Traffic situation awareness method and device, electronic equipment and storage medium

Abstract

This invention provides a traffic situation perception method, device, electronic device, and storage medium, belonging to the field of intelligent transportation technology. The method includes: acquiring multi-dimensional heterogeneous data, including human-dimensional data, vehicle-dimensional data, road-dimensional data, and cloud-dimensional data; acquiring the target scene categories corresponding to the multi-dimensional heterogeneous data and determining the scene priorities corresponding to the target scene categories; dynamically configuring perception resources according to the scene priorities to obtain configured perception resources, which are used to characterize sensor sampling frequency, sensor cooperation mode, and redundant device on / off status; and performing traffic situation perception based on the configured perception resources to obtain traffic situation perception results. The above solution can achieve real-time and accurate assessment of traffic situations, providing lane-level refined information for traffic management departments, supporting accurate decision-making, and effectively alleviating road network congestion and accident risks.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation technology, specifically to a traffic situation perception method, device, electronic device, and storage medium. Background Technology

[0002] Existing traffic situation awareness technologies are mostly limited to the binary fusion of roadside sensor data and vehicle-side basic data, lacking key dimensions such as driver physiological status and cloud-based road network planning. Furthermore, the fusion algorithms are often simple superpositions, failing to address the conflict between spatiotemporal calibration of heterogeneous data and privacy protection. Simultaneously, they lack robustness in extreme scenarios, easily leading to data gaps or misjudgments, and cannot support accurate service delivery. This invention aims to solve the problems of low accuracy, poor real-time performance, and insufficient reliability in existing highway traffic situation awareness systems, which easily cause errors in traffic control decisions, exacerbate road network congestion and accident risks, and severely restrict the intelligent upgrading of highways and the large-scale implementation of vehicle-road cooperative technologies.

[0003] Therefore, developing a precise traffic situation perception method that adapts to complex scenarios and integrates the advantages of new technologies has become an urgent need for the construction of intelligent transportation on highways. Summary of the Invention

[0004] To address the aforementioned problems, this application provides a traffic situation perception method to improve the accuracy of traffic situation perception. The technical solution is as follows: In a first aspect, the present invention provides a traffic situation perception method, comprising: Acquire heterogeneous data across all dimensions, including data on people, vehicles, roads, and the cloud. Obtain the target scene categories corresponding to the heterogeneous data across all dimensions, and determine the scene priorities corresponding to the target scene categories; The sensing resources are dynamically configured according to the scenario priority to obtain the configured sensing resources. The sensing resources are used to characterize the sensor sampling frequency, sensor cooperation mode and the on / off status of redundant devices. Traffic situation perception is performed based on the configuration of perception resources, and the traffic situation perception results are obtained.

[0005] Combining the first aspect and the above implementation methods, in some possible implementation methods, heterogeneous data with full dimensions can be obtained, including: Acquire traffic flow data from the entire network and extract road segment feature data from the traffic flow data; Based on the road segment characteristic data, the congestion coefficient of a single road segment is calculated, and the congestion coefficient of the single road segment is weighted to obtain the road network congestion coefficient. Based on the road network congestion coefficient and road segment travel time, dynamic programming calculations are performed to obtain the optimal travel route cost; The road network congestion coefficient and the cost of the optimal driving route are defined as cloud-based data.

[0006] Combining the first aspect and the above implementation methods, some possible implementation methods, after obtaining heterogeneous data across all dimensions, include: Based on the reference time, the local timestamps in the heterogeneous data of all dimensions are time-synchronized and calibrated to obtain the calibrated timestamps; Transform the spatial coordinates in the heterogeneous data of all dimensions to the target coordinate system to obtain the calibrated spatial coordinates; By binding human dimension data with calibration spatial coordinates, an integrated human-vehicle spatial data block is obtained, which in turn associates human dimension data with road segment-level spatial units.

[0007] Combining the first aspect and the above implementation methods, in some possible implementation methods, the target scene category corresponding to the multi-dimensional heterogeneous data can be obtained, including: The heterogeneous data across all dimensions is matched against preset scene feature thresholds using rules to obtain the matching results; If the matching result has clear scene features, then the matched scene will be identified as the target scene; If the matching result is that the scene features are fuzzy, then the heterogeneous data of all dimensions will be input into the scene classification model to obtain the classification probability of each type of scene. The scene corresponding to the highest classification probability is determined as the target scene category.

[0008] Combining the first aspect and the above implementation methods, among some possible implementation methods, the scene priority of the target scene category is determined, including: Obtain the risk coefficient, impact range coefficient, and response timeliness coefficient for the target scenario category; The priority assessment value is obtained by weighting the risk coefficient, the scope of impact coefficient, and the response time coefficient. Based on the priority evaluation value, the scene priority of the target scene category is determined.

[0009] Combining the first aspect and the above implementation methods, some possible implementation methods involve dynamically configuring perception resources based on scene priority, including: If the scene priority is high priority, the sampling frequency of the core sensor will be increased to the first sampling frequency, and the multi-sensor collaborative sensing mode will be enabled and redundant devices will be turned on. If the scene priority is medium priority, the sampling frequency of the core sensor will be maintained at the second sampling frequency, and the key sensor collaborative sensing mode will be enabled and unnecessary sensors will be turned off. If the scene priority is low, the sampling frequency of the core sensor will be reduced to the third sampling frequency, and the independent sensing mode of a single core sensor will be enabled. In this mode, the first sampling frequency is greater than the second sampling frequency, and the second sampling frequency is greater than the third sampling frequency.

[0010] Combining the first aspect and the above implementation methods, some possible implementation methods dynamically adjust the perception resources according to the scene priority, including: Construct a resource optimization objective function, which aims to minimize total energy consumption and is constrained by ensuring that the perception accuracy meets the minimum perception accuracy requirement corresponding to the scene priority. Based on the resource optimization objective function, the first sampling frequency, the second sampling frequency, or the third sampling frequency is optimized and adjusted.

[0011] Combining the first aspect and the above implementation methods, in some possible implementation methods, heterogeneous data with full dimensions can be obtained, including: The traffic situation awareness model parameters are initialized through the privacy aggregation node and then sent to the client. Train the local model on the client to obtain the local model parameters; The client encrypts the local model parameters to obtain encrypted parameters, and then uploads the encrypted parameters to the privacy aggregation node. The encrypted parameters are aggregated based on privacy aggregation nodes to obtain updated global model parameters. The updated global model parameters are then used for iterative optimization until the global model loss function converges, resulting in the optimal traffic situation perception model, which outputs heterogeneous data across all dimensions.

[0012] In a second aspect, the present invention also provides a traffic situation awareness device, comprising: The data acquisition unit is used to acquire heterogeneous data across all dimensions, including data on people, vehicles, roads, and the cloud. The priority identification unit is used to obtain the target scene category corresponding to the heterogeneous data in all dimensions and determine the scene priority corresponding to the target scene category. The resource configuration unit is used to dynamically configure the sensing resources according to the scenario priority to obtain the configured sensing resources. The sensing resources are used to characterize the sensor sampling frequency, sensor cooperation mode and redundant device switching status. The result feedback unit is used to perform traffic situation perception based on the configured perception resources and obtain the traffic situation perception results.

[0013] Thirdly, the present invention also provides an electronic device, including a memory and a processor, wherein, Memory, used to store programs; The processor, coupled to the memory, executes the program stored in the memory to implement the steps in the traffic situation awareness method in any of the above implementations.

