An intelligent intersection adaptive lighting and traffic signaling system fusing car-road cooperation and visual perception

Abstract

The application discloses a fusion intelligent intersection adaptive lighting and traffic signaling system combining vehicle-road cooperation and visual perception, and belongs to the technical field of intelligent transportation. The system comprises a data fusion module, which is used for collecting vehicle-road cooperation data and visual perception data, and performing space-time alignment and fusion on the vehicle-road cooperation data and the visual perception data to generate a unified global traffic state view; a first processing module, which is used for determining a traffic efficiency cost based on average vehicle delay time and average queue length contained in the unified global traffic state view, and in combination with a preset reference delay time, a preset reference queue length and a weight coefficient; and a second processing module, which is used for determining decision explainability based on a decision model activated by a current system and based on an explainability score preset and bound for the decision model. The application can effectively resist data-level attacks or failures and ensure the robustness of the system.

Description

Technical Field

[0001] The present invention relates to the field of intelligent transportation, and specifically to a smart intersection adaptive lighting and traffic signaling system that integrates vehicle-road collaboration and visual perception. Background Art

[0002] In the field of smart intersection management, existing traffic signal and lighting control systems usually have improving traffic efficiency as the single optimization goal, and their operating logic is based on the idealized premise that all input data is true and accurate.

[0003] This design status quo has led to significant technical bottlenecks and potential risks. First, the system lacks the ability to perceive and diagnose its own operating status. When roadside sensors such as cameras, radars, or vehicle-road cooperative communication units generate erroneous or false data due to failures, bad weather, or malicious attacks, the system cannot effectively identify this contaminated information. Second, many advanced optimization algorithms, such as deep learning models, have opaque decision-making processes, like a black box. This makes the system's decision-making behavior difficult for managers to understand, audit, and trust, bringing compliance and security challenges. In addition, the rigid single-objective optimization strategy makes it lack the ability to flexibly respond to complex situations such as extreme traffic congestion or a decline in the system's internal health. It may even aggravate congestion or cause functional paralysis due to inappropriate control strategies.

[0004] The root cause of the above problems lies in the limitations of existing technical solutions in top-level design. They generally lack a multi-dimensional internal state evaluation system that can quantitatively evaluate the system's own data quality, decision-making stability and logical transparency. The system is merely a passive responder to external traffic information, rather than an intelligent entity that can actively review its own status and actively defend against risks.

[0005] When faced with real risks such as data contamination or algorithm failure, traditional systems are not only unable to guarantee the safety and robustness of intersections, but the credibility of their management and control behaviors cannot be guaranteed, making it difficult to meet the highest requirements for safety, reliability, and transparency of critical urban infrastructure. Summary of the Invention

[0006] The purpose of the present invention is to provide a smart intersection adaptive lighting and traffic signaling system that integrates vehicle-road collaboration and visual perception to solve the problems raised in the above background technology.

[0007] The technical solution of the present invention is to include:

[0008] The data fusion module is used to collect vehicle-road cooperative data and visual perception data, and align and fuse the vehicle-road cooperative data and visual perception data in time and space to generate a unified global traffic status view;

[0009] A first processing module is configured to determine a traffic efficiency cost based on the average vehicle delay time and average queue length included in the unified global traffic status view, combined with a preset reference delay time, reference queue length, and a weight coefficient;

[0010] A second processing module is configured to determine the decision interpretability based on the decision model currently activated by the system and based on the interpretability score preset and bound to the decision model;

[0011] a third processing module for determining a system risk index based on a unified global traffic status view and control instructions output by the system;

[0012] The mode decision module is used to determine the system operation mode based on the system risk index and the preset risk threshold, combined with the traffic efficiency cost and the preset cost upper limit;

[0013] The control execution module is used to activate the corresponding decision model in response to the system operation mode to adjust the control strategy of traffic signaling and lighting.

[0014] Preferably, the third processing module determines the system risk index, including:

[0015] Visual perception data and vehicle-road collaboration data are mapped into a visual density matrix and a collaborative density matrix, respectively. The cross-source conflict degree is determined based on the difference between the visual density matrix and the collaborative density matrix. The system risk index is determined by combining the time average of the cross-source conflict degree and the time series stability of the system output control instructions.

