A real-time data deep processing method and system for traffic state evaluation

Abstract

The application discloses a real-time data deep processing method and system for traffic state evaluation, and relates to the field of Internet of Things.The method comprises the following steps: mapping the obtained multi-source traffic data to a unified space-time reference; calculating the importance score based on the data characteristic attribute to execute a differentiated data transmission strategy, and outputting edge pre-processing data; extracting continuous numerical indicators in the data, combining dynamic credibility to execute feature layer filtering estimation; synchronously extracting heterogeneous independent indicator data, constructing a correction factor based on a conflict coefficient, and substituting the correction factor into an evidence synthesis rule to execute decision layer fusion calculation; comprehensively obtaining preliminary evaluation values based on the feature layer and the decision layer results; introducing a scene correction factor calculated based on real-time environment and event characteristics, dynamically calibrating the preliminary evaluation values, and outputting final traffic state evaluation results; and generating device control instructions and traffic guidance information according to the final traffic state evaluation results. The technical scheme provided by the application improves the accuracy of traffic state evaluation in complex special scenarios.

Description

Technical Field

[0001] This application relates to the field of the Internet of Things, and in particular to a real-time data deep processing method and system for traffic condition assessment. Background Technology

[0002] Currently, intelligent traffic management systems have become a key infrastructure for improving urban traffic efficiency and ensuring road safety. With the rapid development of the Internet of Things and intelligent sensing technologies, traffic management departments have deployed a massive number of cameras, millimeter-wave radars, and other detection devices to collect real-time data on road network operations.

[0003] In related technologies, multi-sensor centralized fusion networks or fixed-weight algorithms are typically used for traffic condition assessment. The entire raw video stream and radar point cloud data collected by various front-end sensors are centrally transmitted back to a central server. Subsequently, in the cloud, a pre-trained deep learning model or a static Kalman filter algorithm is used to perform unified static weight allocation and joint calculation on the feature data uploaded by each sensor node, thereby deriving traffic operation indicators such as vehicle speed and traffic flow for the current road segment, and outputting routine road condition assessment results.

[0004] However, during peak hours, the concurrent transmission of massive amounts of data can easily lead to network bandwidth congestion and system processing delays. In environments such as sudden rainstorms, fog, or traffic accidents, sensors based on different physical principles are easily interfered with and output contradictory underlying detection data. It is difficult to identify and separate these conflicting underlying evidence, which in turn leads to a decrease in the accuracy of traffic condition assessment in complex and special scenarios and can easily cause misjudgments. Summary of the Invention

[0005] This application provides a real-time data deep processing method and system for traffic condition assessment, which can improve the accuracy of traffic condition assessment in complex and special scenarios.

[0006] The first aspect of this application provides a real-time data deep processing method for traffic state assessment, the method comprising: The system receives edge preprocessed data; wherein the edge preprocessed data is obtained by edge nodes acquiring multi-source traffic data and mapping it to a unified spatiotemporal benchmark, then calculating data importance scores based on feature attributes, and sending it after implementing differentiated data transmission strategies at different levels; continuous numerical index data is extracted from the edge preprocessed data, and feature-level filtering state estimation is performed in combination with the dynamic credibility corresponding to each data source, outputting a feature-layer fusion state estimate; heterogeneous independent index data representing traffic state is extracted from the edge preprocessed data, a conflict correction factor is constructed based on the conflict coefficient between each independent index data, and an evidence synthesis rule with the conflict correction factor is used for calculation, outputting a decision-level fusion result; a scene correction factor for the current traffic scenario is calculated by combining real-time acquired environmental feature data and event feature data; a preliminary evaluation value is obtained based on the feature-layer fusion state estimate and the decision-level fusion result, and the preliminary evaluation value is dynamically corrected using the scene correction factor, outputting a final traffic state evaluation result; corresponding equipment control commands and traffic guidance information are generated and output based on the final traffic state evaluation result.

[0007] In the above embodiments, by performing unified spatiotemporal mapping and edge-side hierarchical filtering on multi-source traffic data, the ineffective loss of communication bandwidth is reduced. Combining feature-layer adaptive filtering and noise reduction with decision-layer conflict correction evidence synthesis eliminates the inherent data jumps of a single sensor and decision discrepancies between multi-source heterogeneous sensors. Finally, the introduction of multi-dimensional environmental features to dynamically correct the evaluation values ​​improves the accuracy and robustness of road network state assessment in complex traffic scenarios.

[0008] In conjunction with some embodiments of the first aspect, in some embodiments, the calculation of data importance scores based on the characteristic attributes of the multi-source traffic data under a spatiotemporal reference, and the execution of differentiated data transmission strategies according to the data importance scores to output edge preprocessing data, specifically includes: Extract the road segment risk level, real-time traffic status, and time features corresponding to the multi-source traffic data; assign corresponding weight coefficients to the road segment risk level, the real-time traffic status, and the time features, and perform weighted summation to obtain the data importance score; compare the data importance score with a preset score interval, and execute a full transmission strategy or a dimensionality reduction sampling ratio transmission strategy according to the interval level determined by the comparison result, and output the edge preprocessed data.

[0009] In the above embodiments, the value of data is dynamically quantified and scored by comprehensively considering road segment risks, real-time status, and time characteristics. This allows edge nodes to adaptively switch between full-data or dimensionality-reduced sampling transmission strategies based on the importance of the data. This mechanism ensures the complete and transparent transmission of core traffic status data during high-risk congestion periods while reducing the amount of invalid transmission of redundant data during normal off-peak periods, thus alleviating the communication channel congestion pressure under the edge-cloud collaborative architecture.

[0010] In conjunction with some embodiments of the first aspect, in some embodiments, the step of extracting continuous numerical index data from the edge preprocessing data, combining the dynamic credibility corresponding to each data source to perform feature-level filtering state estimation, and outputting feature-level fusion state estimation values ​​specifically includes: Extract continuous numerical index data from the edge preprocessing data, and calculate the dynamic credibility of each data source based on the historical detection deviation value of each data source and the preset environmental impact coefficient; use the dynamic credibility as a weighting factor to replace the original observation noise covariance matrix in the Kalman filter algorithm and perform filter gain iteration to output the estimated value of the feature layer fusion state.

[0011] In the above embodiments, by deeply integrating historical deviation statistics with external environmental disturbances, the reliability of each data source in the current spatiotemporal context is dynamically quantified and evaluated. This reliability is used as an adaptive weighting factor to directly intervene in the iteration process of the filter gain matrix, enabling the state estimation algorithm to keenly perceive and suppress data distortion caused by environmental degradation or equipment aging, thereby improving the numerical stability of continuous numerical parameters in the feature layer fusion stage.

[0012] In conjunction with some embodiments of the first aspect, in some embodiments, the step of extracting heterogeneous independent indicator data representing traffic state from the edge preprocessing data, constructing a conflict correction factor based on the conflict coefficient between the independent indicator data, and calculating the result using evidence synthesis rules with the conflict correction factor to output the decision-level fusion result specifically includes: Heterogeneous independent index data representing traffic states are extracted from the edge preprocessing data, and basic probability assignment values ​​for each traffic state category are calculated based on the heterogeneous independent index data; the basic probability assignment values ​​are cross-accumulated to extract the conflict coefficient; a conflict correction factor that is inversely proportional to the conflict coefficient is constructed, and the basic probability assignment values ​​are weighted and adjusted using the conflict correction factor; the weighted and adjusted basic probability assignment values ​​are probabilistically synthesized using evidence synthesis rules, and the category corresponding to the highest synthesis probability is selected as the decision layer fusion result.