[0014] Fourthly, the present invention also provides a computer-readable storage medium for storing a computer-readable program or instructions, which, when executed by a processor, can implement the steps of the traffic situation perception method in any of the above implementations.

[0015] The beneficial effects of this invention are as follows: By integrating data from multiple dimensions and employing timestamp calibration and data fusion methods, this application solves the problems of single data dimensions and simple fusion algorithms in existing technologies, achieving effective alignment of multi-source data. Furthermore, by identifying the target scene categories corresponding to the heterogeneous data across all dimensions and dynamically adjusting the operation mode of sensing resources according to the scene priority of the target scene category, system efficiency is improved while ensuring sensing accuracy. Ultimately, this method achieves real-time and accurate assessment of traffic conditions, providing traffic management departments with lane-level refined information, supporting accurate decision-making, and effectively alleviating road network congestion and accident risks. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A schematic diagram of the system architecture of the traffic situation perception method provided by the present invention; Figure 2 A flowchart illustrating the traffic situation perception method provided by the present invention; Figure 3 A schematic diagram of a scenario for the traffic situation perception method provided by the present invention; Figure 4 A schematic diagram of a scenario for the traffic situation perception method provided by the present invention; Figure 5 A flowchart illustrating the traffic situation perception method provided by the present invention; Figure 6 A flowchart illustrating the traffic situation perception method provided by the present invention; Figure 7 A schematic diagram of the traffic situation perception device provided by the present invention; Figure 8 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0019] In the description of the embodiments of the present invention, unless otherwise stated, "multiple" means two or more. "And / or" describes the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0020] The terms "first," "second," etc., used in the embodiments of this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a technical feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature.

[0021] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0022] This invention provides a traffic situation awareness method, device, electronic device, and storage medium, which are described below.

[0023] Please see Figure 1 , Figure 1 This is a system architecture diagram of the traffic situation perception method provided in this application embodiment. The system architecture includes a server 110, a gateway 120, an Internet connection 130, and sensors 140, etc.

[0024] Server 110 refers to a computer system capable of providing certain services. Server 110 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms.

[0025] Gateway 120, also known as an internetwork connector or protocol converter, is a computer system or device that acts as a translator, enabling network interconnection at the transport layer. It bridges the gap between two systems using different communication protocols, data formats, languages, or even completely different architectures. Gateways can also provide filtering and security functions. Messages sent from sensor 140 to server 110 are forwarded to the corresponding server 110 via gateway 120. Similarly, messages sent from server 110 to sensor 140 are also forwarded to the corresponding sensor 140 via gateway 120.

[0026] Sensor 140 includes, but is not limited to, devices such as loop coil sensors, microwave radar sensors, image sensors, and infrared sensors used for real-time monitoring of road traffic flow, vehicle status, and environmental information. Furthermore, sensor 140 can be a single device or a collection of multiple devices. For example, multiple sensors 140 can be interconnected via a local area network to form a single sensor 140. Sensor 140 can communicate and exchange data with the Internet 130 via wired or wireless means.

[0027] Specifically, server 110 receives human-dimensional data and vehicle-dimensional data monitored by sensors 140 in the vehicle (including roadside coils, radars, cameras, weather stations, and in-vehicle OBD, cameras, radars, etc.), as well as road-dimensional data from roadside sensors 140 and cloud-dimensional data obtained from the cloud, which together constitute a full-dimensional heterogeneous data input.

[0028] Server 110 first performs time and space alignment on these multi-source data, and then calls its built-in hybrid traffic situational awareness model for processing: the traffic situational awareness model prioritizes efficient rule matching to process regular scenarios with clear thresholds, such as toll stations and tunnels, and then uses a lightweight CNN model to analyze ambiguous or sudden scenarios, ultimately accurately identifying the target scenario category and its priority for traffic segments. Based on the priority of the output scenarios, the server generates perception resource configuration instructions, which are sent to the sensor network 140 via the Internet 130 and gateway 120, dynamically adjusting its sampling frequency, cooperative mode, and redundancy switch.

[0029] Please see Figure 2 , Figure 2 This is a flowchart illustrating a traffic situation perception method provided in an embodiment of this application. Figure 2 As shown, the method in this application embodiment may include the following steps S201-S204: S201, acquire heterogeneous data across all dimensions.

[0030] In this embodiment, multi-dimensional heterogeneous data refers to a data set encompassing various aspects of the transportation system, acquired from multiple sources and in various formats. This multi-dimensional heterogeneous data includes human-dimensional data, vehicle-dimensional data, road-dimensional data, and cloud-dimensional data. Human-dimensional data may include information such as the driver's physiological state and behavioral characteristics; vehicle-dimensional data may include information such as vehicle speed, location, and operating status; road-dimensional data may include information such as road infrastructure, traffic flow, and environmental conditions; and cloud-dimensional data may include information such as historical traffic patterns, road network planning, and navigation route optimization. These data collectively form the foundation for a comprehensive perception of the traffic situation.

[0031] Specifically, human dimension data can be obtained by combining vehicle-mounted cameras with driver facial recognition technology, or by collecting the driver's physiological indicators through wearable devices; vehicle dimension data can be obtained by the vehicle's own data interface, vehicle-mounted radar, lidar, or cameras, or by roadside equipment such as roadside coil detectors and geomagnetic sensors.

[0032] Road dimension data can be obtained from roadside weather stations, video surveillance systems, traffic flow detectors, etc.; cloud dimension data can be obtained from cloud servers, which can be open data platforms or historical traffic databases of traffic management departments.

[0033] Furthermore, since human, vehicle, road, and cloud data originate from different sensors and data sources, inconsistencies in timestamps or spatial coordinate systems may exist between these different dimensions. Therefore, timestamp calibration of the collected data for each dimension is necessary. For example, a local timestamp can be assigned to each data source, and then time synchronization with a unified NTP server can convert all local timestamps into a common reference time. Simultaneously, all spatial coordinates can be uniformly converted to a preset geographic coordinate system.

[0034] Data fusion is performed on timestamped human, vehicle, road, and cloud data, using cross-node collaborative training and fusion to obtain multi-dimensional heterogeneous data. During data fusion, methods such as Kalman filtering or Bayesian networks can be used to perform weighted averaging or probability inference on different data types, thus obtaining multi-dimensional heterogeneous data.

[0035] S202, obtain the target scene category corresponding to the heterogeneous data in all dimensions, and determine the scene priority corresponding to the target scene category.

[0036] Specifically, the system invokes a pre-defined traffic situation awareness model to identify scenarios from heterogeneous data across all dimensions, thereby determining the target scenario category in which the current user is located and the corresponding scenario priority.

[0037] It should be noted that the training process of the traffic situation awareness model is divided into two stages: First, for regular scenarios with clear thresholds, such as ramps, tunnels, bridges, toll stations, and service areas, a feature database is built based on historical road network and sensor data. Priority rules and thresholds are defined through expert experience, enabling the traffic situation awareness model to directly match feature vectors and preset conditions during the rule matching stage, achieving rapid screening of high-priority scenarios. For ambiguous or sudden scenarios, such as traffic accidents, congestion, and severe weather, a lightweight CNN model is used for deep classification optimization. Its training is based on a large-scale labeled dataset. Through 14-dimensional features in the input layer, including fused road segment type, real-time traffic flow, vehicle speed, occupancy rate, and environmental parameters, spatial and temporal dependencies are learned through three hidden layers. Finally, the 12-dimensional Softmax output layer outputs a probability distribution. The model training uses cross-entropy loss and Adam optimizer iteration, and incorporates random noise and data augmentation to improve generalization ability. The overall model complements the real-time performance of the rule engine with the generalization ability of CNN, balancing speed and accuracy. It can dynamically match priority rules or output probability classifications based on real-time features, accurately identifying the target scenario category in the current area.