[0016] Preferably, the second processing module determines the decision explainability, including:

[0017] A set of preset decision models, where each model in the set is arranged in ascending order of complexity and is associated with a monotonically decreasing preset interpretability score;

[0018] The decision interpretability is determined as a preset interpretability score corresponding to the decision model currently activated by the control execution module.

[0019] Preferably, the mode decision module determines the system operation mode, including:

[0020] When the system risk index is lower than the preset risk warning threshold, the system operation mode is determined to be the optimal efficiency mode;

[0021] When the system risk index is not lower than the risk warning threshold but lower than the preset risk critical threshold, the system operation mode is determined to be the cautious balance mode;

[0022] When the system risk index is not lower than the critical risk threshold, the system operation mode is determined to be a high-confidence mode.

[0023] Preferably, the mode decision module is further used to:

[0024] When it is detected that the traffic efficiency cost exceeds the preset cost upper limit for a continuous preset period of time, the system operation mode is forced to be determined as the cautious trade-off mode regardless of the current system risk index value.

[0025] Preferably, when the system operation mode is the optimal efficiency mode, the control execution module is used to:

[0026] Activate the optimal performance decision model preset in the decision model set and execute the control strategy with the goal of minimizing the traffic efficiency cost.

[0027] Preferably, when the system operation mode is the cautious balance mode, the control execution module is used to:

[0028] Activate the decision model with medium complexity and explainability in the preset decision model set, and execute the control strategy with the goal of optimizing the cost of traffic efficiency while taking into account the explainability of the decision.

[0029] Preferably, when the system operation mode is a high reliability mode, the control execution module is used to:

[0030] Activate the preset rule-based simplest decision model in the decision model set to maximize decision explainability, and isolate the determined contaminated data source based on the composition of the system risk index to execute the preset basic assurance control plan.

[0031] The present invention provides an improved intelligent intersection adaptive lighting and traffic signaling system that integrates vehicle-road collaboration and visual perception. Compared with the existing technology, it has the following improvements and advantages:

[0032] 1. The third processing module in this solution enables the system to self-diagnose by continuously calculating the system risk index. In this mode, the control execution module not only activates the simplest decision model to maximize decision explainability, but also proactively isolates visual data sources identified as contaminated, relying solely on V2X data to operate the basic assurance solution. This mechanism eliminates the system from being a passive receiver of information and instead enables it to actively review its own status, effectively defending against data-level attacks or failures and ensuring system robustness.

[0033] 2. This solution uses a mode decision module to switch between three preset operating modes based on the system risk index, achieving dynamic optimization of multiple objectives. This eliminates the need for a static system operating strategy, dynamically coupling it with its own credibility, and presenting the most appropriate system behavior in different scenarios, a feature not available in existing technologies.

[0034] 3. The mode decision module in this solution incorporates a rule that goes beyond conventional logic: when the efficiency cost of traffic consistently exceeds a preset cost cap, the system is forced to switch to cautious trade-off mode, regardless of its current health. This is to address a unique situation where the system could completely freeze traffic while in an extremely safe, high-confidence mode. This mandatory intervention rule ensures the highest priority of maintaining basic access, adds a safeguard against functional paralysis, and significantly improves the solution's practicality and reliability in the real world. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The present invention will be further explained below in conjunction with the accompanying drawings and Examples:

[0036] Figure 1 This is a flowchart of a smart intersection adaptive lighting and traffic signaling system that integrates vehicle-road collaboration and visual perception.

[0037] Figure 2 This is a diagram of the steps for determining the system operation mode in Example 4. DETAILED DESCRIPTION

[0038] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to specific embodiments.

