction section.

[0029] Furthermore, this disclosure can update the map information of the target construction section based on construction information. In one example, this disclosure can modify the attributes of the target construction section in the map navigation service based on the construction information, for example, changing the number of lanes of the target construction section from 4 to 3, or marking the target construction section as "closed". It should be noted that the construction information output by the multimodal large model can include the construction type and the corresponding confidence level. In one example, if the confidence level is greater than or equal to a preset threshold, the construction information can be considered correct, and the map information of the target construction section can be updated directly using the construction information; if the confidence level is less than the preset threshold, the construction information needs to be sent to a human for review, and the review result is received and the map information of the target construction section is updated according to the review result. If the review result indicates that the construction information is correct, the map information of the target construction section can be updated directly using the construction information; if the review result indicates that the construction information is incorrect, the map information of the target construction section does not need to be updated.

[0030] By employing the above method, this disclosure utilizes the trajectory features of the target driving trajectory to determine the target construction section, filtering out normal traffic areas, reducing the cost of calling subsequent high-computing models and data processing latency, and improving the efficiency of road construction perception. Furthermore, based on the road network information and real-world images of the target construction section, a multimodal large model is used to determine the construction information of the target construction section, enabling a deep understanding of the semantic logic of the construction scene and achieving a refined expression of construction information, thereby improving the accuracy and efficiency of map information updates. In addition, this refined expression of construction information can reduce navigation misjudgments caused by information ambiguity, thus improving the reliability of navigation guidance in road construction scenarios.

[0031] To overcome the problems of low signal-to-noise ratio, coarse perception granularity, and high cost of manual verification of vehicle trajectory data, this disclosure proposes a map information update method driven by the trajectory of intelligent driving vehicles and combined with a multimodal large model (such as a Vision Language Model (VLM)).

[0032] Figure 5 This is a flowchart illustrating a method for updating map information according to an embodiment of this disclosure. Figure 5 As shown, this disclosure can use the trajectory preprocessing module 510 to preprocess the acquired initial driving trajectory to obtain the target driving trajectory; then, the feature extraction module 520 is used to extract features from the target driving trajectory to obtain the trajectory features of each road segment in the target driving trajectory, and then the construction excavation module 530 is used to determine the target construction section (i.e., the section suspected of being under construction); subsequently, the real-scene image of the target construction section collected on-site and the road network information corresponding to the target construction section are called to perform cross-modal information fusion, and based on the fused information (i.e., prompt words), the multimodal large model 540 is used to align the information of the target construction section and to identify the construction information of the target construction section, such as automatically identifying key elements such as construction barriers and cones, and determining the specific construction type, and then using the construction information to update the map information of the target construction section.

[0033] The following content provides a detailed description of the implementation process of the map information update method.

[0034] In some implementations, determining the target driving trajectory includes: Obtain the initial driving trajectory, which includes multiple initial trajectory points; The initial driving trajectory is preprocessed to determine candidate trajectory points; Connect the candidate trajectory points to determine the target driving trajectory.

[0035] In this embodiment, due to malfunctions in the trajectory acquisition equipment and signal errors, the initial driving data obtained from in-vehicle devices, mobile terminal navigation systems, and other devices may be accompanied by noise and errors, typically manifested as sampling anomalies (such as lost trajectory points) and accuracy anomalies (such as position drift). Furthermore, stationary or low-speed driving segments generated by vehicles while waiting at red lights or in traffic congestion can create high-density redundant sampling points within a small area, interfering with the analysis of road construction events. Therefore, this disclosure requires processing such invalid trajectories to improve data quality.

[0036] In this embodiment of the disclosure, the obtained initial driving trajectory includes multiple initial trajectory points. This disclosure can utilize the trajectory preprocessing module 510 to preprocess the initial trajectory points to achieve preprocessing of the initial driving trajectory. This preprocessing process can include the removal of inferior trajectories, such as the removal of drift trajectory points and the removal of stationary trajectory points. The preprocessing method for the initial driving trajectory will be described in detail later.

[0037] After preprocessing the initial driving trajectory, invalid or erroneous initial trajectory points can be filtered out, and the remaining valid points can be considered as candidate trajectory points. Furthermore, in one example, this disclosure can connect the candidate trajectory points in chronological order to determine the target driving trajectory.

[0038] By adopting the above method, this disclosure can determine candidate trajectory points from multiple initial trajectory points of the initial driving trajectory through a preprocessing step, and then connect the candidate trajectory points to obtain the target driving trajectory, thereby improving the quality of the target driving trajectory and providing a reliable data foundation for the subsequent determination of the target construction section, thus improving the efficiency and accuracy of determining construction information.

[0039] The following content details how trajectory features are determined.

[0040] In some implementations, determining trajectory features based on the target driving trajectory includes: Identify at least one road chain contained in the target driving trajectory; At least one path chain is aggregated to obtain an aggregated trajectory; Feature extraction is performed on the aggregated trajectory to determine multiple trajectory features.

[0041] In this embodiment, if only a single link of the target driving trajectory is used as the determination object, it is easily affected by the uneven granularity of the link splitting, which leads to the jitter of the trajectory features. In order to improve the continuity and stability of road construction in the spatial dimension, this disclosure introduces a link aggregation processing mechanism to aggregate adjacent links, and then uses the feature extraction module 520 to calculate the trajectory features of the target driving trajectory.

[0042] In one example, this disclosure can identify at least one road chain based on the geometric features of the target driving trajectory. For example, if discrete line segments appear in a continuous trajectory, these discrete line segments can be considered road chains in the target driving trajectory. In this example, these road chains generally appear at locations where the trajectory amplitude changes significantly (such as locations of road construction).