[0038] S203, dynamically configure the sensing resources according to the scene priority to obtain the configured sensing resources.

[0039] In this embodiment, sensing resources are used to characterize sensor sampling frequency, sensor coordination mode, and the on / off status of redundant devices. After identifying the target scene and determining its priority, the operation mode of the sensing resources needs to be adjusted to adapt to the current requirements.

[0040] Specifically, when the scene priority is high, all available sensors should be enabled for data acquisition and redundant devices should be activated to obtain the most comprehensive and accurate situational information. When the scene priority is medium, only a subset of sensors most critical to the current target scene category should be enabled to work collaboratively.

[0041] When the scenario priority is low, the sampling frequency of the core sensor needs to be reduced to the third sampling frequency, and a single core sensor independent sensing mode needs to be enabled. Here, the core sensor refers to the sensor that undertakes the main sensing task and provides key data sources in the traffic situation awareness system; it is usually the most important and highest-performing sensor type in the system. The single core sensor independent sensing mode means that only one core sensor is enabled for data acquisition and processing, while other auxiliary sensors are in standby or off state.

[0042] It should be noted that the first sampling frequency is higher than the second sampling frequency, and the second sampling frequency is higher than the third sampling frequency. The first, second, and third sampling frequencies can be preset according to the actual sensor performance, data processing capabilities, and the sensing accuracy requirements and energy consumption budget under different priority scenarios. For example, the first sampling frequency can be set to the sensor's maximum operating frequency to ensure the highest sensing accuracy; the second sampling frequency can be set to a balanced frequency that meets normal sensing needs; and the third sampling frequency can be set to the lowest frequency that meets basic sensing needs to maximize energy savings.

[0043] S204, based on the configuration of perception resources, traffic situation perception is performed to obtain traffic situation perception results.

[0044] Specifically, based on the configuration of perception resources, the entire network's traffic situation is assessed in real time and with high precision. This enables the output of detailed traffic situation perception results, such as real-time traffic flow, average speed, vehicle spacing, potential collision risk areas, and the distribution of vehicles with driver fatigue, accurate to the lane level. As a result, traffic management departments can obtain refined and real-time information, enabling them to make rapid decisions.

[0045] In summary, this application addresses the issues of single data dimensions and simple fusion algorithms in existing technologies by integrating multi-dimensional data and employing timestamp calibration and data fusion methods, achieving effective alignment of multi-source data. Through a hybrid recognition process combining rule matching and a lightweight CNN model, it rapidly matches priority scenarios based on rules while utilizing a CNN model to optimize the classification of ambiguous scenarios such as traffic accidents and sudden congestion, significantly improving the accuracy and robustness of identifying sudden scenarios and avoiding misjudgments. Furthermore, based on the priority of the identified scenarios, the operation mode of sensing resources is dynamically adjusted. All sensors and redundant equipment are activated in high-priority scenarios, while the sampling frequency of core sensors is reduced in low-priority scenarios, thereby improving system efficiency while ensuring perception accuracy. Ultimately, this method achieves real-time and accurate assessment of traffic conditions, providing traffic management departments with lane-level refined information to support accurate decision-making and effectively alleviate road network congestion and accident risks.

[0046] If the acquisition methods for cloud-based data are not comprehensive or accurate enough, subsequent traffic situation awareness results may be limited, making it difficult to accurately reflect macro-level traffic conditions and route selection costs. Therefore, please refer to [link / reference needed]. Figure 3 , Figure 3 This is a flowchart illustrating a traffic situation perception method provided in an embodiment of this application. Figure 3 As shown, the method in this application embodiment may include the following steps S301-S304: S301: Obtain traffic flow data for the entire network and extract road segment feature data from the traffic flow data for the entire network.

[0047] In this embodiment of the application, the network traffic flow data refers to real-time or historical traffic data covering all or most sections of the highway network, which may include, but is not limited to, information such as vehicle flow, vehicle speed, occupancy rate, and queue length.

[0048] Specifically, data can be collected in real time through various sensors deployed on highways, or obtained from third-party data sources such as floating car data, mobile phone signaling data, and navigation software data. Road segment feature data are key indicators extracted from network-wide traffic flow data to describe the traffic conditions of specific road segments. These features can include the segment's average speed, traffic volume, density, congestion index, and travel time. Extraction methods can involve data cleaning, aggregation, and statistical analysis to transform raw, fine-grained traffic flow data into representative road segment-level features. For example, the average speed of all vehicles passing through a road segment within a certain time period can be averaged to obtain the segment's average speed as road segment feature data.

[0049] S302 calculates the congestion coefficient of a single road segment based on road segment characteristic data, and then performs a weighted calculation on the congestion coefficient of the single road segment to obtain the road network congestion coefficient.

[0050] In this embodiment of the application, the single-segment congestion coefficient is an indicator for measuring the degree of traffic congestion in a single road segment.

[0051] Specifically, the congestion coefficient of a single road segment is calculated using various methods, such as the ratio of actual vehicle speed to free-flow speed, the ratio of traffic volume to capacity, or the travel time index. For example, when the vehicle speed on a road segment is below a certain threshold, the congestion coefficient can be set to a higher value. Weighted calculation refers to assigning different weights to the congestion coefficients of different road segments when calculating the overall road network congestion coefficient. These weights can be determined based on factors such as the importance of the road segment, such as arterial roads, connecting roads, length, traffic volume, and historical congestion frequency. For example, the main line of a highway can be assigned a higher weight to reflect its greater impact on the overall road network traffic conditions. Weighted calculation methods can employ various complex weighted models, such as simple arithmetic average weighting, entropy-based weighting, and the analytic hierarchy process (AHP).

[0052] S303 uses dynamic programming to calculate the optimal travel path cost based on the road network congestion coefficient and the travel time of the road segment.

[0053] In this embodiment, the road network congestion coefficient is a comprehensive indicator reflecting the overall congestion situation of the entire highway network. Road segment travel time refers to the time required for a vehicle to pass through a specific road segment; it can be a real-time measurement, a historical average, or a predicted value, and is affected by various factors such as road segment length, vehicle speed, and traffic conditions.

[0054] Specifically, it calls a preset dynamic programming algorithm, such as Degas's algorithm or A... The algorithm, along with other path search algorithms, combines the road network congestion coefficient and the travel time of road segments as the weights of the edges to calculate the cost of the optimal travel path.

[0055] S304 defines the road network congestion coefficient and the cost of the optimal driving route as cloud-based data.

[0056] Specifically, using road network congestion coefficients and optimal travel route costs as cloud-based data indicates that this data has undergone advanced computation and aggregation, providing comprehensive and decision-making information. This data can be stored on cloud servers for subsequent use by traffic situation awareness models to provide more comprehensive traffic condition assessments and predictions.

[0057] In summary, this application acquires traffic flow data from the entire highway network and extracts segment feature data to calculate the congestion coefficient of each segment. This congestion coefficient is then aggregated through weighted calculations to form the overall network congestion coefficient, thus comprehensively reflecting the overall traffic operation status of the highway network. By combining the network congestion coefficient and segment travel time, dynamic programming is used to calculate the optimal travel path cost under the current traffic conditions, providing important reference for users' travel decisions. By using these macro-level network congestion coefficients and optimal travel path costs as cloud-level data, the consideration of macro-level traffic operation and travel costs in multi-dimensional heterogeneous data is supplemented. This approach enables traffic situational awareness methods to understand and analyze traffic conditions from a broader perspective, focusing not only on the congestion of local segments but also on the connectivity and efficiency of the entire road network, as well as the potential costs for users when choosing routes. This provides more comprehensive and in-depth data support for subsequent accurate traffic situational awareness.