[0039] Example 1:

[0040] See also Figure 1 The present invention provides a smart intersection adaptive lighting and traffic signaling system that integrates vehicle-road collaboration and visual perception, including:

[0041] The data fusion module is used to collect vehicle-road cooperative data and visual perception data, and align and fuse the vehicle-road cooperative data and visual perception data in time and space to generate a unified global traffic status view;

[0042] A first processing module is configured to determine a traffic efficiency cost based on the average vehicle delay time and average queue length included in the unified global traffic status view, combined with a preset reference delay time, reference queue length, and a weight coefficient;

[0043] A second processing module is configured to determine the decision interpretability based on the decision model currently activated by the system and based on the interpretability score preset and bound to the decision model;

[0044] a third processing module for determining a system risk index based on a unified global traffic status view and control instructions output by the system;

[0045] The mode decision module is used to determine the system operation mode based on the system risk index and the preset risk threshold, combined with the traffic efficiency cost and the preset cost upper limit;

[0046] A control execution module, configured to activate a corresponding decision model in response to a system operation mode to adjust the control strategy of traffic signals and lighting;

[0047] A smart intersection adaptive lighting and traffic signaling system that integrates vehicle-road collaboration and visual perception. In this embodiment, it includes several functional modules that work together to achieve traffic control that can adaptively balance multiple objectives based on its own health status and external traffic efficiency.

[0048] The operation of the system starts with the data fusion module; the module is designed to build a comprehensive and redundantly checked panorama of intersection traffic; in this embodiment, the data fusion module is configured to perform multimodal data acquisition through dedicated hardware deployed on the roadside; a roadside unit captures vehicle-road cooperative data that meets the C-V2X standard, and the data stream provides the system with high-frequency dynamic information of the connected vehicle, such as real-time position, velocity vector and acceleration; vehicle-road cooperative data refers to data packets that are actively broadcast outward by the vehicle in accordance with a specific communication protocol. Its function is to provide high-precision vehicle motion status, and its source is the vehicle-mounted communication and positioning unit; the module uses multiple high-definition cameras deployed at key locations at the intersection to collect The visual perception data is collected; the visual data stream is input into an edge computing device, which runs an optimized computer vision algorithm for real-time detection, identification, and continuous tracking of all non-connected vehicles, non-motorized vehicles, pedestrians and other traffic participants within the field of view; the algorithm further extracts macro traffic flow parameters from the visual information, such as the queue length and traffic volume of each lane; the two heterogeneous data streams of collected vehicle-road collaboration data and visual perception data are then subjected to a spatiotemporal alignment process within the module, that is, mapping and matching based on a unified timestamp and coordinate system, and merging into a unified global traffic status view; the view provides a unified and reliable data foundation for all subsequent processing modules;

[0049] To achieve adaptive control of lighting, the data fusion module is also configured to integrate and process the following data related to lighting decisions:

[0050] Ambient lighting data: Lux meters and ambient light sensors deployed at intersections collect real-time data on natural light intensity, providing an accurate basis for the system to determine whether it is daytime, nighttime, or dusk.

[0051] Weather condition data: This can be accessed through third-party weather service APIs or by leveraging existing cameras to identify adverse weather conditions such as rain, snow, and fog that affect visibility through image analysis algorithms.

[0052] Special traffic participant data: Based on the existing visual detection algorithm, it enhances the detection and location tracking of vulnerable traffic participants such as pedestrians and non-motor vehicles;

[0053] These new data streams are aligned with existing traffic data in time and space, integrating them into a more comprehensive global view of traffic and environmental conditions, providing data support for subsequent joint decision-making on adaptive lighting and traffic signaling.

[0054] Based on the global traffic state view, a first processing module is used to determine the traffic efficiency cost. The module aims to quantify the degree of congestion and delay at the intersection into a scalar indicator that can be used for system optimization. In this embodiment, the determination of the traffic efficiency cost follows a mathematical model derived from the service level theory of traffic engineering. To ensure the dimensional consistency of the physical quantities in the model, all inputs are normalized. The traffic efficiency cost is calculated as follows:

[0055]

[0056] in: : Traffic efficiency cost, a dimensionless scalar, which quantifies the traffic efficiency cost that the system needs to pay and is calculated by this module;

[0057] : A preset dimensionless weight coefficient, which sums to 1. Its function is to balance the relative importance of delay and congestion in cost assessment. It is pre-calibrated by the traffic management department according to the control targets at different time periods;

[0058] : Average vehicle delay time, in seconds, which reflects the deterioration of traffic experience. It is derived from the data fusion module in the preset time window. Calculated based on statistics of all vehicle data;

[0059] : Reference delay time, in seconds. It serves as a normalized benchmark for delay time and is based on the 85th percentile value obtained from urban traffic management regulations or statistical analysis of historical data.