[0013] In the above embodiments, a correction factor for the inverse proportional mapping is dynamically constructed by quantifying the degree of cross-conflict among heterogeneous independent indicators in the probability allocation space. The decision weights of highly conflicting evidence vectors are adaptively weakened before evidence synthesis, enhancing the decision-making layer's fault-tolerant judgment capability when dealing with contradictions in multi-source data.

[0014] In conjunction with some embodiments of the first aspect, in some embodiments, after extracting continuous numerical index data from the edge preprocessing data, combining the dynamic credibility corresponding to each data source to perform feature-level filtering state estimation, and outputting the feature-level fusion state estimate, the method further includes: When concurrent high-frequency transient jumps are detected in the estimated value of the feature layer fusion state within a preset time window, and the environmental feature data characterizes an abnormal scene of precipitation or road surface attachment superimposed with low visibility, it is determined that the multi-source sensor concurrent resonance interference mode is entered. Simultaneously, the micro-motion frequency domain distribution features in the echo signal of the active detection device within this time sequence, as well as the local morphological dynamic disturbance features of the corresponding video feature data in the target area, are extracted. The micro-motion frequency domain distribution features and the local morphological dynamic disturbance features are then subjected to cross-modal spatiotemporal correlation analysis to construct a representation of the real... The microphysical coupling matrix is ​​used to determine the physical property of target motion causing splashing or disturbance of the road surface medium. Target data that causes high-frequency transient jumps is input into the microphysical coupling matrix for cross-comparison and identification. Target data that does not have a specific spatiotemporal coupling relationship is identified as pseudo-target heterogeneous data induced by physical environment distortion and multipath reflection clutter and is removed. For the remaining real target data after removing pseudo-target heterogeneous data, a filtering algorithm with robust adjustment and adaptive attenuation characteristics is introduced to iterate the state gain again and output the reconstructed feature layer fusion state estimate.

[0015] In the above embodiments, by introducing cross-modal joint spatiotemporal verification of active echo frequency domain and passive video morphology under meteorological conditions, a coupling matrix reflecting the physical disturbance of the road surface medium is constructed. This matrix can identify and remove false target clutter excited by multipath effects. Combined with secondary reconstruction using a robust algorithm, the link for environmental noise to propagate to the higher-level state machine is severed, ensuring the resilience of the traffic state perception system against environmental interference in harsh scenarios.

[0016] In conjunction with some embodiments of the first aspect, in some embodiments, after the step of introducing a filtering algorithm with robust adjustment and adaptive attenuation characteristics to iterate the state gain again for the remaining real target data after stripping and removing heterogeneous data of false targets, and outputting the reconstructed feature layer fusion state estimate, the method further includes: When the reconstructed feature layer fusion state estimate is detected to show a sudden disappearance of the target group in a continuous spatial segment, and the state estimate in the preceding time series shows a continuous deceleration trend, it is determined that the current road segment has entered an abnormal scene superimposed with a secondary static scene of congestion and stagnation. The static occlusion features of the environmental background in the video feature data of the corresponding spatial segment in the current time series are extracted, and the detection signal occlusion shadow features formed by the echo signal of the active detection device behind the target are extracted simultaneously. The static occlusion features and the detection signal occlusion shadow features are spatially topologically mapped to construct a static spatial occupancy matrix representing the physical stagnation volume of the real target. The spatial coordinates of the sudden disappearance of the target group are backtracked into the static spatial occupancy matrix for secondary verification, and the spatial coordinates falling within the effective area of ​​the static spatial occupancy matrix are extracted. The extracted spatial coordinates are marked as the real stagnation target data that was mistakenly removed, and a static state vector approaching zero is assigned to them. The real stagnation target data is forcibly incorporated into the filtering algorithm with robust adjustment and adaptive attenuation characteristics for state update, and the corrected feature layer fusion state estimate is output.

[0017] In the above embodiments, to address the issue of targets being mistakenly filtered out as background by radar sensors due to prolonged stagnation caused by severe congestion, a spatial occupancy matrix is ​​constructed by fusing static video occlusion and radar back shadows. This restores the data of mistakenly removed stagnant targets and resets the stationary vector, compensating for the low-speed perception blind spot of the microwave detection system and ensuring the physical consistency of traffic flow reconstruction in congestion-related secondary scenarios.

[0018] In conjunction with some embodiments of the first aspect, in some embodiments, the step of extracting heterogeneous independent indicator data representing traffic state from the edge preprocessing data, constructing a conflict correction factor based on the conflict coefficient between the independent indicator data, and calculating the result using evidence synthesis rules with the conflict correction factor to output the decision-level fusion result specifically includes: Heterogeneous independent index data representing traffic conditions are extracted from the edge preprocessing data. Basic probability assignment values ​​for each traffic condition category are calculated, and the conflict coefficient between the heterogeneous independent index data is extracted. When the conflict coefficient exceeds a preset extreme conflict threshold, and the corresponding heterogeneous independent index data exhibits a significantly different step distribution of motion states in the longitudinal topology of the road space, a traffic shock wave abnormal conflict mode is triggered, and the conventional evidence synthesis rule is suspended. The dynamic state gradient difference between the heterogeneous independent index data is calculated along the longitudinal topology of the road space, and the spatial node coordinates where the dynamic state gradient difference exceeds the critical safety threshold are identified as state-abrupt shock wave fronts. For the state-abrupt shock wave fronts and their associated sections in the direction of oncoming traffic, probability synthesis based on the conflict correction factor is skipped, and their decision category is anchored to the highest warning level category, and output as the decision-layer fusion result.

[0019] In the above embodiments, by capturing in real time the step gradients of contradictory motion states on the longitudinal right-of-way topology, traffic shock fronts that may induce chain-reaction collisions are identified. When high-risk shock wave patterns are detected, the smooth convergence logic of conventional evidence synthesis is interrupted, improving the risk response rating of the upstream impact area of ​​the shock wave. This eliminates the time lag effect of conventional statistical algorithms in responding to sudden and severe events, providing a significant advance warning for traffic guidance and interception.

[0020] In a second aspect, embodiments of this application provide a real-time data deep processing system for traffic condition assessment. The real-time data deep processing system for traffic condition assessment includes: one or more processors and a memory; the memory is coupled to the one or more processors and is used to store computer program code, the computer program code including computer instructions, and the one or more processors call the computer instructions to cause the real-time data deep processing system for traffic condition assessment to perform the method described in the first aspect and any possible implementation thereof.

[0021] Thirdly, embodiments of this application provide a computer program product containing instructions that, when the computer program product is run on a real-time data deep processing system for traffic condition assessment, cause the real-time data deep processing system for traffic condition assessment to perform the method described in the first aspect and any possible implementation thereof.

[0022] Fourthly, embodiments of this application provide a computer-readable storage medium including instructions that, when executed on a real-time data deep processing system for traffic state assessment, cause the real-time data deep processing system for traffic state assessment to perform the method described in the first aspect and any possible implementation thereof.

[0023] Understandably, the real-time data deep processing system for traffic state assessment provided in the second aspect, the computer program product provided in the third aspect, and the computer storage medium provided in the fourth aspect are all used to execute the real-time data deep processing method for traffic state assessment provided in the embodiments of this application. Therefore, the beneficial effects they can achieve can be referred to the beneficial effects in the corresponding methods, and will not be repeated here.