[0043] Furthermore, this disclosure aggregates at least one road chain to obtain an aggregated trajectory. Specifically, this disclosure can merge at least one road chain that is geographically continuous and logically belongs to the same trajectory, based on the road chain's connectivity, the consistency of road attributes (such as road direction, number of lanes, etc.), and spatial adjacency, to obtain an aggregated trajectory. It can be understood that the aggregated trajectory can be considered as an optimized trajectory of the target driving trajectory, which presents a continuous target driving trajectory.

[0044] In this embodiment of the disclosure, the feature extraction module 520 can be used to extract features from the aggregated trajectory to determine multiple trajectory features. In this disclosure, trajectory features can characterize the driving behavior performed by the user in the road segment corresponding to the aggregated trajectory (i.e., the optimized target driving trajectory). In other words, trajectory features can be considered as a mapping between the state of the road segment corresponding to the aggregated trajectory and the user's driving behavior.

[0045] In some implementations, the trajectory features include at least one of lane centerline offset, lane crossing amount, and lane change ratio.

[0046] In this embodiment of the disclosure, lane centerline offset can refer to the vertical distance between the actual driving trajectory of a vehicle and the theoretical centerline of the corresponding lane in the current road segment. For example, in a road construction scenario, when a lane is occupied by construction, vehicles in that lane will squeeze into adjacent lanes, and their driving trajectories will shift to one side, thus providing a basis for judging the road construction status of the road segment. Lane crossing volume can refer to the total number of times a vehicle's trajectory line crosses the lane dividing line during driving. For example, in a scenario of lane borrowing construction, vehicles need to cross solid or dashed lines to enter the opposite lane, resulting in an increase in crossing volume. Lane change ratio can refer to the proportion of the entire trajectory where lane changes have occurred to the total trajectory. This ratio can be the ratio between trajectory lengths or the ratio between the number of trajectory points. For example, in a scenario of road closure construction, road construction makes the original lanes of the road segment completely unusable, and all vehicles must change lanes, which increases the lane change ratio of the road segment.

[0047] In one example, the feature extraction module 520 can spatially register the aggregated trajectory with the lane-level layer of the high-precision map, and use a projection algorithm to calculate the lateral deviation sequence of the trajectory points of the aggregated features relative to the lane centerline, thereby statistically determining the lane centerline offset. At the same time, the feature extraction module 520 calculates the lane crossing volume by detecting the topological intersection events between the aggregated trajectory and the lane boundary line. In addition, the feature extraction module 520 can determine the lane change ratio by calculating the ratio of the length (or number of trajectory points) of the trajectory with lane changes to the total length (or number of trajectory points) of all trajectories in the aggregated trajectory.

[0048] In this embodiment of the disclosure, multiple determined trajectory features can be used to verify the authenticity of road construction in the future, while driving the map navigation service to shift from "static rules" to a closed-loop implementation of "dynamic perception-decision-feedback".

[0049] By employing the above approach, this disclosure introduces a road chain aggregation strategy, reconstructing adjacent discrete road chains with consistent attributes into spatially continuous and semantically complete aggregated trajectories, reducing feature abrupt noise caused by road chain boundary cutting. Furthermore, feature calculation is performed based on the smooth and continuous aggregated trajectories, enabling key indicators such as lane centerline offset and lane crossing volume to accurately reflect the overall driving behavior of vehicles on the road segment. This process improves the signal-to-noise ratio and robustness of trajectory features, providing a data foundation for subsequent determination of target construction sections and deep inference of multimodal large models, thereby enhancing the accuracy and timeliness of road construction event detection.

[0050] Furthermore, this disclosure can combine trajectory features to perform anomaly detection on road segments, thereby determining the target construction road segment. The following content details the method for determining the target construction road segment.

[0051] This disclosure can fuse multiple trajectory features, namely, the multi-dimensional trajectory features of lane centerline offset, lane crossing volume, and lane change ratio, to construct the health of road segments in the target driving trajectory, thereby identifying and determining target construction sections; and then using the target construction sections as high-priority target areas to dynamically trigger real-time acquisition of real-scene images and multi-modal large model inference process.

[0052] In some implementations, the target construction section is determined based on trajectory characteristics, including: Multiple trajectory features are fused to determine the health of each segment in the target driving trajectory; Based on the health status of each road segment, the target construction section is determined.

[0053] In some implementations, multiple trajectory features are fused to determine the health of each segment of the target driving trajectory, including: For each road segment, the trajectory features corresponding to the road segment are fused using preset fusion weights to determine the feature fusion result of the road segment; The health of a road segment is determined based on the feature fusion results of the road segment.

[0054] In this embodiment of the disclosure, the preset fusion weights are configured differently based on different trajectory characteristics (i.e., lane centerline offset, lane crossing volume, and lane change ratio) to differentiate the sensitivity of road construction events. For example... Figure 5As shown, for each segment in the target driving trajectory, the construction excavation module 530 can call the preset fusion weights and use the preset fusion weights to perform linear weighted fusion of the trajectory features of multiple dimensions corresponding to the segment, generating a feature fusion result that can reflect the degree of driving abnormality of the segment.

[0055] Furthermore, the construction and excavation module 530 can utilize a mapping function to convert the feature fusion results into a health score characterizing the traffic quality and safety status of a road segment. In one example, the stronger the abnormal signal in the feature fusion results of a road segment, the lower the health score of that road segment.

[0056] In this embodiment of the disclosure, the construction excavation module 530 can determine the target construction section based on the health status. For example, a lower health status means a more serious deviation of the trajectory from the normal state, and a greater likelihood of road construction.