[0058] Because heterogeneous data across all dimensions originates from different sensors, systems, and platforms, these data often have their own independent local timestamps and spatial coordinate systems, leading to inconsistencies in time and space. This spatiotemporal mismatch severely impacts the accuracy of subsequent data fusion and the precision of traffic situation awareness, making it difficult to extract unified and reliable traffic information from massive amounts of heterogeneous data. Therefore, please refer to [link / reference needed]. Figure 4 , Figure 4 This is a flowchart illustrating a traffic situation perception method provided in an embodiment of this application. Figure 4 As shown, the method in this application embodiment may include the following steps S401-S403: S401 performs time synchronization calibration on local timestamps in heterogeneous data across all dimensions based on a reference time, resulting in a calibrated timestamp.

[0059] Specifically, the precise time synchronization service provided by the BeiDou Navigation Satellite System is used as the reference time to calibrate the local timestamps of various data sources in the heterogeneous data of all dimensions.

[0060] S402 transforms the spatial coordinates in the heterogeneous data of all dimensions to the target coordinate system to obtain the calibrated spatial coordinates.

[0061] Specifically, spatial coordinates from different data sources, such as GPS, inertial navigation systems, and map data, are uniformly converted to a preset target coordinate system using coordinate transformation algorithms. Alternatively, data obtained directly from BeiDou high-precision positioning services can be used, as it inherently possesses unified and high-precision spatial coordinates, serving as a benchmark for other data conversions.

[0062] S403 binds human dimension data with calibration spatial coordinates to obtain a human-vehicle spatial integrated data block, thereby associating human dimension data with road segment-level spatial units.

[0063] Specifically, the calibrated human-dimensional data is spatially overlaid with highway network geographic information system data to determine the road segment or intersection to which each human-dimensional data point belongs. Data containing calibrated timestamps, calibrated spatial coordinates, and integrated human-vehicle spatial data blocks are defined as calibrated data. This step defines the final data structure after spatiotemporal synchronization calibration, ensuring that all key spatiotemporal information and the spatial correlation of human-dimensional data are integrated into a unified data unit. This can be represented as a structured data record, where each record contains a unique calibrated timestamp, a set of calibrated spatial coordinates, and an integrated human-vehicle spatial data block describing the correlation between human-dimensional data and that spatiotemporal point.

[0064] In summary, this application utilizes the BeiDou spatiotemporal reference for precise time synchronization calibration and spatial coordinate transformation, ensuring high temporal and spatial alignment of information from different data sources and eliminating data discrepancies and conflicts. Simultaneously, binding human-dimensional data with calibrated spatial coordinates and linking it to road segment-level spatial units provides a clear geographic context for human activity, enabling the integration of multi-dimensional data such as people, vehicles, roads, and cloud within a unified and precise spatiotemporal framework. This not only significantly improves the quality and reliability of the calibrated data but also lays a solid foundation for subsequent cross-node collaborative training, data fusion, and traffic scene recognition, thereby achieving a more accurate and comprehensive perception of highway traffic conditions.

[0065] In one feasible implementation, in addition to acquiring heterogeneous data across all dimensions, the following specific steps are performed: The traffic situation awareness model parameters are initialized through the privacy aggregation node and then sent to the client. Train the local model on the client to obtain the local model parameters; The client encrypts the local model parameters to obtain encrypted parameters, and then uploads the encrypted parameters to the privacy aggregation node. The encrypted parameters are aggregated based on privacy aggregation nodes to obtain updated global model parameters. The updated global model parameters are then used for iterative optimization until the global model loss function converges, resulting in the optimal traffic situation perception model, which outputs heterogeneous data across all dimensions.

[0066] It's important to note that the client is a device capable of capturing vehicle, pedestrian, cloud, and road-level data. The privacy aggregation node is a core component of the federated learning architecture, its main function being to coordinate the client's training process and securely aggregate the encrypted parameters uploaded by the client. This privacy aggregation node typically does not directly access the raw data but instead processes the encrypted model updates. Its implementation can include servers based on trusted execution environments or aggregation servers employing secure multi-party computation protocols to ensure the security and privacy of the aggregation process. Traffic situation awareness model parameters refer to the set of parameters maintained by the privacy aggregation node during the federated learning process, representing the current state of the traffic situation awareness model, such as the weights and biases of the neural network. These parameters are initialized at the start of training and updated locally by the aggregation client in each iteration.

[0067] The client trains the model based on the dimensional data stored locally and uploads the model parameters obtained from the local training to the privacy aggregation node.

[0068] Local model training refers to the process by which a client independently trains a traffic situation awareness model using its local dimensional data after receiving the model's parameters. This process aims to adapt the model to the characteristics of the local data, generating local model parameters that reflect the distribution of that data. Local model training can employ various machine learning algorithms, such as deep learning and support vector machines, to optimize model performance. Local model parameters are the model weights, biases, and other parameters obtained by the client after completing local model training. These parameters reflect the model's learning results on specific client-side local data. Local model parameters form the basis for subsequent encryption processing and uploading to privacy aggregation nodes for aggregation.

[0069] Encryption refers to the technical means by which clients protect the privacy of local model parameters before uploading them. Its purpose is to prevent malicious parties from stealing or reverse-engineering local model parameters during transmission and aggregation, thereby leaking the privacy of the original data. Encryption can employ various techniques, such as differential privacy mechanisms, which obscure individual data contributions by adding noise; or homomorphic encryption, which allows computation on ciphertext without decryption.

[0070] Encrypted parameters refer to the local model parameters after being encrypted. These parameters exist in ciphertext form and cannot be directly deciphered by unauthorized parties, thus protecting the client's privacy. Encrypted parameters are the input for aggregation processing by the privacy aggregation node.

[0071] Aggregation processing refers to the process by which a privacy-aggregating node summarizes and integrates encrypted parameters uploaded by multiple clients. The purpose of aggregation processing is to extract common knowledge from the local learning results of each client and incorporate it into the traffic situation awareness model. Aggregation processing typically employs secure aggregation protocols, such as federated averaging algorithms, to complete the aggregation without decrypting the local model parameters. The updated traffic situation awareness model parameters refer to the new set of traffic situation awareness model parameters obtained by the privacy-aggregating node after completing the aggregation processing. These parameters integrate the local learning results of all participating clients and represent the latest state of the traffic situation awareness model in the current iteration.

[0072] The convergence of the loss function of a traffic situation awareness model refers to the gradual decrease and stabilization of the model's loss function value during iterative optimization, reaching the preset convergence condition. This typically indicates that the model's performance has reached a good level, and further training will yield limited performance improvements. The optimal traffic situation awareness model refers to the final model obtained after the loss function of the traffic situation awareness model has converged. This model has undergone multiple rounds of iterative optimization, incorporating knowledge from all clients, and possesses good generalization ability and predictive performance.