[0024] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: 1. This application reduces the ineffective loss of communication bandwidth by performing unified spatiotemporal mapping and edge-side hierarchical filtering on multi-source traffic data. By combining feature-level adaptive filtering and noise reduction with decision-level conflict correction evidence synthesis, it eliminates the inherent data jumps of a single sensor and decision discrepancies between multi-source heterogeneous sensors. Finally, it introduces multi-dimensional environmental features to dynamically correct the evaluation values, improving the accuracy and robustness of road network state assessment in complex traffic scenarios.

[0025] 2. This application dynamically quantifies and scores the value of data by comprehensively considering road segment risks, real-time status, and time characteristics, enabling edge nodes to adaptively switch between full-data or dimensionality-reduced sampling transmission strategies based on the importance of the data. This mechanism ensures the complete transmission of core traffic status data during high-risk congestion periods while reducing the invalid transmission of redundant data during normal off-peak periods, thus alleviating the communication channel congestion pressure under the edge-cloud collaborative architecture.

[0026] 3. This application dynamically and quantitatively evaluates the credibility of each data source in the current spatiotemporal context by deeply integrating historical deviation statistics with external environmental disturbances. This credibility is then used as an adaptive weighting factor to directly intervene in the iteration process of the filter gain matrix, enabling the state estimation algorithm to keenly perceive and suppress data distortion caused by environmental degradation or equipment aging, thereby improving the numerical stability of continuous numerical parameters in the feature layer fusion stage. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the overall architecture of the real-time data deep processing system for traffic condition assessment provided in the embodiments of this application; Figure 2 This is a flowchart illustrating a real-time data deep processing method for traffic condition assessment in an embodiment of this application. Figure 3 This is a logical diagram illustrating the dual-track fusion of multi-source data feature layer and decision layer provided in the embodiments of this application; Figure 4 This is a schematic diagram of the process for dynamic traffic state calibration based on scene correction factors provided in the embodiments of this application; Figure 5 This is another flowchart illustrating the real-time data deep processing method for traffic condition assessment in this application embodiment; Figure 6 This is an exemplary hardware structure diagram of a real-time data deep processing system for traffic condition assessment in this application embodiment. Detailed Implementation

[0028] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification and appended claims of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to and includes any or all possible combinations of one or more of the listed items.

[0029] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0030] Figure 1 This is a schematic diagram of the overall architecture of the real-time data deep processing system for traffic condition assessment provided in an embodiment of this application. Figure 1 As shown, the system's logical architecture is divided from bottom to top into a data acquisition layer, an edge preprocessing layer, a cloud-based deep processing layer, an evaluation and decision-making layer, and an output execution layer. Specifically, the data acquisition layer is used to access multi-source raw traffic data and buffer and aggregate it through a cache queue; the edge preprocessing layer is deployed at roadside nodes close to the data source to perform data cleaning, format standardization, spatiotemporal alignment, and compression and filtering operations to generate edge preprocessed data; the cloud-based deep processing layer receives the above edge preprocessed data and performs parallel processing by the feature layer and the decision-making layer; the evaluation and decision-making layer combines multi-dimensional features to identify environmental scenarios and uses scenario correction factors to dynamically calibrate the basic evaluation values; the output execution layer distributes the final traffic state evaluation results to various terminals and control devices, and uses a closed-loop feedback mechanism to feed back the execution data to achieve online iteration and updates of the model.

[0031] Figure 2 This is a flowchart illustrating a real-time data deep processing method for traffic condition assessment as described in this application. The method can be executed by a cloud server or a traffic data processing center and includes the following steps: S101. Receive edge preprocessing data; wherein, the generation process of the edge preprocessing data includes: edge nodes acquiring multi-source traffic data and mapping the multi-source traffic data to a unified spatiotemporal reference.

[0032] In the initial stage of receiving edge preprocessed data, the system needs to uniformly map the set of traffic flow parameters (i.e., multi-source traffic data) collected from multiple channels such as roadside cameras, radar, geomagnetic sensors, and internet terminals to a spatiotemporal reference system containing a specified geographic coordinate system and standardized timestamps. To achieve this spatiotemporal reference registration, the edge computing nodes parse the original local coordinates and local hardware time of each device when processing asynchronously reported data packets. For the spatial dimension, the system constructs a homogeneous transformation matrix containing rotation matrices and translation vectors, and transforms the local coordinate coefficient values ​​of each device to the world geodetic coordinate system through matrix multiplication. For the temporal dimension, the system establishes a unified reference sampling frequency and uses a polynomial interpolation algorithm to calculate the corresponding data values ​​of each data source at a common reference time point, thereby eliminating the temporal domain deviation caused by asynchronous sampling.

[0033] At the engineering implementation level, spatiotemporal registration mechanisms have multiple feasible paths. For example, the system can extract the global timestamps distributed by the network time protocol to perform difference calculations and alignment on the data transmitted back by each device, and perform affine transformations on the original device locations based on the anchor point coordinates reserved in the high-precision map to complete spatial calibration. In addition, a clock synchronization compensation model based on Kalman filtering can be constructed, incorporating historical communication delay data of the devices to derive the current delay compensation amount, and then importing the time-aligned data into the coordinate system transformation equation to obtain the corresponding coordinates in a unified space.

[0034] For abnormal scenarios involving sudden interruptions in local road network satellite positioning and network timing signals, the system introduces a spatiotemporal reference maintenance mechanism based on inertial evolution. Under this mechanism, the system extracts historical spatiotemporal snapshot data from the moment before signal loss and activates the crystal oscillator drift compensation algorithm of the local hardware clock to maintain internal time increment. Based on this, and considering the topological relationships of neighboring non-blind zone nodes, a graph neural network is used to calculate the relative displacement vectors of target groups within the current blind zone. Finally, the local relative coordinates, combined with the drift compensation time, are re-derived and transformed into estimated values ​​under a virtual spatiotemporal reference. This mechanism ensures the continuity and reliability of traffic state assessment calculations in communication island environments.

[0035] S102. The edge node calculates the data importance score based on the feature attributes of the multi-source traffic data under the spatiotemporal reference, executes a differentiated data transmission strategy according to the data importance score, and sends the edge preprocessed data to the cloud.

[0036] To optimize limited network uplink bandwidth, edge computing nodes calculate the transmission value (i.e., data importance score) of data packets based on physical quantity fields (i.e., feature attributes) reflecting traffic flow conditions before sending data to the cloud. The system first parses the data packets, extracting feature attributes such as the state variance value representing the degree of vehicle driving anomalies and the congestion index of the current road segment. These are multiplied by a preset weighting constant determined based on statistical analysis and linearly accumulated to obtain a quantified importance score. Subsequently, the system reads the router's current uplink bandwidth utilization to set a dynamic grading threshold. Based on this threshold, the system executes targeted differentiated data transmission strategies: when the score is greater than the high-priority threshold, a full-load transmission strategy is implemented to retain the complete payload; when the score is in a lower range, a dimensionality reduction transmission strategy is implemented by framing or truncating redundant fields according to a set sampling rate. Data packets filtered and processed by the above strategies are finally merged into a pre-processed edge data subset and pushed to the sending queue.