[0057] In one example, this disclosure can pre-set a road segment health threshold. The construction excavation module 530 compares the road segment health with the road segment health threshold to determine the target construction road segment. For example, if the road segment health is lower than the road segment health threshold, or if the road segment shows a significant downward trend, then the road segment is determined to be an abnormal road segment affected by road construction, and thus the road segment is identified as the target construction road segment.

[0058] Using the above method, multi-dimensional trajectory features of road segments are fused using preset fusion weights to generate feature fusion results that reflect the traffic status of road segments. Furthermore, the feature fusion results are transformed into intuitive health scores, realizing a dynamic health assessment of the traffic quality of road segments. Finally, target construction road segments are determined based on the health scores, improving the accuracy of identifying suspected construction areas and providing a data foundation for the determination of subsequent construction information.

[0059] In some embodiments, this disclosure also includes: Build a feature database; The trajectory features and the corresponding target driving trajectory are saved to the feature database.

[0060] In this embodiment, a feature database supporting spatiotemporal indexing can be designed to achieve the accumulation and instant retrieval of historical data. In one example, the core of constructing the feature database lies in designing a storage architecture capable of responding simultaneously to both temporal and spatial dimensions. In other words, the feature database created in this disclosure includes a composite index mechanism using spatial windows as geographical partitions and time windows as sorting keys. This composite index mechanism can lock all data records for a specific road segment within a specific time period (such as "the past 7 days").

[0061] In this embodiment of the disclosure, saving the trajectory features and the corresponding target driving trajectory to the feature database can refer to binding and storing the calculated trajectory features with the target driving trajectory that generated the trajectory features.

[0062] For example, during the data entry stage, this disclosure can implement a spatial alignment and temporal archiving mechanism: spatially, trajectory features can be automatically mapped to specific road segments within the target driving trajectory, ensuring the continuity and consistency of trajectory features at different times on the same road segment in physical space; temporally, trajectory features can carry accurate time tags and be sequentially categorized into the corresponding time windows. This spatiotemporal dual-dimensional storage mechanism can improve the efficiency of subsequent trajectory feature retrieval based on the target driving trajectory. For instance, when performing a trajectory feature retrieval task such as "past 7 days," it is not necessary to traverse the entire feature database; only all trajectory features within the specified time window for the designated road segment need to be directly extracted, thereby allowing the calculation of the road segment's health.

[0063] In this way, by creating a feature database and saving trajectory features and corresponding target driving trajectories into the feature database, we can provide real-time road condition awareness for the navigation engine, provide long-term evolution trend basis for map updates, and provide rich behavioral data support for user profile construction.

[0064] The following content details how construction information is determined.

[0065] In some implementations, based on road network information and real-world images of the target construction section, a multimodal large model is used to determine the construction information of the target construction section, including: Based on the road network information and real-scene images of the target construction section, prompt words are constructed; Input the prompt word into the multimodal large model, and the multimodal large model will determine the construction information of the target construction section.

[0066] In this embodiment, the multimodal large model can be a visual language large model (VLM). In one example, this disclosure can introduce parameter optimization techniques such as instruction tuning and low-rank adaptation (LoRA) to optimize the VLM for domain adaptation. In this example, the optimization process can transform professional road construction judgment standards and specifications into a structured instruction set, guiding the VLM to deeply internalize the semantic logic and discrimination rules of the construction scenario. This allows the VLM to maintain its generalization ability while following industry standards for reasoning, thereby achieving high-precision and interpretable multimodal semantic understanding in complex construction scenarios.

[0067] like Figure 5As shown, this disclosure uses the road network information (such as road network structure and attributes) of the target construction section as prior knowledge, combines it with recently collected or real-time collected real-scene image sequences of the target construction section, constructs prompt words, and uses the prompt words to drive a multimodal large model 540 to perform cross-modal reasoning and high-precision semantic discrimination, thereby determining the construction information of the target construction section. Here, the construction information may include structured construction intelligence information, such as construction type, confidence level corresponding to the construction type, construction location, etc.

[0068] Understandably, the target construction section is determined by trajectory features, which can reflect the real-time dynamic changes in road capacity. Real-scene images can intuitively present the physical attributes of the target construction section, such as the number of lanes, the distribution of median strips, signs and markings, and the roadside environment. Cue words (the content used to drive the multimodal large model 540 to analyze construction information) further provide key contextual clues to help identify specific elements such as construction signs, cones, and water-filled barriers, thereby jointly constructing a comprehensive perception of the real environment of the road section.

[0069] By employing the above method, prompt words are determined based on road network information and real-world images, providing contextual information as a reasoning benchmark for the multimodal large model. This enables the model to align road network information with real-world images, thereby identifying construction information for the target road section. This improves the accuracy and timeliness of determining construction information, providing data support for subsequent map updates.

[0070] It should be noted that this disclosure requires processing of the road network information and real-scene images of the target construction section to construct prompt words.

[0071] In some implementations, prompts are constructed based on road network information and real-world images of the target construction section, including: The road network information of the target construction section is encoded to obtain the encoded road network information, which includes descriptive information about the road network information. Based on the encoded road network information and real-world images, prompt words are constructed.

[0072] In this embodiment of the disclosure, the process of encoding the road network information of the target construction section refers to converting the structured attributes of the target construction section into descriptive text in natural language form, i.e., descriptive information about the road network. For example, the road network information of the target construction section may include: the name of the target construction section is "XX Expressway"; the direction of travel is southbound; the location reference is K12+500 to K13+800 (i.e., from kilometer 12.5 to kilometer 13.8 of the expressway); the two leftmost fast lanes are closed; and the speed limit is 40 kilometers per hour. After encoding the road network information of the target construction section, the resulting descriptive information of the road network can be: "In the southbound direction of XX Expressway, from kilometer marker K12+500 to K13+800, due to road construction, the two leftmost fast lanes are temporarily closed, and traffic control with a speed limit of 40 kilometers per hour is implemented in this area."