[0073] In another feasible implementation, the method further includes the steps of: adding differential privacy noise to the local model parameters to obtain noisy parameters; performing a first encryption process on the noisy parameters to obtain a first encrypted parameter; if the local model parameters correspond to human dimension data, then superimposing a homomorphic encryption mask on the first encrypted parameter to obtain a second encrypted parameter, so as to determine the second encrypted parameter as the encrypted parameter; if the local model parameters do not correspond to human dimension data, then the first encrypted parameter is determined as the encrypted parameter.

[0074] It should be noted that local model parameters refer to the set of model weights, biases, and other numerical values ​​obtained by the client locally using its calibrated data. Differential privacy noise is a technique that protects individual privacy by adding random noise to the data. Differential privacy noise can be added using various mechanisms. For example, random noise following a Laplace distribution can be added to the local model parameters based on a Laplace mechanism, or random noise following a Gaussian distribution can be added based on a Gaussian mechanism, to achieve privacy protection of the local model parameters while meeting a certain privacy budget.

[0075] The noisy parameters are the local model parameters after differential privacy processing, which obscures the contribution of individual data points while maintaining the overall statistical characteristics of the data. The first encryption process is a fundamental encryption operation performed on the noisy parameters to protect the confidentiality of the data during transmission and prevent unauthorized access. This encryption process can employ symmetric encryption algorithms, such as Advanced Encryption Standard (AES), which encrypts and decrypts data using a shared key; or it can employ asymmetric encryption algorithms, such as RSA, which uses a public key for encryption and a private key for decryption. The first encrypted parameters are the noisy parameters after the first encryption process.

[0076] Human-dimensional data involves highly sensitive information such as individual behavior, identity, or preferences, such as a driver's driving habits and travel routes. Homomorphic encryption masks refer to the use of homomorphic encryption technology to further protect the privacy of data. Homomorphic encryption allows direct computation on encrypted data without prior decryption, thus preventing the aggregation node from obtaining the original local model parameter information when aggregating encrypted parameters.

[0077] A homomorphic encryption mask can be applied to the first encryption parameter itself, or it can be generated as a homomorphic encryption mask and superimposed on the first encryption parameter. For example, the Paillier homomorphic encryption scheme can be used, which supports addition operations on encrypted data, allowing the aggregation node to sum the encrypted parameters uploaded by multiple clients without decryption.

[0078] The second encryption parameter is the local model parameter corresponding to the human dimension data, which is the parameter obtained by overlaying a homomorphic encryption mask on the first encryption parameter. The encryption parameter is the parameter that is finally uploaded to the privacy aggregation node, and its form varies depending on whether it corresponds to human dimension data.

[0079] Please see Figure 5 , Figure 5 This is a flowchart illustrating a traffic situation perception method provided in an embodiment of this application. Figure 5 As shown, the method in this application embodiment may include the following steps S501-S504: S501 performs rule matching between heterogeneous data across all dimensions and preset scene feature thresholds to obtain matching results.

[0080] Specifically, based on the scene classification model, heterogeneous data of all dimensions are matched with preset scene feature thresholds by rules. The preset scene feature thresholds refer to a series of feature parameters and their corresponding numerical ranges or logical conditions that are pre-set to identify specific traffic scenes. These thresholds can be determined based on historical data or simulation and are used to quickly determine whether the fused data conforms to the typical characteristics of a known scene.

[0081] For example, we can set "traffic volume greater than X and average vehicle speed less than Y" as the feature threshold for congestion scenarios. Rule matching is a method that compares and judges data based on predefined rules and thresholds. By comparing various features in the fused data with preset scenario feature thresholds, it determines whether the data meets the recognition conditions of a specific scenario.

[0082] The matching result is characterized by clear scene features, which means that the features in the fused data are highly consistent with the feature threshold of a certain preset scene, and there is no confusion with the features of other scenes. This means that the system can directly identify the current scene with a high degree of confidence.

[0083] S502, if the matching result shows clear scene features, then the matched scene will be determined as the target scene.

[0084] S503. If the matching result is that the scene features are fuzzy, then input the heterogeneous data of all dimensions into the scene classification model to obtain the classification probability of each type of scene.

[0085] Specifically, in S402-S403, when the rule matching result is clear, the matched specific scene is directly used as the current target scene. A matching result of ambiguous scene features means that the features in the fused data fail to completely match the feature thresholds of any preset scene, or partially match the feature thresholds of multiple scenes simultaneously, making it impossible to determine the current scene clearly through simple rules. This may occur in transitional states, abnormal situations, or when the data is incomplete.

[0086] When rule matching fails to explicitly identify a scene, heterogeneous data across all dimensions is input into a scene classification model to obtain the classification probabilities for each scene. This scene classification model is a machine learning model used to classify input data, assigning it to one of several predefined scene categories. Scene classification models can utilize learned complex patterns and feature relationships for deeper judgments; implementation methods can include support vector machines, neural networks, decision trees, random forests, etc. The classification probability for each scene represents the likelihood that the input data belongs to each predefined scene category after classification, typically expressed as probability values. These probability values ​​reflect the model's confidence in different scenes.

[0087] S504, determine the scene corresponding to the highest classification probability as the target scene category.

[0088] Specifically, in S503-S504, the classification probability of each scene is the likelihood that the input data belongs to each predefined scene category after the scene classification model classifies it. This probability is usually expressed as a probability value, reflecting the model's confidence in different scenes. Determining the scene with the highest classification probability as the target scene involves selecting the scene with the highest probability value from the output of the scene classification model's probabilities for each scene category. This is a common decision-making strategy used to make the best judgment under uncertainty.

[0089] For example, suppose we need to identify three traffic scenarios: "congestion," "slow traffic," and "smooth traffic." We can preset the feature thresholds as follows: "traffic volume greater than X and average speed less than Y" for congestion scenarios; "traffic volume between X and Z and average speed between Y and W" for slow traffic scenarios; and "traffic volume less than Z and average speed greater than W" for smooth traffic scenarios. When heterogeneous data from all dimensions is input, rule matching is performed first. If the data shows that traffic volume is significantly higher than X and speed is significantly lower than Y, the matching result indicates a clear scenario feature, and it is directly identified as a "congestion" scenario. If the data shows that traffic volume and speed are near the boundaries of X and Z, Y and W, or exhibit abnormal fluctuations, making it impossible to clearly match any scenario, the matching result is an ambiguous scenario feature. In this case, the fused data is input into a pre-trained neural network scenario classification model. This model will output classification probabilities such as "congestion: 0.4, slow traffic: 0.55, smooth traffic: 0.05." Since "slow traffic" has the highest classification probability, it is identified as the target scenario.

[0090] In summary, this application combines the speed of rule matching with the robustness of scene classification models. It enables efficient and direct identification of scenes with well-defined features, while allowing for refined judgment through deep learning models in scenes with ambiguous features, thus significantly improving the accuracy and adaptability of traffic scene recognition. This allows subsequent dynamic allocation of sensing resources and traffic situation awareness to be based on more reliable scene information, thereby enhancing the overall intelligence level of traffic management and control.

[0091] In practical applications, accurately and comprehensively assessing the priorities of different traffic scenarios to guide the dynamic allocation of subsequent sensing resources is a problem that needs to be solved. Inaccurate or insufficient priority assessment may lead to unreasonable allocation of sensing resources, hindering effective response to sudden or high-risk traffic events, thus affecting the accuracy and response efficiency of traffic situational awareness. Based on this, please refer to [link to relevant documentation]. Figure 6 , Figure 6 This is a flowchart illustrating a traffic situation perception method provided in an embodiment of this application. Figure 6As shown, the method in this application embodiment may include the following steps S601-S606: S601, obtain the risk coefficient, impact range coefficient, and response time coefficient of the target scenario category.