[0037] In practical deployment, this differentiated transmission strategy can be implemented based on various data structures and algorithms. The system can establish a priority queue data structure based on scores, pushing data packets into corresponding level queues according to their scores, and prioritizing clearing high-scoring queues by polling and allocating network port sending resources. Alternatively, the system can classify information entropy retention levels based on score values, call principal component analysis algorithms to remove redundant dimensions from low-scoring data packets, and upload the compressed low-scoring data and uncompressed high-scoring data via a multiplexed channel.

[0038] In extreme communication scenarios where backbone communication links are suddenly interrupted due to geological disasters or power outages, the system introduces a semantic relay transmission mechanism based on a multi-hop ad hoc network. When the probe detects consecutive timeouts of WAN heartbeat packets, the system activates the backup radio frequency module of the short-range wireless ad hoc network built into the edge node and searches for undamaged neighboring nodes within the communication radius based on a distance vector routing algorithm. At this time, the conventional transmission strategy is forcibly downgraded to a core feature semantic extraction mode, extracting a concise semantic message containing accident coordinates and congestion length, which is relayed back to the cloud via the multi-hop network. This compensation mechanism effectively maintains the server's macro-level traffic assessment capabilities under extremely limited conditions.

[0039] S103. Extract continuous numerical index data from the edge preprocessing data, combine the dynamic credibility of each data source to perform feature-level filtering state estimation, and output the feature-level fusion state estimation value.

[0040] After receiving and parsing the edge preprocessed data, the cloud server first extracts continuous numerical indicators such as speed and traffic, which can be measured using real numbers and change smoothly over time. To address fluctuations in the quality of multi-source data, the system calculates a quantitative indicator of the anti-interference capability of each data source in the current time period, namely dynamic reliability. Specifically, the system retrieves the observation variance of each data source over historical periods, establishes the dynamic reliability value by calculating the reciprocal of the variance, and incorporates this value into the observation noise covariance matrix of the Kalman filter algorithm. During state transition and gain iteration, data sources with higher reliability are assigned smaller observation noise variances, thus occupying a greater influence weight when updating the state vector in the filter gain. This mechanism uses a recursive algorithm to infer the true state of the system, ultimately outputting a single-feature-layer fused state estimate that eliminates random errors from individual devices.

[0041] There are multiple evolutionary implementations of state estimation mechanisms at the feature level. One approach is to construct a polynomial-fitted state transition matrix, using dynamic confidence to adjust particle resampling weights in the particle filter algorithm, and performing a weighted summation of the state vectors of high-weight particles to obtain the fusion result. Another method involves establishing an extended Kalman filter equation that includes a forgetting factor, substituting dynamic confidence into the derivative of the Kalman gain, and using a nonlinear Jacobian matrix to linearize and iteratively update the continuous numerical index data.

[0042] Faced with concurrent resonance interference from multiple sensors caused by weather and the potential for false rejection of stationary vehicles in congested areas, the system introduces cross-modal microphysical coupling and static space occupancy matrix verification logic. If a high-frequency transient jump is detected in the state estimate within the time window corresponding to the physical acceleration limit, and the environment is characterized by precipitation or low visibility, the system determines that it has entered the concurrent resonance interference mode. The system simultaneously extracts the micro-motion frequency domain distribution characteristics of the echoes from active detection devices and the local morphological dynamic disturbance characteristics of the target area in the video, and conducts spatiotemporal correlation analysis to construct a microphysical coupling matrix. By comparing the specific physical correlations of water splashes or airflow disturbances caused by actual vehicle movement, the system identifies and removes false target data induced by environmental distortion, and then introduces a robust filtering algorithm to reconstruct the state estimate. If the reconstructed estimate shows a group disappearance of targets in a continuous spatial segment with an initial deceleration trend, the system determines that it has entered a congested and stationary scene. At this time, the system extracts the static occlusion features of the video background and the shadow features of radar detection signals, and constructs a static space occupancy matrix through topological mapping. The system imports the disappearance coordinates into the matrix for secondary verification, marks the coordinates within the effective area as real stationary target data, assigns a stationary state vector approaching zero, and forces it into the filtering algorithm for update. This mechanism filters out false alarm interference and compensates for the detection blind spot of stationary targets, ensuring the confidence of state estimation in complex environments.

[0043] S104. Extract heterogeneous independent index data representing traffic status from the edge preprocessing data, construct a conflict correction factor based on the conflict coefficient between each independent index data, and calculate it using the evidence synthesis rule with the conflict correction factor, and output the decision layer fusion result.

[0044] While feature layer calculations are performed, the system conducts decision-level inference based on the qualitative judgment results (i.e., heterogeneous independent index data) that appear as discrete category labels within the data packets. The system first converts the qualitative descriptions of each sensor into basic probability assignment functions for each set of state hypotheses. By traversing the probability assignments of pairwise data sources, the system accumulates the product of probabilities where the intersection is an empty set to quantify the degree of contradiction between conclusions from different data sources, thus deriving a conflict coefficient. Subsequently, the system utilizes a negative exponential decay or smoothed inverse proportional function (e.g., calculated using the formula W). c =1 / (1 +αK ij A conflict correction factor is constructed that is inversely proportional to the conflict coefficient in a generalized sense. This factor is then multiplied as a weighting coefficient into the basic probability allocation function, which forces highly conflicting probabilities to be assigned to the unknown set, reducing the interference of contradictory evidence on the final probability. Finally, the adjusted probabilities are substituted into the evidence synthesis rules (i.e., algebraic operational rules), and the category corresponding to the highest joint probability is taken as the fusion result of the decision layer.

[0045] Evidence synthesis with conflict correction factors has several extended implementations. The system can generate a dedicated discount matrix based on the extracted conflict coefficients, use this matrix to perform attenuation operations on the original probability assignments, and then perform orthogonal summation to obtain the state class probabilities. Furthermore, the system can also calculate the Pearson correlation coefficient between independent indicator data to measure the evidence gap, calculate the correction factor based on the gap and assign it to local support, and then use a weighted average rule to output a unified decision conclusion.

[0046] In situations where sudden road network disruptions lead to conflicting heterogeneous data, the system implements traffic shock front identification and forced risk anchoring logic. When the conflict coefficient exceeds a preset safety threshold (extreme conflict threshold), and the heterogeneous independent index data exhibits an extremely contradictory step distribution in the longitudinal topology of the road space, the system determines that a high-risk traffic shock front conflict mode has been triggered and proactively suspends conventional evidence synthesis calculations. The system calculates the state gradient difference between adjacent spatial nodes along the longitudinal topology of the road, and identifies the spatial coordinates where the dynamic slope exceeds the safety threshold as a state-abrupt shock front. For this front and its associated sections in the direction of oncoming traffic, the system skips the probability numerical smoothing process for conflict correction and forcibly anchors its decision category to the highest level of safety risk category, which is then output as the decision-level fusion result. This mechanism can overcome the neutralization effect of conventional probability synthesis on local high-risk early warnings, ensuring a rapid response to the fault characteristics of sudden accidents.