[0073] In one example, the descriptive information about the road network can provide a basis for determining the construction type. Here, construction types can include road occupancy construction, lane borrowing construction, diversion construction, road widening construction, road closure construction, and roadside construction. In road occupancy construction, road construction occupies one or more lanes of a road segment, causing some lanes to be impassable, while the remaining lanes maintain their original driving direction. In lane borrowing construction, one side of the road is completely or partially closed, and the opposite lane or adjacent non-motorized vehicle lane is temporarily used for vehicle passage. In diversion construction, the original road segment is completely unusable, requiring the formation of a temporary passage through other surrounding roads. In road widening construction, to increase road capacity, the roadbed is widened to both sides or one side of the existing road, and new lanes are added. In road closure construction, the entire road segment is completely closed to all vehicles. In roadside construction, construction activities are limited to the area on the side of the road and do not affect the normal passage of motor vehicles.

[0074] This disclosure can encode the road network information of the target construction section to determine the descriptive information of the road network information, and then construct prompt words based on the descriptive information of the road network information and real scene images, so as to enable the multimodal large model 540 to infer the construction type of the target construction section.

[0075] In one example, this disclosure can construct prompts based on encoded road network information and real-world images. The prompts can include role settings, background information, visual information, definitions of various construction types, and task instructions. For example, this disclosure can define the role of the multimodal large model 540 as a traffic expert, inject encoded road network information (i.e., descriptive information about the road network) as background information into the prompts, and inject real-world images as visual information into the prompts. Finally, the multimodal large model 540 is required to compare the background information and the visual information to determine the construction information (including construction type, confidence level corresponding to the construction type, and construction location, etc.).

[0076] In some implementations, the descriptive information for road network information includes at least one of lane number information, lane direction information, lane traffic status information, and roadside attribute information.

[0077] In this embodiment of the disclosure, lane quantity information can be used as a basis for distinguishing between road occupation construction and road widening construction. For example, after encoding the road network information of the target construction section, the lane quantity information of the target construction section is determined. The multimodal large model 540 can compare the lane quantity information with the actual passable lane width and number in the real scene image to identify "lane reduction" caused by construction occupation or "lane increase" caused by expansion, and thus determine whether the construction type is road occupation construction or road widening construction.

[0078] Lane direction information can be used as a basis for determining lane borrowing for construction, as it clarifies the direction of travel for each lane in the target construction section. For example, by encoding the road network information of the target construction section, the lane direction information of the target construction section is determined. The multimodal large model 540 can then compare the lane direction information with the direction of travel in the real-world image to identify the "change in lane direction" caused by road construction, and thus determine the construction type as lane borrowing construction.

[0079] Lane traffic status information can be used as a basis for distinguishing between detour construction and road closure construction. For example, after encoding the road network information of the target construction section, the lane traffic status information of the target construction section is determined. The multimodal large model 540 can compare the lane traffic status information with the traffic status of the road in the real scene image to identify "road interruption" caused by road construction, and then determine the construction type as closure construction or detour construction based on the subsequent trajectory of the vehicle.

[0080] Roadside attribute information can be used as a basis for determining roadside construction. For example, roadside attribute information can describe the status of sidewalks, green belts, and building setback zones. Multimodal large model 540 can determine whether construction activities encroach on the motor vehicle lane based on roadside attribute information. If the construction activities do not encroach on the motor vehicle lane, the construction type can be determined to be roadside construction.

[0081] The following examples illustrate the content and format of prompt words.

[0082] In one example, after encoding the road network information of the target construction section, a prompt word can be constructed based on the encoded road network information (i.e., descriptive information about the road network) and a real-world image. This prompt word can include five parts: (1) Role setting: You are a professional traffic condition analysis expert.

[0083] (2) Background information: The current target construction section is an urban arterial road. The original design was a two-way six-lane road. Currently, there are 3 lanes from west to east and 2 lanes from east to west. The lanes in the two directions are separated by double yellow lines. Vehicles are prohibited from crossing the center line. There are no temporary access roads.

[0084] (3) Visual information: that is, the real-scene images of the target construction section collected.

[0085] (4) Definitions of various construction types, namely, the corresponding definitions of road occupation construction, road borrowing construction, road diversion construction, road widening construction, road closure construction and roadside construction.

[0086] (5) Task instructions: Compare background information and visual information, and based on the definition of each construction type, determine the construction information of the current target construction section (construction type, confidence level of construction type and construction location, etc.), and output it in text format.

[0087] By adopting the above method, this disclosure encodes the road network information of the target construction section into a natural language description containing dimensions such as the number of lanes, direction, traffic status and roadside attributes, and co-constructs prompt words with real-scene images, thereby improving the semantic understanding accuracy and logical reasoning ability of the multimodal large model in the construction scenario, and thus improving the accuracy of the multimodal large model in determining construction information, providing reliable data support for updating map information.

[0088] The following content details how map information is updated.

[0089] In some implementations, the construction information includes the construction type and the confidence level corresponding to the construction type; Based on construction information, the map information of the target construction section is updated, including: If the confidence level corresponding to the construction type is greater than or equal to the preset threshold, the map information of the target construction section is updated based on the construction information. If the confidence level corresponding to the construction type is less than the preset threshold, the verification result of the construction information is received, and the map information of the target construction section is updated according to the verification result.