[0092] In this embodiment, the risk coefficient of a target scenario refers to an indicator that quantitatively assesses the potential hazard level or probability of occurrence of a specific traffic scenario. The risk coefficient can be determined based on historical accident data, real-time event reports, and traffic flow prediction models. For example, accident scenarios involving hazardous materials transport vehicles typically have a higher risk coefficient, while minor traffic congestion scenarios have a relatively lower risk coefficient. This risk coefficient can also be obtained by comprehensively judging various factors such as vehicle type, road conditions, and weather conditions involved in the scenario.

[0093] The impact range coefficient is an indicator that measures the spatial and temporal breadth of the impact of a specific traffic scenario on the surrounding road network and traffic flow. This coefficient can reflect the extent of traffic delays or congestion caused by the scenario, or the length of roads affected. For example, a prolonged accident occurring on a multi-lane main road will have a significantly higher impact range coefficient than a short-term malfunction occurring on a secondary road. The determination of this coefficient can rely on traffic simulation models, road network topology analysis, or real-time traffic data analysis.

[0094] S602, the priority assessment value is obtained by weighting the risk coefficient, the scope of impact coefficient and the response time coefficient.

[0095] Specifically, the priority evaluation formula is as follows:

[0096] in, For risk coefficient, For the influence range coefficient, For response time coefficient, The risk coefficient is weighted and the preset value is 0.5. The weight of the influence range coefficient is set to 0.3. The response timeliness coefficient is weighted and preset to 0.2. The total weight coefficient is 1 to ensure the rationality of the evaluation logic.

[0097] In emergency scenarios =1, in high-risk routine scenarios =0.7; In low-to-medium risk routine scenarios =0.4; In low-risk, routine scenarios =0.2. Influence range coefficient. Calculated based on the percentage of lanes affected by each scenario. Response timeliness coefficient. The value is 1 for scenarios requiring immediate response, 0.7 for scenarios requiring key monitoring, and 0.3 for routine monitoring scenarios.

[0098] S603, determine the scene priority of the target scene category based on the priority evaluation value.

[0099] In this embodiment, scene priority is the result of classifying priority evaluation values, typically including discrete levels such as high priority, medium priority, and low priority. These levels provide traffic management departments with clear decision-making basis, guiding the allocation and scheduling of sensing resources. For example, priority evaluation values ​​can be divided into different intervals, with each interval corresponding to a priority level.

[0100] S604 If the scene priority is high priority, the sampling frequency of the core sensor will be increased to the first sampling frequency, and the multi-sensor collaborative sensing mode will be enabled and redundant devices will be turned on.

[0101] In this embodiment, the multi-sensor collaborative sensing mode activates all available sensors for data acquisition when the scene priority is high, and processes the data using a multi-source data fusion algorithm to obtain the most comprehensive and accurate situational information. The redundant device on / off status refers to the enabled or disabled state of standby or auxiliary devices. Conversely, when sensing demand is low, unnecessary redundant devices can be disabled to save energy.

[0102] S605 If the scene priority is medium priority, the sampling frequency of the core sensor will be maintained at the second sampling frequency, and the key sensor collaborative sensing mode will be enabled and unnecessary sensors will be turned off.

[0103] In the embodiments of this application, the key sensor collaborative sensing mode refers to enabling only some of the most critical sensors in the current scene to work collaboratively when the scene priority is medium priority.

[0104] S606 If the scene priority is low, the sampling frequency of the core sensor will be reduced to the third sampling frequency, and the independent sensing mode of the single core sensor will be enabled.

[0105] In this embodiment, the single-core sensor independent sensing mode refers to using only one core sensor for data acquisition and processing, while other auxiliary sensors are in standby or off state. The core sensor refers to the sensor that undertakes the main sensing task and provides key data sources in the traffic situation perception system; it is usually the most important and highest-performing sensor type in the system. The core sensor can be a high-precision millimeter-wave radar, lidar, high-definition camera, or an intelligent roadside unit with edge computing capabilities, etc.

[0106] It should be noted that in S604-S606, the first sampling frequency, second sampling frequency, and third sampling frequency can be preset according to the actual sensor performance, data processing capabilities, and the sensing accuracy requirements and energy consumption budget under different priority scenarios. For example, the first sampling frequency can be set to the sensor's maximum operating frequency to ensure the highest sensing accuracy; the second sampling frequency can be set to a balanced frequency that meets normal sensing needs; and the third sampling frequency can be set to the minimum frequency that meets basic sensing needs to maximize energy savings.

[0107] For example, if a high-priority scenario of "multi-vehicle rear-end collision ahead" is detected, the control module will immediately send instructions to the core millimeter-wave radar and high-definition camera of the roadside unit, increasing the radar's sampling frequency from the conventional 20Hz (second sampling frequency) to 50Hz (first sampling frequency) and the camera's frame rate from 30fps to 60fps. Simultaneously, the system will activate all available radars, cameras, and lidars on that road segment, putting them into a multi-sensor collaborative perception mode. The edge computing unit will then perform real-time fusion processing of multi-source data to accurately identify the location of the accident vehicles, the extent of damage, and the distribution of personnel. Furthermore, the system will also activate backup communication links and additional edge computing modules (redundant equipment).

[0108] If the system identifies a low-priority scenario of "stable traffic flow with no abnormal events," the control module will reduce the sampling frequency of the core millimeter-wave radar from 20Hz (second sampling frequency) to 10Hz (third sampling frequency) and reduce the camera frame rate to 15fps. In this case, only one core millimeter-wave radar is used for independent sensing, while other auxiliary sensors (such as some cameras and LiDAR) are in standby or off to minimize energy consumption. If the system identifies a medium-priority scenario of "increased traffic flow in localized road sections with a slight congestion trend," the control module will maintain the sampling frequency of the core millimeter-wave radar at 20Hz (second sampling frequency) and the camera frame rate at 30fps. The system will enable a key sensor collaborative sensing mode using the core millimeter-wave radar and some high-definition cameras. For example, only cameras covering congested areas will be used for collaborative sensing, while sensors in other non-essential areas will be turned off to optimize resource consumption while ensuring sensing accuracy.

[0109] In summary, this application enables refined and dynamic adjustment of sensing resources based on the priority of different traffic scenarios. In high-priority scenarios, the system can rapidly enhance its sensing capabilities, ensuring the acquisition of the most comprehensive and accurate situational information, thereby providing strong support for emergency response and decision-making. In medium-priority scenarios, the system can rationally allocate resources while ensuring sensing performance, avoiding unnecessary waste. In low-priority scenarios, the system can maximize energy savings and extend equipment lifespan. This intelligent resource allocation strategy significantly improves the efficiency, accuracy, and reliability of traffic situational awareness, while optimizing system operating costs, making the entire sensing system more adaptable and economical.

[0110] In one feasible implementation, when dynamically adjusting sensing resources based on scene priority, the following is specifically executed: Construct a resource optimization objective function, which aims to minimize total energy consumption and is constrained by ensuring that the perception accuracy meets the minimum perception accuracy requirement corresponding to the scene priority.

[0111] In this embodiment, the resource optimization objective function is a mathematical model used to quantify and evaluate the merits of system resource allocation schemes under specific constraints. Its construction typically involves identifying the variables to be optimized (e.g., sampling frequency), quantifying the objective (e.g., energy consumption), and defining constraints (e.g., sensing accuracy). The establishment of this function is fundamental to achieving intelligent and efficient resource management, transforming complex system performance indicators into computable mathematical expressions.