[0047] Figure 3This is a logical diagram illustrating the dual-track fusion of multi-source data feature layer and decision layer provided in an embodiment of this application. Combined with... Figure 3 As shown, in the cloud processing stage, the system performs traffic splitting calculations based on the different forms of perceived data. For continuous numerical index data (such as vehicle speed data collected by devices), the system inputs it into the feature layer fusion module. In this module, the system dynamically adjusts the observation noise covariance and filter gain in the filtering algorithm by combining the dynamic credibility and environmental influence coefficients corresponding to each data source to suppress high-error data and output the feature layer fusion state estimate. Simultaneously, for heterogeneous independent index data representing traffic conditions (such as flow rate and occupancy status), the system inputs it into the decision layer fusion module. This module assigns basic probabilities to each evidence source, constructs a conflict correction factor by extracting the conflict coefficient between evidence, and uses this correction factor to weight and adjust the evidence with discrepancies before performing evidence synthesis. Finally, the underlying conclusions of the dual-track processing are uniformly output as a comprehensive index set containing the fusion state, eliminating the perceptual discrepancies of multi-source heterogeneous data in the spatiotemporal dimensions. S105. Combine real-time acquired environmental feature data and event feature data to calculate the scene correction factor for the current traffic scenario.

[0048] The theoretical operating state of road traffic flow is often suppressed and disturbed by abnormal external environments. Therefore, the system acquires relevant objective physical quantities by asynchronously listening to meteorological service interfaces and traffic management platform broadcasts. The system extracts environmental characteristic data representing visibility and precipitation intensity, as well as event characteristic data representing traffic accidents and road occupancy information. Subsequently, the system numerically encodes the above multidimensional parameters, for example, mapping rainfall to hydroplaning coefficients and accident-occupied roads to capacity reduction coefficients. The system constructs an evaluation function for nonlinear surface fitting, substituting the encoded parameters as independent variables into the calculation. The dependent variable value produced by this function is the scene correction factor, which represents the attenuation ratio of the current mixed complex scene to the ideal road capacity, and is used to compensate for the theoretical evaluation value in subsequent steps.

[0049] At the level of specific mathematical modeling, the derivation method of the scene correction factor can be replaced. The system can establish a fuzzy inference rule base containing environmental input nodes, perform fuzzification processing on multi-source feature data using membership functions, and then solve for the real-valued correction factor using the centroid method after fuzzy logic operations. In addition, the system can also construct a multivariate regression analysis equation, match and retrieve the statistical characteristics of traffic efficiency attenuation under similar weather and event overlap conditions in historical data, and input real-time features into the regression equation to calculate the scene correction ratio for the current road condition.

[0050] S106. Based on the feature layer fusion state estimate and the decision layer fusion result, obtain the preliminary evaluation value, and use the scene correction factor to dynamically correct the preliminary evaluation value, and output the final traffic state evaluation result.

[0051] After the underlying state machine estimation and evidence synthesis operations are completed, and external constraint factors are ready, the system performs a comprehensive calibration of the state baseline results. First, the system uses the congestion level determined by the decision-level fusion results as a mathematical constraint interval, mapping the real-valued results of the feature-level fusion state estimates to this interval. This yields a pure traffic flow theoretical calculation baseline, without considering the suppression of unforeseen events—the preliminary assessment value. Subsequently, the system retrieves pre-calculated scenario correction factors and applies them as discount multipliers to this preliminary assessment value. This dynamically corrected algebraic multiplication operation essentially subtracts the speed and flow reduction caused by adverse weather or physical road occupancy affecting vehicle driving intentions and dynamic braking limits. The output is the final traffic state assessment result that conforms to the constraints of the current complex environmental conditions.

[0052] The calibration calculation process for the preliminary assessment values ​​can also integrate mechanical and probabilistic models. The system can analyze the average vehicle speed parameter in the preliminary assessment values ​​and substitute the scene correction factor into the safety distance calculation term in the vehicle following model based on Newtonian mechanics, resolving the model's differential equations to output the correction value. Alternatively, this process can be based on Markov chain expansion, where the preliminary assessment values ​​are established as prior state variables, a state transition probability decay matrix is ​​constructed using the scene correction factor, and the final posterior state distribution vector after environmental disturbance correction is derived through matrix multiplication.

[0053] Figure 4 This is a schematic diagram illustrating the process of dynamic traffic state calibration based on scene correction factors provided in an embodiment of this application. Figure 4 As shown, during the evaluation and calibration phase, the system receives the fused feature data and simultaneously extracts real-time external weather, time, and event data. The system inputs these multi-dimensional features into a scene recognition classifier to identify the specific environmental pattern of the current road network and classify it into corresponding scene labels such as regular commuting, severe weather, sudden accidents, or large-scale events. Subsequently, the system obtains preliminary evaluation values ​​based on the underlying fusion results and extracts scene correction factors that match the current classification scene. The system performs a composite calculation by overlaying the preliminary evaluation values ​​and scene correction factors, deducting the suppression and reduction of traffic efficiency caused by environmental interference and sudden events, and outputs the final traffic state evaluation result under the composite scene, generating corresponding decision control instructions accordingly. Furthermore, the system continuously collects actual road condition evolution data after the decision is issued as feedback samples, triggering parameter iteration in the online learning module to dynamically adjust the factor weights under each scene, ensuring the adaptability of the system's evaluation logic to the evolution of complex actual road conditions.

[0054] S107. Generate corresponding equipment control commands and traffic guidance information based on the final traffic condition assessment results.

[0055] At the end of a single computation cycle, the system drives the actual traffic engineering response based on the final set of traffic state parameters. On one hand, the system analyzes the segment queue length and node delay indices included in the final evaluation results, substituting them into the road network green wave coordination control algorithm or single-point timing optimization algorithm to solve for the optimal signal cycle and green ratio parameters of the target intersection. Subsequently, the system encapsulates messages according to the communication protocol format agreed upon by the underlying execution hardware (such as traffic signals or ramp controllers), generating equipment control commands for distribution. On the other hand, the system extracts the congestion range coordinate set and vehicle speed distribution data from the results, calls a natural language generation template to stitch and integrate the above data into structured text containing descriptions of slow-moving distances and detour route suggestions, generates traffic guidance information for public travelers, and pushes it to the variable message signs on the road for rendering and display.

[0056] The distribution of control instructions and guidance information has multiple scalability pathways. On the control side, the system can extract traffic volume saturation parameters, introduce macroscopic basic map (MFD) models to generate control instructions that constrain the traffic rate at the boundary of the control area, and distribute them to distributed roadside computing units via a dedicated wireless communication network. On the guidance side, the system can convert the assessment results into standard structured traffic condition release source data, perform text-to-speech (TTS) encoding and synthesis on the generated text guidance information, and then directly broadcast the voice guidance information to vehicle terminals within the congestion-affected area using vehicle-mounted shortwave broadcast frequencies or vehicle-to-everything (V2X) communication interfaces.

[0057] In the above embodiments, by performing unified spatiotemporal mapping and edge-side hierarchical filtering on multi-source traffic data, the ineffective loss of communication bandwidth is reduced. Combining feature-layer adaptive filtering and noise reduction with decision-layer conflict correction evidence synthesis eliminates the inherent data jumps of a single sensor and decision discrepancies between multi-source heterogeneous sensors. Finally, the introduction of multi-dimensional environmental features to dynamically correct the evaluation values ​​improves the accuracy and robustness of road network state assessment in complex traffic scenarios.

[0058] In other embodiments of this application, when there is a surge in concurrent road network sensing data, network congestion and delays in critical data transmission may occur due to full data backhaul. The real-time data deep processing method for traffic state assessment provided in this application can calculate data scores based on multi-dimensional traffic characteristics and perform differentiated dimensionality reduction transmission to ensure priority backhaul of high-value data.