[0090] In this embodiment of the disclosure, the confidence level corresponding to the construction type can refer to the quantitative score of the credibility or certainty of the construction type when the multimodal large model 540 outputs a specific construction type (such as road borrowing construction, road occupation construction, etc.). It is usually a value between 0 and 1, reflecting how confident the multimodal large model 540 is that the target construction section belongs to a certain specific construction type based on the encoded road network information and real scene images.

[0091] In the embodiments disclosed herein, such as Figure 5As shown, the confidence level determination module 550 can be used to determine the update method for map information. In one example, this disclosure can set an appropriate threshold (i.e., a preset threshold) in the confidence level determination module 550, and then the confidence level determination module 550 can compare the confidence level corresponding to the construction type with the preset threshold to determine the update method for map information.

[0092] In this embodiment of the disclosure, when the confidence level determination module 550 determines that the confidence level corresponding to the construction type is greater than or equal to a preset threshold, the subsequent map update process can be directly triggered. That is, this disclosure can utilize the map information update module 570 to directly update the map information based on the construction information. In one example, the map information update module 570 can, based on the construction information, perform real-time corrections to the map information of the target construction section, such as adjusting lane direction, number of lanes, updating traffic attributes, or marking temporary traffic control status.

[0093] In this embodiment, if the confidence level determination module 550 determines that the confidence level corresponding to the construction type is less than a preset threshold, the construction information is ambiguous, and the map information is not directly updated. In this case, the construction information can be sent to the review module 560, which then initiates a review process involving a human or advanced expert system. In one example, reviewers or advanced algorithms will combine more contextual information to perform a secondary determination of the construction information, generating a clear review result. Further, in one example, the review module 560 can determine whether to proceed with the subsequent map information update process based on the review result. For example, if the review result indicates that the construction information is incorrect, the review module 560 may not send the construction information to the map information update module 570; if the review result indicates that the construction information is correct, the review module 560 may send the construction information to the map information update module 570, which will receive the review result and proceed with the subsequent map information update process accordingly.

[0094] Using the above method, this disclosure implements a confidence level strategy based on the construction information output by the multimodal large model, and constructs a differentiated processing closed loop that combines "direct updating of map information with high-confidence construction information" with "manual review of low-confidence construction information". This realizes full lifecycle management of road construction events from online release and real-time updates to offline removal, and improves the timeliness and accuracy of map information.

[0095] This disclosure, through multi-source data fusion and refined spatiotemporal processing, enables the establishment of an update mechanism based on detailed construction information in map navigation services. This mechanism not only allows path planning algorithms to generate differentiated traffic strategies (such as implementing specific route optimization for different scenarios like lane borrowing and closures), but also enhances the accuracy of traffic guidance reminders. For example, this disclosure can reduce vague descriptions and instead adopt a tiered warning mechanism, triggering voice prompts 1 kilometer, 500 meters, and 200 meters before the vehicle enters the construction area, respectively. Simultaneously, the construction area markers are displayed on the navigation interface, overlaid with lane-level guidance lines, thereby achieving comprehensive visual guidance from macro-level warnings to micro-level lane directions, improving driving safety and traffic efficiency.

[0096] The following section details how to preprocess the initial driving trajectory.

[0097] In some implementations, the initial driving trajectory is preprocessed to determine candidate trajectory points, including: From multiple initial trajectory points of the initial driving trajectory, identify drift trajectory points, where the distance between a drift trajectory point and its adjacent initial trajectory point is greater than or equal to a preset distance threshold. Other initial trajectory points besides the drift trajectory points are identified as candidate trajectory points.

[0098] Figure 6 This is a schematic diagram of preprocessing an initial driving trajectory according to an embodiment of the present disclosure. Figure 1 In this embodiment, each initial trajectory point in the initial driving trajectory can be traversed, and the distance (such as Euclidean distance) between each trajectory point and its adjacent initial trajectory points can be calculated. Generally, this disclosure uses the same sampling frequency for the initial trajectory points, meaning that the distance between adjacent initial trajectory points is the same. If the distance between an initial trajectory point and its adjacent initial trajectory points is greater than or equal to a preset distance threshold, then the initial trajectory point can be determined to be a drift trajectory point caused by signal loss or equipment failure. This drift trajectory point typically exhibits a long-distance jump that violates the laws of physical motion within a very short time.

[0099] like Figure 6 As shown, there are six initial trajectory points, namely A, B, C, D, E, and F. By calculating the distance between each initial trajectory point and its adjacent initial trajectory points, it can be determined that initial trajectory point C is a drift trajectory point. In other words, the distance between the drift trajectory point (i.e., initial trajectory point C) and its adjacent initial trajectory points (i.e., initial trajectory points B and D) is greater than or equal to a preset distance threshold. Furthermore, this disclosure can eliminate the drift trajectory point (i.e., initial trajectory point C) and determine the remaining initial trajectory points (i.e., A, B, D, E, and F) as candidate trajectory points.

[0100] In one example, given that the initial trajectory points are essentially point samples representing the road surface features, exhibiting spatial clustering (i.e., under normal driving conditions, the initial trajectory points present a continuous, high-density banded distribution on the road), this disclosure can utilize a density-based spatial clustering algorithm (DBSCAN) to determine drift trajectory points based on the distribution pattern of the initial trajectory points. In this example, this disclosure can divide closely connected initial trajectory points within a high-density area into "core clusters," representing the actual driving path, while initial trajectory points far from the core clusters can be identified as drift trajectory points. In this approach, a preset distance threshold can be used as the basis for clustering the initial trajectory points. For example, if the distance between adjacent initial trajectory points is less than the preset distance threshold, the adjacent initial trajectory points can be grouped into one category; if the distance between adjacent initial trajectory points is greater than or equal to the preset distance threshold, then one of the adjacent initial trajectory points may be a drift trajectory point.