[0060] : The average queue length of each entrance lane, in meters, is used to reflect the physical congestion level of the intersection. Its source is the data fusion module in the same time window. The data is obtained through visual analysis or data fusion statistics;

[0061] : Reference queue length, in meters. It serves as a normalized benchmark for queue lengths and is derived from the 85th percentile value based on intersection design specifications or statistical analysis of historical data.

[0062] A second processing module is used to determine decision explainability; the module aims to quantify the transparency and auditability of the current system decision logic and provide a quantitative standard for system compliance; in this embodiment, decision explainability The determination is not done through complex real-time calculations, but is directly bound to the decision model currently running the system; decision interpretability refers to the degree to which the control decisions made by the system can be understood and reviewed by humans. It serves as a quantitative indicator of system credibility, and its source is the preset score bound to the currently activated decision model; there is a set of decision models preset inside the module, in which the models are sorted according to their algorithmic complexity and transparency; when the system switches the decision model, the module immediately outputs the preset interpretability score corresponding to the new model as the decision interpretability indicator of the current system ;

[0063] A third processing module is used to determine the system risk index; the module aims to build an early warning mechanism for quantitative monitoring of the internal state of the system, and proactively discover signs of data pollution attacks or internal decision-making logic confusion; in this embodiment, the determination of the system risk index is based on a comprehensive analysis of the global traffic state view output by the data fusion module and the control instructions output by the subsequent control execution module; the system risk index refers to a comprehensive quantitative assessment of the consistency of data and decision-making stability within the system, and its function is to judge the reliability of the current operation of the system. The source is calculated by this module based on cross-source data conflicts and the volatility of control instruction sequences; the module calculates a dimensionless system risk index by evaluating the consistency between data from different sources and the stability of the system output behavior ;

[0064] A mode decision module as the core decision unit of the system is used to determine the system operation mode based on the system risk index and the preset risk threshold, combined with the traffic efficiency cost and the preset cost upper limit; the module is based on the current internal state of the system. Reflection and external expression, by Reflection, select the most appropriate operation strategy; in this embodiment, the module is mainly based on the system risk index calculated by the third processing module and the traffic efficiency cost calculated by the first processing module Make decisions and Compare with two pre-set risk thresholds and monitor Whether it exceeds the preset upper limit, thereby switching between multiple predefined operating modes;

[0065] A control execution module is used to respond to the system operating mode determined by the mode decision module, activate the decision model corresponding to the mode, and then adjust the traffic signal and lighting control strategy. This module is the ultimate executor of the system strategy. It receives instructions from the mode decision module and loads and activates a specific model from a preset set of decision models. The activated model then generates control instructions based on the real-time traffic view provided by the data fusion module. For example, it can adjust the timing of traffic lights, change the direction of variable lanes, or adjust the intensity and coverage of intersection lighting.

[0066] The system in this embodiment implements a closed-loop, self-perceiving intelligent traffic control system by constructing three core internal state indicators: traffic efficiency cost, decision explainability, and system risk index, and establishing a set of adaptive mode switching and control logic based on these indicators. Compared with traditional systems that only optimize traffic efficiency, this system can monitor its own data quality and decision stability in real time while ensuring smooth traffic flow, and quantify the credibility of its decision-making behavior. When the system risk index decreases, it can proactively downgrade to a safer and more transparent operating mode, effectively resisting the risks brought by data pollution attacks or algorithm failures, thereby greatly improving the robustness, security, and credibility of the entire smart intersection infrastructure.