[0059] like Figure 5 The diagram shown is another flowchart illustrating a real-time data deep processing method for traffic condition assessment provided in this application, including the following steps: S201. Receive edge preprocessing data; wherein, the generation process of the edge preprocessing data includes: edge nodes acquiring multi-source traffic data and mapping the multi-source traffic data to a unified spatiotemporal reference.

[0060] Step S201 and Figure 2 Step S101 in the illustrated embodiment is similar and can be found in the description of step S101, which will not be repeated here.

[0061] S202, The edge node extracts the road segment risk level, real-time traffic status and time characteristics corresponding to the multi-source traffic data.

[0062] Road segment risk level refers to a quantitative indicator of road safety threat assessed based on historical accident rates or road geometric alignment attributes; real-time traffic status is a classification label used to represent the degree of road congestion or the smoothness of traffic flow at the current moment; time characteristics refer to the attribution of the data collection time on the timeline of the whole day, which is usually used to distinguish between morning and evening peak hours and off-peak hours.

[0063] The system performs this feature extraction step after the spatiotemporal reference of multi-source traffic data is unified and cached. The system reads the aligned data stream label dictionary. It then calls the road network static attribute table and dynamic weather interface pre-installed in the edge node memory to match and derive the road segment risk level value corresponding to the current coordinates. The system uses a speed extraction function to calculate the average spatial headway of vehicles and substitutes it into the traffic flow density formula to classify real-time traffic conditions as congested or smooth. Finally, the system extracts the data packet generation timestamp and uses time modulo arithmetic to extract the current hour and minute variables, thereby constructing a time feature vector containing periodic patterns.

[0064] In some embodiments, feature parameters can be extracted in multiple ways: Optionally, the system parses the geographic location information of traffic data messages; carries this location information to access the locally built-in historical accident black spot database to obtain a basic risk value; combines the current ambient light index to perform a reduction calculation on the basic risk value to output the road segment risk level; Optionally, the system inputs the time-series vehicle coordinates into a sliding window speed measurement function to calculate the instantaneous speed; calculates the vehicle speed variance and queue length within the window; substitutes the speed variance and queue length into a preset fuzzy inference matrix to output the corresponding real-time traffic state parameters. It is understood that other methods can also be used to extract multi-dimensional features, which are not limited here.

[0065] S203. The edge node is assigned a corresponding weight coefficient to the road segment risk level, the real-time traffic status and the time feature respectively, and the weighted sum is performed to obtain the data importance score.

[0066] Weighting coefficients are proportional constants used to measure the contribution of each data feature to the overall traffic status assessment; weighted summation is an algebraic operation process that multiplies the normalized values ​​of each feature by their corresponding weights and then performs linear summation; data importance score is a scalar measure of the priority transmission value of a certain sensing data packet under limited network bandwidth.

[0067] The system reads the pre-defined weight vectors corresponding to the road segment risk level, real-time traffic status, and time features from the local configuration file. The system constructs a feature column vector from the feature scalar values ​​of these three dimensions and performs an inner product operation between the feature column vector and the weight row vector. When a feature value is missing from some data sources, the system uses a linear interpolation model or historical moving average to smoothly fill in the missing value, ensuring dimensional consistency in the matrix operations. The dot product of the feature matrices represents the data importance score, indicating the transmission priority of the data packet.

[0068] In some embodiments, the weighted summation calculation of importance scores can be implemented in several ways: Optionally, the system reads the locally stored hierarchical analysis weight configuration file; parses the floating-point weight parameters corresponding to each feature label; multiplies the normalized feature value with the corresponding weight and executes an adder instruction to obtain the score; Optionally, the system collects the feature distribution matrix from historical abnormal traffic events; calls the principal component analysis algorithm to calculate the information entropy contribution of each feature dimension as a dynamic weight; updates the weight pool in memory according to the dynamic weight coefficient and performs a vector dot product to obtain the final score. It is understood that other methods can also be used to calculate the score, and this is not limited here.

[0069] S204. The edge node compares the data importance score with a preset score range. Based on the range level determined by the comparison result, it executes either a full transmission strategy or a dimensionality reduction sampling ratio transmission strategy, outputs the edge preprocessed data, and sends it to the cloud.

[0070] The preset scoring range refers to the numerical boundary range used to divide data priority levels, which is derived by R&D personnel based on experimental data of communication carrying capacity under different network bandwidth conditions; the full transmission strategy refers to a data forwarding mechanism that sends the original data packets to the cloud with the complete payload without any compression or discarding; the dimensionality reduction sampling ratio transmission strategy refers to a network transmission method that reduces the data volume by removing redundant fields or reducing the sampling frequency; edge preprocessed data refers to the traffic data set that is finally stored in the gateway sending queue after being filtered and processed by the aforementioned strategies.

[0071] Before calculating the data importance score for each data packet and preparing for WAN backhaul, the system executes this policy routing step. The system calls its internal comparator logic unit to determine the range of values ​​within which the current data importance score falls. If the score falls into the high-priority range, the system maintains the original payload structure of the data packet, encapsulates it with the highest-level network service quality label, and executes the full transmission policy. If the score falls into the low-priority range, the system calls a downsampling algorithm to truncate some data frames from the continuous video stream, or calls a feature mapping function to remove non-critical dimension attributes from the radar point cloud data packets. It then executes a dimensionality reduction sampling ratio transmission policy according to the calculated compression ratio, unifies the data packets processed by different policies, and outputs them as edge preprocessing data.

[0072] In some embodiments, the execution of differentiated transmission strategies and data output can be achieved in several ways: Optionally, the system inputs the data importance score into a binary search tree to locate the corresponding threshold interval; extracts keyframes containing complete background information for low-priority video data streams; discards predicted frames with small inter-frame differences and repackages them into low-bitrate streaming media files for output; Optionally, the system pushes data packets into different priority buffer queues based on the score interval level; calls principal component analysis functions to remove redundant coordinate variables for low-score radar queues; merges the retained principal component features, assembles them into a simplified message, and pushes it to the sending port. It is understood that other methods can also be used to achieve dimensionality reduction transmission of data, which are not limited here.

[0073] In some embodiments, when faced with sudden interruptions to the backbone communication link due to geological disasters or power outages, the real-time data deep processing system for traffic condition assessment can also perform semantic relay transmission based on a multi-hop ad hoc network. Specifically, the system detects continuous timeouts of wide area network heartbeat packets using probes; wakes up the backup radio frequency module of the short-range wireless ad hoc network built into the edge node; searches for non-paralyzed neighboring nodes within the communication radius based on a distance vector routing algorithm; and forcibly downgrades the full transmission strategy to a core feature semantic extraction mode, extracting a simplified semantic message containing accident coordinates and congestion length, and relaying this message back to the cloud via a multi-hop network. This method improves the accuracy of traffic condition assessment in complex and special scenarios.

[0074] S205. Extract continuous numerical index data from the edge preprocessing data, combine the dynamic credibility of each data source to perform feature-level filtering state estimation, and output the feature-level fusion state estimation value.

[0075] In some embodiments, when faced with complex situations where the quality of multi-sensor data fluctuates due to external environmental interference, high-error data interference can be reduced by calculating the dynamic reliability of the data source and reconstructing the noise covariance matrix of the filtering algorithm.