[0101] By using the above method, this disclosure identifies drift trajectory points by the distance between adjacent points, which achieves accurate filtering of noise data in the initial driving trajectory, improves the continuity and smoothness of trajectory point data, reduces the interference of drift trajectory points on subsequent trajectory feature calculations, and provides a highly reliable data foundation for road construction excavation.

[0102] In some implementations, the initial driving trajectory is preprocessed to determine candidate trajectory points, including: If the number of initial trajectory points within the preset range is greater than or equal to the number threshold, the target trajectory point is determined within the preset range. The target trajectory points within the preset range and the initial trajectory points outside the preset range are determined as candidate trajectory points.

[0103] Figure 7 This is a schematic diagram of preprocessing an initial driving trajectory according to an embodiment of the present disclosure. Figure 2 .like Figure 7 As shown, when the vehicle is stationary or traveling at low speed, redundant initial trajectory points may be generated within a certain range (i.e., Figure 7 (The black markers in the image) These redundant initial trajectory points have similar timestamps in the initial driving trajectory, or their sampling positions are repeated or too close.

[0104] In this embodiment of the disclosure, if the number of initial trajectory points gathered within a preset range is greater than or equal to a threshold, it indicates that the vehicle is stationary or traveling at low speed within that area, thus generating redundant initial trajectory points. Furthermore, this disclosure can utilize a compression algorithm to determine a target trajectory point from the preset range, such as determining the geometric center point, time midpoint, first initial trajectory point, and last initial trajectory point within the preset range as the target trajectory point, and deleting other initial trajectory points within the preset range, thereby achieving thinning and smoothing of the trajectory data within the preset range.

[0105] Finally, this disclosure can merge the target trajectory points determined after compressing the preset range with the initial trajectory points outside the preset range to jointly determine the candidate trajectory points.

[0106] By employing the above method, this disclosure compresses high-density initial trajectory points within a preset range into representative single target trajectory points, while retaining initial trajectory points outside the preset range, and determining candidate trajectory points accordingly. This process reduces redundant data while preserving the vehicle's parking position information, reducing trajectory distortion caused by dense initial trajectory points, and thus improving the accuracy and efficiency of subsequent trajectory feature calculations.

[0107] In some implementations, the initial driving trajectory is preprocessed to determine candidate trajectory points, including: Identify initial trajectory points that are not located on the road from multiple initial trajectory points of the initial driving trajectory; Using a time-series probability model, initial trajectory points that are not located within a road are matched to the corresponding road to obtain matched trajectory points; Candidate trajectory points are determined based on the matched trajectory points and the initial trajectory points located on the road in the initial driving trajectory.

[0108] Figure 8 This is a schematic diagram of preprocessing an initial driving trajectory according to an embodiment of the present disclosure. Figure 3 .like Figure 8 As shown, due to multiple factors such as positioning errors, coordinate system transformation deviations, and limitations in map accuracy, vehicle trajectory data often exhibits spatial offset, i.e. Figure 8 The initial trajectory points shown are not located within the road network. To accurately determine the vehicle's actual position within the road network topology, map matching processing is required.

[0109] In this embodiment of the disclosure, a spatial geometric verification is first performed on each initial trajectory point in the initial driving trajectory. In one example, by overlaying and comparing the coordinates of the initial trajectory points with the road network layer of a high-precision map, initial trajectory points that are not located within the road can be filtered out.

[0110] Furthermore, this disclosure can employ a temporal probability model to match initial trajectory points not located within roads to their corresponding roads, thereby obtaining matched trajectory points, i.e. Figure 8 The image shows the initial trajectory points within the road after matching. In one example, the temporal probability model can be a Hidden Markov Model (HMM).

[0111] Specifically, Hidden Markov Models can correct the initial trajectory points that are not located on the road, rather than simply projecting the nearest neighbor. The process of matching the initial trajectory point with the road includes two core probability calculation processes: (1) calculating the emission probability, that is, calculating the geometric distance probability from the initial trajectory point that is not located on the road to the candidate road segment. Generally speaking, the closer the distance, the higher the probability that the initial trajectory point is located on the candidate road segment; (2) calculating the transition probability, that is, combining the time series context, calculating the logical probability of transitioning from the matched road segment at the previous time to the current candidate road segment. In one example, calculating the transition probability requires considering the topological connectivity between road segments, the consistency of driving direction, and whether the calculated speed between the two points conforms to physical laws.

[0112] Furthermore, the Hidden Markov Model, through the Viterbi algorithm, can find the optimal path among all possible road segment combinations that maximizes the product of the "emission probability" and the "transition probability," thereby accurately matching the initial trajectory point that is not located on the road to the most logical real road and generating the matching trajectory point.

[0113] Finally, this disclosure can perform spatiotemporal fusion of the matched trajectory points with the initial trajectory points located within the road to determine candidate trajectory points.

[0114] By adopting the above approach, this disclosure introduces a matching algorithm based on a temporal probability model, which not only considers the geometric distance between the initial trajectory point not located on the road and the road, but also incorporates the spatiotemporal continuity constraints of vehicle motion. Through global optimization, the initial trajectory point not located on the road is matched to the corresponding road, thereby reducing the geometric distortion caused by the initial trajectory point positioning noise and improving the accuracy and reliability of subsequent trajectory feature calculations.