[0112] The resource optimization objective function aims to minimize total energy consumption. In a traffic situational awareness system, this means adjusting the configuration of sensing resources to minimize the total electrical energy consumed by the system when completing sensing tasks. This typically includes the operating energy consumption of the sensors themselves, data transmission energy consumption, data processing energy consumption, and the standby or operating energy consumption of redundant equipment.

[0113] Based on the resource optimization objective function, the first sampling frequency, the second sampling frequency, or the third sampling frequency is optimized and adjusted.

[0114] Specifically, after identifying the scene priority, the first, second, or third sampling frequency is used as a variable to be optimized. Based on real-time parameters of the current traffic scene (e.g., traffic flow, event density) and the sensor's own performance and power consumption models, the total energy consumption is expressed as a function of these sampling frequencies. Simultaneously, perception accuracy is also modeled as a function of the sampling frequency, and a corresponding minimum perception accuracy threshold is set based on the scene priority. Subsequently, the system invokes an optimization algorithm, using the constructed resource optimization objective function as a basis, to iteratively search and determine the specific values ​​of the first, second, or third sampling frequency that minimize the total energy consumption, while satisfying the perception accuracy constraint.

[0115] In summary, by constructing a resource optimization objective function with the goal of minimizing total energy consumption and constrained by meeting the minimum perception accuracy requirements corresponding to the scenario priority, these sampling frequencies are finely optimized and adjusted. This enables the perception system to intelligently achieve the optimal balance between energy consumption and perception performance when facing different traffic scenarios, thereby significantly improving the energy efficiency and sustainable operation capability of the entire traffic situation perception system. This optimization and adjustment mechanism makes the allocation of perception resources more flexible and efficient.

[0116] based on Figure 1 The system architecture will be discussed below. Figure 7 This application provides a detailed description of the traffic situation perception device provided in its embodiments. It should be noted that... Figure 7 The traffic situation awareness device in the application is used to perform the functions described herein. Figure 2 - Figure 6 The methods shown in the embodiments are illustrated for ease of explanation, showing only the parts relevant to the embodiments of this application. For specific technical details not disclosed, please refer to this application. Figure 2 - Figure 6 In the embodiment shown, the traffic situation awareness device 700 may include a data acquisition unit 701, a priority identification unit 702, a resource allocation unit 703, and a result feedback unit 704, as detailed below: The data acquisition unit 701 is used to acquire heterogeneous data across all dimensions, including data on people, vehicles, roads, and the cloud. The priority identification unit 702 is used to obtain the target scene category corresponding to the heterogeneous data in all dimensions and determine the scene priority corresponding to the target scene category. The resource configuration unit 703 is used to dynamically configure the sensing resources according to the scene priority to obtain the configured sensing resources. The sensing resources are used to characterize the sensor sampling frequency, sensor cooperation mode and redundant device switching status. The result feedback unit 704 is used to perform traffic situation perception based on the configuration of perception resources and obtain traffic situation perception results.

[0117] In one embodiment, the data acquisition unit 701 is further configured to perform the following specific actions: Acquire traffic flow data from the entire network and extract road segment feature data from the traffic flow data; Based on the road segment characteristic data, the congestion coefficient of a single road segment is calculated, and the congestion coefficient of the single road segment is weighted to obtain the road network congestion coefficient. Based on the road network congestion coefficient and road segment travel time, dynamic programming calculations are performed to obtain the optimal travel route cost; The road network congestion coefficient and the cost of the optimal driving route are defined as cloud-based data.

[0118] In one embodiment, the data acquisition unit 701 is further configured to perform the following specific actions: Based on the reference time, the local timestamps in the heterogeneous data of all dimensions are time-synchronized and calibrated to obtain the calibrated timestamps; Transform the spatial coordinates in the heterogeneous data of all dimensions to the target coordinate system to obtain the calibrated spatial coordinates; By binding human dimension data with calibration spatial coordinates, an integrated human-vehicle spatial data block is obtained, which in turn associates human dimension data with road segment-level spatial units.

[0119] In one embodiment, the priority identification unit 702 is further configured to perform the following specific actions: The heterogeneous data across all dimensions is matched against preset scene feature thresholds using rules to obtain the matching results; If the matching result has clear scene features, then the matched scene will be identified as the target scene; If the matching result is that the scene features are fuzzy, then the heterogeneous data of all dimensions will be input into the scene classification model to obtain the classification probability of each type of scene. The scene corresponding to the highest classification probability is determined as the target scene category.

[0120] In one embodiment, the priority identification unit 702 is further configured to perform the following specific actions: Obtain the risk coefficient, impact range coefficient, and response timeliness coefficient for the target scenario category; The priority assessment value is obtained by weighting the risk coefficient, the scope of impact coefficient, and the response time coefficient. Based on the priority evaluation value, the scene priority of the target scene category is determined.

[0121] In one embodiment, the resource allocation unit 703 is further configured to perform the following specific actions: If the scene priority is high priority, the sampling frequency of the core sensor will be increased to the first sampling frequency, and the multi-sensor collaborative sensing mode will be enabled and redundant devices will be turned on. If the scene priority is medium priority, the sampling frequency of the core sensor will be maintained at the second sampling frequency, and the key sensor collaborative sensing mode will be enabled and unnecessary sensors will be turned off. If the scene priority is low, the sampling frequency of the core sensor will be reduced to the third sampling frequency, and the independent sensing mode of a single core sensor will be enabled. In this mode, the first sampling frequency is greater than the second sampling frequency, and the second sampling frequency is greater than the third sampling frequency.

[0122] In one embodiment, the resource allocation unit 703 is further configured to perform the following specific actions: Construct a resource optimization objective function, which aims to minimize total energy consumption and is constrained by ensuring that the perception accuracy meets the minimum perception accuracy requirement corresponding to the scene priority. Based on the resource optimization objective function, the first sampling frequency, the second sampling frequency, or the third sampling frequency is optimized and adjusted.

[0123] In one embodiment, the data acquisition unit 701 is further configured to perform the following specific actions: The traffic situation awareness model parameters are initialized through the privacy aggregation node and then sent to the client. Train the local model on the client to obtain the local model parameters; The client encrypts the local model parameters to obtain encrypted parameters, and then uploads the encrypted parameters to the privacy aggregation node. The encrypted parameters are aggregated based on privacy aggregation nodes to obtain updated global model parameters. The updated global model parameters are then used for iterative optimization until the global model loss function converges, resulting in the optimal traffic situation perception model, which outputs heterogeneous data across all dimensions.

[0124] The traffic situation perception device 700 provided in the above embodiments can realize the technical solutions described in the above traffic situation perception method embodiments. The specific implementation principles of each module or unit can be found in the corresponding content in the above traffic situation perception method embodiments, and will not be repeated here.

[0125] like Figure 8 As shown, the present invention also provides an electronic device 800. The electronic device 800 includes a processor 801, a memory 802, and a display 803. Figure 8 Only some components of the electronic device 800 are shown, but it should be understood that it is not required to implement all the components shown, and more or fewer components may be implemented instead.

[0126] In some embodiments, processor 801 may be a central processing unit (CPU), microprocessor, or other data processing chip, used to run program code stored in memory 802 or process data, such as the traffic situation perception method of the present invention.