[0076] Specifically, after extracting continuous numerical indicators such as vehicle speed and space occupancy, the system initiates a quantitative assessment of the data source reliability. The system extracts historical detection deviation values ​​for each sensor from the underlying logs over past operating cycles; these values ​​characterize the inherent hardware error distribution of the equipment. Simultaneously, the system reads a preset environmental impact coefficient, which is a fixed reduction parameter derived by researchers based on signal penetration attenuation experimental data from optical and microwave sensors under different meteorological conditions such as rainfall and fog. The system inputs this historical detection deviation value and the environmental impact coefficient into an evaluation function for cross-multiplication and reciprocal calculation to determine the dynamic reliability of each data source in the current time period. Through this coupled calculation, the system can quantify the true usability level of a specific sensor under current meteorological conditions, avoiding blindly trusting optical detection equipment when line of sight is obstructed. Subsequently, the system establishes a Kalman filter state transition machine, using the dynamic reliability as a function variable of the weighting factors to dynamically replace the originally statically configured original observation noise covariance matrix in the Kalman filter algorithm. During the filtering gain iteration, data sources with lower dynamic confidence values ​​are assigned a larger observation noise variance variable. This allows the Kalman gain matrix to adaptively reduce the weight of the problematic data source on the posterior estimation of the system state when calculating the state update. This matrix substitution mechanism ensures that the underlying filtering process no longer relies on fixed confidence assumptions, but dynamically allocates fusion weights based on changes in the external environment, guaranteeing that the output feature layer fused state estimate always converges towards the high-confidence data source.

[0077] By introducing environmental constraints and historical biases to calculate dynamic reliability, and using this to reconstruct the observation noise matrix of the Kalman filter, the system can adaptively reduce the weights of data sources affected by environmental disturbances during gain iteration. This overcomes the weakness of fixed-parameter filtering models that are prone to failure under sudden environmental changes, and improves the reliability of feature layer data fusion.

[0078] S206. Extract heterogeneous independent index data representing traffic status from the edge preprocessing data, construct a conflict correction factor based on the conflict coefficient between each independent index data, and calculate it using the evidence synthesis rule with the conflict correction factor, and output the decision layer fusion result.

[0079] In some embodiments, when faced with contradictions and disagreements in traffic state determination based on multi-source heterogeneous data, the interference of highly conflicting data on decision-making results can be reduced by introducing conflict factors and reconstructing evidence synthesis rules.

[0080] Specifically, after extracting heterogeneous and independent indicator data representing traffic conditions, such as discrete road condition labels identified by video and traffic flow qualitative levels determined by radar, the system uses a preset probability mapping function to calculate basic probability allocation values ​​for each traffic condition assumption category (such as smooth flow, slow flow, congestion, etc.). The preset probability mapping function refers to a probability transformation mathematical relationship derived by those skilled in the art based on the physical characteristics of different sensors and historical classification statistical distributions. Through this mapping transformation, the system unifies heterogeneous data from different dimensions into a standard probability measure space, facilitating subsequent algebraic operations. Next, the system iterates through the probability allocation values ​​generated by each data source, performing cross-accumulation operations on the probability products of different data sources pointing to different state categories (i.e., the intersection is an empty set), thereby extracting a conflict coefficient representing the degree of contradiction between data. Extracting the conflict coefficient quantifies the degree of discrepancy between the judgment results of different sensors, providing an objective mathematical basis for subsequent weight adjustments. Subsequently, the system constructs a conflict correction factor, inversely proportional to the conflict coefficient, using an algebraic form that subtracts from the conflict coefficient or exhibits a similar reciprocal decay. This correction factor is then used as a discount multiplier to weight and adjust the original basic probability assignment values. This approach allocates a portion of the effective probability quality from highly conflicting data sources to the unknown set, artificially weakening the dominance of highly contradictory data sources in the global synthesis. Finally, the system substitutes the weighted basic probability assignment values ​​into the Dempster-Schaffer evidence synthesis formula to perform orthogonal summation and synthesis operations, calculating the final joint probability of each state hypothesis. The category corresponding to the highest synthesis probability is then selected as the decision-level fusion result. This step integrates local evidence into a global conclusion through probability synthesis, ensuring the objectivity of qualitative judgments.

[0081] By extracting the conflict coefficient of heterogeneous data and constructing an inverse correction factor, the basic probability allocation values ​​are weighted and adjusted before evidence synthesis, enabling the system to adaptively weaken the interference of contradictory evidence. This avoids the computational paradoxes that traditional algorithms produce under severe data conflict and improves the reliability of multi-source heterogeneous data decision-making fusion.

[0082] S207. Combine real-time acquired environmental feature data and event feature data to calculate the scene correction factor for the current traffic scenario.

[0083] S208. Based on the feature layer fusion state estimate and the decision layer fusion result, obtain the preliminary evaluation value, and use the scene correction factor to dynamically correct the preliminary evaluation value, and output the final traffic state evaluation result.

[0084] S209. Generate corresponding equipment control commands and traffic guidance information based on the final traffic condition assessment results.

[0085] Steps S205-S209 and Figure 2Steps S103-S107 in the illustrated embodiment are similar and can be found in the descriptions of steps S103-S107, which will not be repeated here.

[0086] In the above embodiments, by prioritizing the transmission of high-value abnormal data with limited network bandwidth resources, the uplink volume of normal redundant data is significantly reduced. This avoids communication link congestion and node processing delays caused by concurrent data backhaul, and improves the system's computational efficiency and network robustness in handling large-scale sensing data under limited bandwidth conditions.

[0087] The following describes an exemplary real-time data deep processing system 300 for traffic condition assessment provided in an embodiment of this application. Figure 6 This is an exemplary hardware structure diagram of a real-time data deep processing system 300 for traffic condition assessment provided in an embodiment of this application.

[0088] In some embodiments, the real-time data deep processing system 300 for traffic condition assessment is a computer device or includes a computer device. The computer device includes a processor, memory, and a network interface connected via a system bus. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the computer device stores data. The network interface of the computer device is used to communicate with other external terminals or servers via a network connection. In some embodiments, the network interface can be a wired network interface; in some embodiments, the network interface can also be a wireless network interface. When the computer program is executed by the processor, it implements the methods in the embodiments of this application.

[0089] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0090] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0091] As used in the above embodiments, depending on the context, the term "when..." can be interpreted as "if...", "after...", "in response to determining...", or "in response to detecting...". Similarly, depending on the context, the phrase "when determining..." or "if (the stated condition or event) is interpreted as "if determining...", "in response to determining...", "when (the stated condition or event) is detected", or "in response to detecting (the stated condition or event)".

[0092] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive), etc.

[0093] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A real-time data deep processing method for traffic condition assessment, characterized in that, include: Receive edge preprocessed data; wherein, the edge preprocessed data is obtained by the edge node from multi-source traffic data and mapped to a unified spatiotemporal reference, and then the data importance score is calculated based on feature attributes, and a differentiated data transmission strategy is executed in a hierarchical manner before being sent; Extract continuous numerical index data from the edge preprocessing data, combine the dynamic credibility of each data source to perform feature-level filtering state estimation, and output the feature-level fusion state estimation value. Heterogeneous independent index data representing traffic status are extracted from the edge preprocessing data. A conflict correction factor is constructed based on the conflict coefficient between each independent index data. The evidence synthesis rule with the conflict correction factor is used to calculate and output the decision layer fusion result. By combining real-time acquired environmental and event feature data, the scene correction factor for the current traffic scenario is calculated. Based on the feature layer fusion state estimate and the decision layer fusion result, a preliminary assessment value is obtained, and the preliminary assessment value is dynamically corrected using the scene correction factor to output the final traffic state assessment result. Based on the final traffic condition assessment results, corresponding equipment control commands and traffic guidance information are generated.