[0067] This solution establishes a multi-dimensional status assessment system, the core of which lies in three precisely defined quantitative indicators:

[0068] Traffic efficiency cost The construction formula is:

[0069] ,

[0070] in, : Traffic efficiency cost, : weight coefficient, : average delay, : Reference delay, : average queue length, : Reference queue length; In the physical sense, the formula is not a simple sum, but a linear combination of the delay in the time dimension and the congestion in the space dimension in an abstract cost space; through the reference value and The normalization process of two physical quantities with different dimensions is mapped into dimensionless scalars, so that they can be compared and weighed in a unified framework; the cost function The value of represents the macro cost that the transportation system needs to pay to maintain the current operating state and is the primary objective function of system optimization.

[0071] Example 2

[0072] The third processing module determines the system risk index, including:

[0073] Visual perception data and vehicle-road collaboration data are mapped into a visual density matrix and a collaborative density matrix, respectively. The cross-source conflict degree is determined based on the difference between the visual density matrix and the collaborative density matrix. The system risk index is determined by combining the time average of the cross-source conflict degree and the time series stability of the system output control instructions.

[0074] In this embodiment, the method for determining the system risk index of the third processing module is further limited; the module calculates the system risk index through a two-stage process. ;

[0075] The first stage is the calculation of cross-source conflict degree. To achieve the calculation, the physical space of the intersection is virtualized into a grid; at any moment , the data in the global traffic state view provided by the data fusion module are processed separately; the visual perception data are mapped into a visual density matrix , whose elements Representative Grid At the moment At the same time, the vehicle-road collaboration data is also processed and mapped into a collaboration density matrix with the same dimension ,element represents the vehicle density calculated from V2X data at the same time for the same grid. Both matrices are normalized to eliminate dimensionality differences.

[0076] To enable those skilled in the art to implement, a specific, non-restrictive implementation of the mapping process is provided:

[0077] For the visual density matrix Generation: For the HD camera video stream at the intersection, a pre-trained target detection deep learning model, such as YOLOv5, SSD or FasterR-CNN model, is used to identify vehicles, non-motor vehicles and other traffic participants in the field of view in real time, and obtain their bounding boxes in the image coordinate system. Through the pre-calibrated camera parameters, a homography transformation matrix is ​​calculated. The matrix can map the pixel coordinates on the image plane to the virtualized grid coordinates of the physical space of the intersection. For each detected target, the center point coordinates of its bounding box are mapped to In the grid; at the moment , grid The corresponding matrix elements The value of , that is, the number of target center points determined to fall within the grid;

[0078] For the synergy density matrix Generation: The processing is relatively straightforward, at the same time , collects the location information broadcast by all connected vehicles through C-V2X communication, usually high-precision GPS coordinates; converts geographic coordinates into the same local spatial coordinate system used by the virtualized grid, and Broadcasts its own position and its coordinates fall into the grid The connected vehicles in the network are counted, and the count value is determined as the collaborative density matrix in The value of the position ;

[0079] Through the above steps, the heterogeneous raw data can be converted into a density matrix with the same dimension and unified physical meaning, which is convenient for the subsequent calculation of cross-source conflict degree. Lay the foundation;

[0080] Based on the cross-source conflict degree at the moment is determined as the root mean square error between these two matrices:

[0081]

[0082] in: :time The cross-source conflict degree is a dimensionless scalar that quantifies the degree of inconsistency between data from two different sources and is calculated by this module.

[0083] : The dimension of the virtualized grid is a positive integer. Its function is to define the granularity of spatial discretization, which is derived from the system preset; : Row index of the virtualized grid; : column index of the virtualized grid; :time;

[0084] In practice, the choice of granularity aims to balance spatial resolution and computational overhead. A preferred setting is to set the physical size of each grid cell to be comparable to the width of a typical lane, for example, a 3m x 3m square grid. This setting ensures that each grid cell roughly corresponds to the space of a parking space, helping to accurately reflect the queuing and distribution of vehicles while avoiding the unnecessary computational burden caused by overly fine grids.