[0127] In some embodiments, processor 801 may be a single server or a group of servers. The server group may be centralized or distributed. In some embodiments, processor 801 may be local or remote. In some embodiments, processor 801 may be implemented on a cloud platform. In one embodiment, the cloud platform may include a private cloud, public cloud, hybrid cloud, community cloud, distributed cloud, intranet, multi-cloud, etc., or any combination thereof.

[0128] In some embodiments, memory 802 may be an internal storage unit of electronic device 800, such as a hard disk or memory of electronic device 800. In other embodiments, memory 802 may also be an external storage device of electronic device 800, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on electronic device 800.

[0129] Furthermore, the memory 802 may include both internal storage units of the electronic device 800 and external storage devices. The memory 802 is used to store application software and various types of data installed on the electronic device 800.

[0130] In some embodiments, display 803 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. Display 803 is used to display information from electronic device 800 and to display a visual user interface. Components 801-803 of electronic device 800 communicate with each other via a system bus.

[0131] In one embodiment, when processor 801 executes the traffic situation awareness program in memory 802, the following steps can be implemented: Acquire heterogeneous data across all dimensions, including data on people, vehicles, roads, and the cloud. Obtain the target scene categories corresponding to the heterogeneous data across all dimensions, and determine the scene priorities corresponding to the target scene categories; The sensing resources are dynamically configured according to the scenario priority to obtain the configured sensing resources. The sensing resources are used to characterize the sensor sampling frequency, sensor cooperation mode and the on / off status of redundant devices. Traffic situation perception is performed based on the configuration of perception resources, and the traffic situation perception results are obtained.

[0132] It should be understood that when the processor 801 executes the traffic situation awareness program in the memory 802, in addition to the functions mentioned above, it can also perform other functions, as detailed in the description of the corresponding method embodiments above.

[0133] Furthermore, this embodiment of the invention does not specifically limit the type of electronic device 800 mentioned. Electronic device 800 can be a mobile phone, tablet computer, personal digital assistant (PDA), wearable device, laptop computer, or other portable electronic device. Exemplary embodiments of portable electronic devices include, but are not limited to, portable electronic devices running iOS, Android, Microsoft, or other operating systems. The aforementioned portable electronic device can also be other portable electronic devices, such as a laptop computer with a touch-sensitive surface (e.g., a touch panel). It should also be understood that in some other embodiments of the invention, electronic device 800 may not be a portable electronic device, but rather a desktop computer with a touch-sensitive surface (e.g., a touch panel).

[0134] Accordingly, this application also provides a computer-readable storage medium for storing computer-readable programs or instructions. When the programs or instructions are executed by a processor, they can implement the steps or functions of the traffic situation perception methods provided in the above-described method embodiments.

[0135] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.), and the computer program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0136] The traffic situation perception method, device, electronic device, and storage medium provided by the present invention have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A traffic situation perception method, characterized in that, include: Acquire heterogeneous data across all dimensions, including data on people, vehicles, roads, and the cloud. Obtain the target scene category corresponding to the full-dimensional heterogeneous data, and determine the scene priority corresponding to the target scene category; The sensing resources are dynamically configured according to the scenario priority to obtain the configured sensing resources. The sensing resources are used to characterize the sensor sampling frequency, sensor cooperation mode and redundant device on / off status. Traffic situation perception is performed based on the configured perception resources to obtain traffic situation perception results.

2. The method according to claim 1, characterized in that, The human-dimensional data represents the driver's physiological state data, the road-dimensional data represents roadside perception data and environmental perception data, and the vehicle-dimensional data represents the vehicle's driving state. The acquisition of multi-dimensional heterogeneous data includes: Acquire traffic flow data from the entire network and extract road segment feature data from the traffic flow data of the entire network; Based on the road segment characteristic data, calculate the congestion coefficient of a single road segment, and then perform a weighted calculation on the congestion coefficient of the single road segment to obtain the road network congestion coefficient; Based on the road network congestion coefficient and road segment travel time, dynamic programming calculations are performed to obtain the optimal travel route cost; The road network congestion coefficient and the optimal driving route cost are determined as cloud-dimensional data.

3. The method according to claim 1, characterized in that, After acquiring the full-dimensional heterogeneous data, the method further includes: Based on the reference time, the local timestamps in the full-dimensional heterogeneous data are time-synchronized and calibrated to obtain the calibrated timestamps. The spatial coordinates in the full-dimensional heterogeneous data are transformed to the target coordinate system to obtain the calibrated spatial coordinates; The human dimension data is bound to the calibration spatial coordinates to obtain a human-vehicle spatial integrated data block, thereby associating the human dimension data with the road segment-level spatial unit.

4. The method according to claim 1, characterized in that, The process of determining the scene priority of the target scene category includes: Obtain the risk coefficient, impact range coefficient, and response timeliness coefficient of the target scenario category; The priority evaluation value is obtained by weighting the risk coefficient, the scope of influence coefficient and the response time coefficient. Based on the priority evaluation value, the scene priority of the target scene category is determined.

5. The method according to claim 4, characterized in that, The dynamic configuration of sensing resources according to the scene priority includes: If the scenario priority is high priority, the sampling frequency of the core sensor will be increased to the first sampling frequency, and the multi-sensor collaborative sensing mode will be enabled and redundant devices will be activated. If the scenario priority is medium priority, the sampling frequency of the core sensor will be maintained at the second sampling frequency, and the key sensor collaborative sensing mode will be enabled and unnecessary sensors will be turned off. If the scenario priority is low, the sampling frequency of the core sensor is reduced to the third sampling frequency, and the independent sensing mode of the single core sensor is enabled, wherein the first sampling frequency is greater than the second sampling frequency, and the second sampling frequency is greater than the third sampling frequency.

6. The method according to claim 5, characterized in that, The step of dynamically adjusting the perception resources according to the scene priority includes: Construct a resource optimization objective function, which aims to minimize total energy consumption and is constrained by ensuring that the perception accuracy meets the minimum perception accuracy requirement corresponding to the scenario priority. Based on the resource optimization objective function, the first sampling frequency, the second sampling frequency, or the third sampling frequency is optimized and adjusted.

7. The method according to claim 1, characterized in that, The acquisition of heterogeneous data across all dimensions includes: The traffic situation awareness model parameters are initialized through the privacy aggregation node, and the traffic situation awareness model parameters are then sent to the client. Train the local model on the client to obtain the local model parameters; The client encrypts the local model parameters to obtain encrypted parameters, and then uploads the encrypted parameters to the privacy aggregation node. The encrypted parameters are aggregated based on the privacy aggregation node to obtain updated global model parameters. Iterative optimization is then performed based on the updated global model parameters until the global model loss function converges, resulting in the optimal traffic situation perception model, which outputs heterogeneous data across all dimensions.

8. A traffic situation awareness device, characterized in that, The device includes: The data acquisition unit is used to acquire heterogeneous data across all dimensions, including human-dimensional data, vehicle-dimensional data, road-dimensional data, and cloud-dimensional data. A priority identification unit is used to obtain the target scene category corresponding to the full-dimensional heterogeneous data and determine the scene priority corresponding to the target scene category. The resource configuration unit is used to dynamically configure the sensing resources according to the scenario priority to obtain the configured sensing resources. The sensing resources are used to characterize the sensor sampling frequency, sensor cooperation mode and redundant device switching status. The result feedback unit is used to perform traffic situation perception based on the configured perception resources and obtain traffic situation perception results.

9. An electronic device, characterized in that, The electronic device includes: Memory, used to store executable program code; A processor for calling and running the executable program code from the memory, causing the electronic device to perform the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed, implements the method as described in any one of claims 1 to 7.

Images