[0115] Furthermore, in this embodiment of the disclosure, the candidate trajectory points obtained after preprocessing the initial driving trajectory are connected in chronological order to determine the target driving trajectory.

[0116] This disclosure addresses key pain points such as high costs of professional data collection, delayed data updates, and the difficulty of accurately and efficiently updating construction information using existing trajectory mining and image recognition technologies. It proposes a map information update method based on the fusion of intelligent driving vehicle trajectories and real-world images. This method introduces a Visual Language Model (VLM) and reduces noise interference from single trajectory data through multi-source information fusion, achieving high-precision automated updates of construction information. This not only reduces the cost of manual review but also improves the real-time update performance of construction information.

[0117] This disclosure represents not only a technological iteration and upgrade, but also a leap forward in service paradigm. Through deep integration with navigation systems, a five-dimensional intelligent road condition perception hub integrating "people, vehicles, road sections, environment, and construction events" is constructed, providing highly reliable, adaptable, and personalized map navigation services. It has practical value in areas such as improving traffic management efficiency, upgrading personalized services, and assisted driving vehicle-road cooperation, enhancing the accuracy and stability of construction event judgment in complex scenarios. In the future, this technology can be extended to a wider range of traffic scenarios, evolving from single construction detection to intelligent prediction across all scenarios, becoming a core infrastructure for next-generation intelligent navigation and smart transportation systems.

[0118] This disclosure also proposes a map information updating device. Figure 9 This is a schematic diagram of the structure of a map information updating device 900 according to an embodiment of the present disclosure, including: The trajectory determination module 910 is used to determine the target driving trajectory; The road segment determination module 920 is used to determine the trajectory characteristics based on the target driving trajectory, and to determine the target construction road segment based on the trajectory characteristics; The construction information determination module 930 is used to determine the construction information of the target construction section based on the road network information and real-scene images of the target construction section, using a multimodal large model. Map update module 940 is used to update the map information of the target construction section based on construction information.

[0119] In some implementations, the construction information determination module 930 is used for: Based on the road network information and real-scene images of the target construction section, prompt words are constructed; Input the prompt words into the multimodal large model, and the multimodal large model will determine the construction information of the target construction section.

[0120] In some implementations, the construction information determination module 930 is used for: The road network information of the target construction section is encoded to obtain encoded road network information, which includes descriptive information about the road network information. Based on the encoded road network information and real-world images, prompt words are constructed.

[0121] In some implementations, the descriptive information for road network information includes at least one of lane number information, lane direction information, lane traffic status information, and roadside attribute information.

[0122] In some implementations, the construction information includes the construction type and the confidence level corresponding to the construction type; Map update module 940 is used for: If the confidence level corresponding to the construction type is greater than or equal to the preset threshold, the map information of the target construction section is updated based on the construction information. If the confidence level corresponding to the construction type is less than the preset threshold, the verification result of the construction information is received, and the map information of the target construction section is updated according to the verification result.

[0123] In some implementations, the road segment determination module 920 is used for: Identify at least one road chain contained in the target driving trajectory; At least one path chain is aggregated to obtain an aggregated trajectory; Feature extraction is performed on the aggregated trajectory to determine multiple trajectory features.

[0124] In some implementations, the trajectory features include at least one of lane centerline offset, lane crossing amount, and lane change ratio.

[0125] In some implementations, the road segment determination module 920 is used for: Multiple trajectory features are fused to determine the health of each segment in the target driving trajectory; Based on the health status of each road segment, the target construction section is determined.

[0126] In some implementations, the road segment determination module 920 is used for: For each road segment, the trajectory features corresponding to the road segment are fused using preset fusion weights to determine the feature fusion result of the road segment; The health of a road segment is determined based on the feature fusion results of the road segment.

[0127] In some implementations, the trajectory determination module 910 is used for: Obtain the initial driving trajectory, which includes multiple initial trajectory points; The initial driving trajectory is preprocessed to determine candidate trajectory points; Connect the candidate trajectory points to determine the target driving trajectory.

[0128] In some implementations, the trajectory determination module 910 is used for: From multiple initial trajectory points of the initial driving trajectory, identify drift trajectory points, where the distance between the drift trajectory point and its adjacent initial trajectory points is greater than or equal to a preset distance threshold. Other initial trajectory points besides the drift trajectory points are identified as candidate trajectory points.

[0129] In some implementations, the trajectory determination module 910 is used for: If the number of initial trajectory points within the preset range is greater than or equal to the number threshold, the target trajectory point is determined within the preset range. The target trajectory points within the preset range and the initial trajectory points outside the preset range are determined as candidate trajectory points.

[0130] In some implementations, the trajectory determination module 910 is used for: Identify initial trajectory points that are not located on the road from multiple initial trajectory points of the initial driving trajectory; Using a time-series probability model, initial trajectory points that are not located within a road are matched to the corresponding road to obtain matched trajectory points; Candidate trajectory points are determined based on the matched trajectory points and the initial trajectory points located on the road in the initial driving trajectory.

[0131] The specific functions and examples of each module and submodule of the apparatus in this disclosure can be found in the relevant descriptions of the corresponding steps in the above method embodiments, and will not be repeated here.

[0132] The acquisition, storage, and application of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0133] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0134] Figure 10 A schematic block diagram of an example electronic device 1000 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0135] like Figure 10 As shown, device 1000 includes a computing unit 1001, which can perform various appropriate actions and processes according to a computer program stored in read-only memory (ROM) 1002 or a computer program loaded into random access memory (RAM) 1003 from storage unit 1008. The RAM 1003 may also store various programs and data required for the operation of device 1000. The computing unit 1001, ROM 1002, and RAM 1003 are interconnected via bus 1004. Input / output (I / O) interface 1005 is also connected to bus 1004.