[0085] : Visual density matrix in The value of position is a dimensionless real number, which comes from the processed visual perception data;

[0086] : The collaborative density matrix is The value of position is a dimensionless real number, which comes from the processed vehicle-road cooperative data;

[0087] The second stage is the synthesis of the final system risk index; the module calculates the Within, the above cross-source conflict Time average ; At the same time, it also analyzes the control instruction vector output by the control execution module In the same time window Calculate the mean of the time series within and variance To ensure dimensional consistency, the indicator used is the square of the coefficient of variation of the control instruction sequence , which is a dimensionless value; system risk index It is determined by taking the weighted sum of the time average of the cross-source conflict degree and the time series stability of the system output control instructions:

[0088]

[0089] in: :time The system risk index, a dimensionless scalar, is finally calculated by this module;

[0090] : A preset dimensionless weight, which sums to 1. Its role is to balance the relative impact of data-level conflicts and system behavior-level instability. Its source is preset by the system designer;

[0091] : In the time window The time average value within a period of time is used to smooth instantaneous data jitter and reflect the average data quality over a period of time. It comes from the first stage calculation of this module.

[0092] :Control instruction sequence in time window The variance within the system is used to quantify the volatility or uncertainty of the system output, which comes from the control execution module;

[0093] :Control instruction sequence in time window The mean value within is used as the benchmark for variance normalization and its source is the control execution module;

[0094] : A very small positive constant, such as , which is used to prevent the The denominator is zero due to its tendency to approach zero, which ensures the mathematical robustness of the calculation. This is due to the system default.

[0095] It should be noted that the systemic risk index It is an indicator that quantifies system risk. The higher the value, the higher the degree of data conflict or decision instability, that is, the worse the system risk index;

[0096] Through this specific way of determining health, the system can not only Indicators identify conflicts across data sources caused by sensor failures, malicious attacks, or severe weather, and can also be used to The indicator monitors potential oscillations or confusion in the system's own decision-making logic. This integrated monitoring mechanism makes the assessment of the system risk index more comprehensive and profound, enabling earlier and more accurate early warning of potential risks, thus providing a more reliable basis for subsequent adaptive mode switching decisions.

[0097] System risk index of the present invention The definition of is calculated as follows:

[0098] ,

[0099] in, : Systematic risk index, : weight coefficient, : average cross-source conflict degree, : Control instruction variance, : control instruction mean, : Stability constant; the formula has profound physical connotations; the first term It comes from the calculation of the root mean square error between two independent observation channels, namely visual perception and vehicle-road cooperative data matrix, which quantifies the information entropy or uncertainty of the information received by the system from the outside world; a high The value of , means that there is a significant conflict between the physical reality described by the two channels, suggesting data contamination or sensor failure; the second is the square of the coefficient of variation of the control instruction sequence, which quantifies the inherent entropy or chaos of the system's own output behavior; a high value of this term means that the system decision logic may fall into an oscillatory or unstable state; therefore, As a comprehensive indicator, it measures the overall stability and reliability of the system from both the input and output levels. The formula is self-consistent, and its components are calculated from the system's previous modules and can be logically derived completely.

[0100] The present invention realizes a fundamental transformation from passive response to active defense. The existing technology usually responds passively to the changes in traffic flow, and its operation premise is that all input data are true and valid. The third processing module in this solution calculates the system risk index. The continuous calculation of the system enables the system to have the ability of self-diagnosis; for example, when the camera in an area is maliciously blocked, resulting in a huge deviation between the visual density matrix and the V2X collaborative density matrix, the cross-source conflict degree will rise sharply, pushing up ; The model decision module will refer to the preset risk threshold based on the changes , forcing the system to switch to high confidence mode; in this mode, the control execution module not only activates the simplest decision model to maximize decision interpretability, but also The composition of The contribution of this project is to actively isolate visual data sources that are judged to be contaminated, and rely only on V2X data to run the basic security solution. This mechanism makes the system no longer a passive receiver of information, but an active reviewer of its own status, which can effectively resist attacks or failures at the data level and ensure the robustness of the system.