2. The method according to claim 1, characterized in that, The feature attributes of the multi-source traffic data based on the spatiotemporal reference are used to calculate the data importance score, and a differentiated data transmission strategy is executed according to the data importance score to output edge preprocessing data, specifically including: Extract the road segment risk level, real-time traffic status, and time characteristics corresponding to the multi-source traffic data; The risk level of the road segment, the real-time traffic status, and the time feature are each assigned a corresponding weight coefficient and then summed in a weighted manner to obtain the data importance score. The data importance score is compared with a preset score range. Based on the range level determined by the comparison result, a full transmission strategy or a dimensionality reduction sampling ratio transmission strategy is executed respectively to output the edge preprocessed data.

3. The method according to claim 1, characterized in that, The step of extracting continuous numerical index data from the edge preprocessing data, combining the dynamic reliability of each data source with feature-level filtering state estimation, and outputting feature-layer fusion state estimation values ​​specifically includes: Extract continuous numerical index data from the edge preprocessing data, and calculate the dynamic reliability of each data source based on the historical detection deviation value of each data source and the preset environmental impact coefficient. The dynamic confidence level is used as a weighting factor to replace the original observation noise covariance matrix in the Kalman filter algorithm, and the filter gain is iterated to output the estimated value of the feature layer fusion state.

4. The method according to claim 1, characterized in that, The process involves extracting heterogeneous independent indicator data representing traffic conditions from the edge preprocessing data, constructing a conflict correction factor based on the conflict coefficient between the independent indicator data, and calculating the result using evidence synthesis rules with the conflict correction factor. The result is then used to output the decision-level fusion result, specifically including: Extract heterogeneous independent index data representing traffic state from the edge preprocessing data, and calculate the basic probability assignment value for each traffic state category based on the heterogeneous independent index data; The conflict coefficient is extracted by cross-accumulating the basic probability allocation values. Construct a conflict correction factor that is inversely proportional to the conflict coefficient, and use the conflict correction factor to perform a weighted adjustment on the basic probability allocation value; The basic probability allocation values ​​after weighted adjustment are synthesized using evidence synthesis rules, and the category corresponding to the highest synthesis probability is selected as the fusion result of the decision layer.

5. The method according to claim 1, characterized in that, After extracting continuous numerical index data from the edge preprocessing data, combining it with the dynamic reliability corresponding to each data source to perform feature-level filtering state estimation, and outputting the feature-level fusion state estimate, the method further includes: When the estimated value of the feature layer fusion state is detected to have concurrent high-frequency transient jumps within a preset time window, and the environmental feature data is characterized as an abnormal scene of precipitation or road surface attachment superimposed with low visibility, it is determined that the multi-source sensor concurrent resonance interference mode is entered. Simultaneously extract the micro-motion frequency domain distribution characteristics of the echo signal from the active detection device within this time sequence, as well as the local morphological dynamic disturbance characteristics of the corresponding video feature data in the target area; By performing cross-modal spatiotemporal correlation analysis on the frequency domain distribution characteristics of the micro-motion and the local morphological dynamic disturbance characteristics, a micro-physical coupling matrix is ​​constructed to characterize the physical property of the road surface medium splashing or disturbance excited by the motion of the real target. The target data that causes high-frequency transient jumps is input into the microphysical coupling matrix for cross-comparison and identification. Target data that does not have a specific spatiotemporal coupling relationship is identified as false target heterogeneous data induced by physical environment distortion and multipath reflection clutter and is then removed. For the remaining real target data after removing the heterogeneous data of false targets, a filtering algorithm with robust adjustment and adaptive attenuation characteristics is introduced to re-iterate the state gain and output the reconstructed feature layer fusion state estimate.

6. The method according to claim 5, characterized in that, After the remaining real target data, after removing and eliminating heterogeneous data of false targets, is subjected to a filtering algorithm with robust adjustment and adaptive attenuation characteristics to re-iterate the state gain and output the reconstructed feature layer fusion state estimate, the method further includes: When it is detected that the reconstructed feature layer fusion state estimate shows a sudden disappearance of the target group in a continuous spatial segment, and the state estimate in the preceding time series shows a continuous deceleration trend, it is determined that the current road segment has entered a secondary static scene of abnormal scene superimposed with congestion and stagnation. Extract the static occlusion features of the environmental background from the video feature data of the corresponding spatial segment in the current time series, and simultaneously extract the detection signal occlusion shadow features formed by the echo signal of the active detection device behind the target; The static occlusion features and the detection signal occlusion shadow features are spatially topologically mapped to construct a static spatial occupancy matrix that represents the physical stationary volume of the real target. The spatial coordinates of the sudden disappearance of the target group are backtracked into the static spatial occupancy matrix for secondary verification, and the spatial coordinates falling within the effective area of ​​the static spatial occupancy matrix are extracted. The extracted spatial coordinates are marked as the real lingering target data that was mistakenly rejected, and they are assigned a static state vector that approaches zero. The real target data is forcibly fed into the filtering algorithm with robust adjustment and adaptive attenuation characteristics for state update, and the corrected feature layer fusion state estimate is output.

7. The method according to claim 1, characterized in that, The process involves extracting heterogeneous independent indicator data representing traffic conditions from the edge preprocessing data, constructing a conflict correction factor based on the conflict coefficient between the independent indicator data, and calculating the result using evidence synthesis rules with the conflict correction factor. The result is then used to output the decision-level fusion result, specifically including: Extract heterogeneous independent index data representing traffic state from the edge preprocessing data, calculate the basic probability assignment value for each traffic state category, and extract the conflict coefficient between the heterogeneous independent index data. When the conflict coefficient exceeds the preset extreme conflict threshold, and the corresponding heterogeneous independent index data shows an extremely contradictory step distribution of motion states in the longitudinal topology of the road space, it is determined that a high-risk conflict mode of traffic shock wave is triggered, and the regular evidence synthesis rule is suspended. The dynamic state gradient difference between the heterogeneous independent index data is calculated along the longitudinal topology of the road space, and the spatial node coordinates where the dynamic state gradient difference exceeds the critical safety threshold are identified as state-abrupt shock fronts. For the state-change shock front and its associated vehicle direction section, skip the probability synthesis based on the conflict correction factor, forcibly anchor its decision category to the highest level of safety risk category, and output it as the decision layer fusion result.

8. A real-time data deep processing system for traffic condition assessment, characterized in that, The real-time data deep processing system for traffic condition assessment includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code including computer instructions, and the one or more processors call the computer instructions to cause the real-time data deep processing system for traffic condition assessment to perform the method as described in any one of claims 1-7.

9. A computer program product containing instructions, characterized in that, When the computer program product is run on a real-time data deep processing system for traffic condition assessment, the real-time data deep processing system for traffic condition assessment performs the method as described in any one of claims 1-7.

10. A computer-readable storage medium comprising instructions, characterized in that, When the instruction is executed on a real-time data deep processing system for traffic condition assessment, the real-time data deep processing system for traffic condition assessment performs the method as described in any one of claims 1-7.

Images