[0136] Multiple components in device 1000 are connected to I / O interface 1005, including: input unit 1006, such as keyboard, mouse, etc.; output unit 1007, such as various types of monitors, speakers, etc.; storage unit 1008, such as disk, optical disk, etc.; and communication unit 1009, such as network card, modem, wireless transceiver, etc. Communication unit 1009 allows device 1000 to exchange / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0137] The computing unit 1001 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 1001 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 1001 performs the various methods and processes described above, such as map information updating methods. For example, in some embodiments, the map information updating method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 1008. In some embodiments, part or all of the computer program may be loaded and / or installed on device 1000 via ROM 1002 and / or communication unit 1009. When the computer program is loaded into RAM 1003 and executed by the computing unit 1001, one or more steps of the map information updating method described above may be performed. Alternatively, in other embodiments, the computing unit 1001 may be configured to perform a map information update method by any other suitable means (e.g., by means of firmware).

[0138] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0139] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0140] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on 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 of the foregoing.

[0141] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0142] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0143] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0144] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0145] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for updating map information, comprising: Determine the target driving trajectory; Based on the target driving trajectory, determine the trajectory characteristics, and based on the trajectory characteristics, determine the target construction section; Based on the road network information and real-scene images of the target construction section, the construction information of the target construction section is determined using a multimodal large model; Based on the construction information, the map information of the target construction section is updated.

2. The method according to claim 1, wherein, Based on the road network information and real-scene images of the target construction section, and using a multimodal large model, the construction information of the target construction section is determined, including: Based on the road network information and real-scene images of the target construction section, prompt words are constructed; The prompt words are input into the multimodal large model, and the multimodal large model determines the construction information of the target construction section.

3. The method according to claim 2, wherein, The prompt words are constructed based on the road network information and real-scene images of the target construction section, including: The road network information of the target construction section is encoded to obtain encoded road network information, which includes descriptive information about the road network information. The prompt words are constructed based on the encoded road network information and the real-scene image.

4. The method according to claim 3, wherein, The descriptive information for the road network information includes at least one of the following: lane number information, lane direction information, lane traffic status information, and roadside attribute information.

5. The method according to any one of claims 1-4, wherein, The construction information includes the construction type and the confidence level corresponding to the construction type; The step of updating the map information of the target construction section based on the construction information includes: If the confidence level corresponding to the construction type is greater than or equal to a preset threshold, the map information of the target construction section is updated based on the construction information. If the confidence level corresponding to the construction type is less than a preset threshold, the verification result of the construction information is received, and the map information of the target construction section is updated according to the verification result.

6. The method according to any one of claims 1-5, wherein, The step of determining trajectory features based on the target driving trajectory includes: Identify at least one road chain contained in the target driving trajectory; The at least one path chain is aggregated to obtain an aggregated trajectory; Feature extraction is performed on the aggregated trajectory to determine multiple trajectory features.

7. The method according to claim 6, wherein, The trajectory features include at least one of lane centerline offset, lane crossing amount, and lane change ratio.

8. The method according to any one of claims 1-7, wherein, The determination of the target construction section based on the trajectory features includes: Multiple trajectory features are fused to determine the health of each segment in the target driving trajectory; Based on the health status of each road segment, the target construction road segment is determined.

9. The method according to claim 8, wherein, The process of fusing multiple trajectory features to determine the health of each segment in the target driving trajectory includes: For each of the aforementioned road segments, the trajectory features corresponding to the road segments are fused using preset fusion weights to determine the feature fusion result of the road segments; Based on the feature fusion results of the road segment, the health of the road segment is determined.

10. The method according to any one of claims 1-9, wherein, Determining the target driving trajectory includes: Obtain an initial driving trajectory, which includes multiple initial trajectory points; The initial driving trajectory is preprocessed to determine candidate trajectory points; The candidate trajectory points are connected to determine the target driving trajectory.

11. The method according to claim 10, wherein, The preprocessing of the initial driving trajectory to determine candidate trajectory points includes: From the plurality of initial trajectory points of the initial driving trajectory, a drift trajectory point is identified, wherein the distance between the drift trajectory point and the adjacent initial trajectory point is greater than or equal to a preset distance threshold. Other initial trajectory points besides the drift trajectory points are determined as candidate trajectory points.

12. The method according to claim 10, wherein, The preprocessing of the initial driving trajectory to determine candidate trajectory points includes: If the number of initial trajectory points within a preset range is greater than or equal to a number threshold, then a target trajectory point is determined within the preset range. The target trajectory points within the preset range and the initial trajectory points outside the preset range are determined as the candidate trajectory points.

13. The method according to claim 10, wherein, The preprocessing of the initial driving trajectory to determine candidate trajectory points includes: From the plurality of initial trajectory points of the initial driving trajectory, identify initial trajectory points that are not located on the road; Using a time-series probability model, the initial trajectory points that are not located on the road are matched to the corresponding roads to obtain the matched trajectory points; The candidate trajectory points are determined based on the matched trajectory points and the initial trajectory points located on the road in the initial driving trajectory.

14. A map information updating device, comprising: The trajectory determination module is used to determine the target driving trajectory; The road segment determination module is used to determine trajectory features based on the target driving trajectory, and to determine the target construction road segment based on the trajectory features; The construction information determination module is used to determine the construction information of the target construction section based on the road network information and real-scene images of the target construction section, using a multimodal large model. The map update module is used to update the map information of the target construction section based on the construction information.

15. An electronic device comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-13.

16. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-13.

17. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1-13.

Images