[0101] Example 3

[0102] The second processing module determines the decision explainability, including:

[0103] A set of preset decision models, where each model in the set is arranged in ascending order of complexity and is associated with a monotonically decreasing preset interpretability score;

[0104] The decision interpretability is determined as a preset interpretability score corresponding to the decision model currently activated by the control execution module;

[0105] This embodiment further defines the method for determining decision interpretability of the second processing module in the system of Example 1; the core of the module lies in a preset, structured mapping relationship rather than complex dynamic calculations;

[0106] In order to clarify its internal logic, the system presets a set of decision models during the design phase. The decision model set refers to a library that contains all the alternative algorithm models that may be activated by the system in different modes. Its function is to provide the system with multi-level decision-making capabilities. The source is pre-developed and integrated by system developers according to application requirements. The models in this set are arranged in ascending order according to their intrinsic algorithm complexity. For example, model It is a deterministic model based on simple if-then rules, with clear decision paths and easy review; with index As the size of the model increases, the complexity of the model increases until the model , which may be a complex model based on deep reinforcement learning. Although it has the best performance in optimizing traffic efficiency, its decision-making process is opaque and difficult to be directly explained;

[0107] Corresponding to the sorting, the system is for each decision model in the set Each static binding has a preset interpretability score The preset explainability score is a dimensionless value jointly determined by domain experts and legal compliance experts based on the algorithmic principles, auditability and transparency of each model. It serves to provide a quantitative benchmark for the credibility of each decision logic. The score is set to be monotonically decreasing, that is, to meet the requirements of ; For example, a completely transparent rule model Score Can be set to 0.9, and the most complex deep learning model Score It may be set to 0.2;

[0108] Based on the above assumptions, during the operation of the system, decision explainability Directly determined as the decision model currently activated by the control execution module The corresponding preset explainability score ; When the mode decision module instructs to switch to the model When the second processing module does not need to perform any calculations, it only needs to query and output Binding score As the current That's it;

[0109] This static binding of explainability to specific decision models provides a deterministic, auditable, and quantitative standard. It avoids the complex and potentially controversial dynamic assessment of the abstract concept of explainability. It ensures that the system's credibility at any given moment is a clear, pre-defined, and unalterable attribute, greatly enhancing the system's transparency and compliance with regulatory scrutiny. Operators and managers can clearly understand the degree of transparent logic used in decisions made under certain modes.

[0110] Explainability of decision-making Through the decision model Static binding score To determine, that is ,in, : Decision explainability, :Model The preset score of : The index of the decision model represents the position of the model in the set arranged in ascending order of complexity. The principle of this design is that the interpretability of an algorithm is its intrinsic property, determined by its mathematical structure, rather than a dynamically changing quantity. By pre-calibrating the indicators by domain experts according to the algorithmic principles of each model, the indicators can be quantified, providing a quantitative basis at the legal and ethical levels for subsequent decision-making.

[0111] Example 4

[0112] See Figure 2 As shown, the mode decision module determines the system operation mode, including:

[0113] When the system risk index is lower than the preset risk warning threshold, the system operation mode is determined to be the optimal efficiency mode;

[0114] When the system risk index is not lower than the risk warning threshold but lower than the preset risk critical threshold, the system operation mode is determined to be the cautious balance mode;

[0115] When the system risk index is not lower than the critical risk threshold, the system operation mode is determined to be a high-confidence mode;

[0116] In this embodiment, the core logic used by the mode decision module in the system to determine the system operation mode is further limited; the logic is based on the system risk index Comparison of two key thresholds;

[0117] The system presets two risk thresholds: a risk warning threshold and a critical risk threshold The risk threshold is the critical value used to classify the system health status level. It serves as the trigger condition for state transition and is set by statistically analyzing a large amount of historical or simulation data to identify typical system state turning points.

[0118] A feasible and specific setting method is as follows: Under the condition of normal system operation, collect a large number of system risk index data, for example, for several consecutive weeks. The data is used to form a benchmark data set, and the data set is statistically analyzed to determine the risk warning threshold. It can be set as the 95th percentile of the data distribution, representing the statistical boundary of slight abnormalities in the system. By injecting fault simulation or finding known sensor failures and data attack events in historical data, the system can obtain a clear risk critical threshold. can be set to the observed value under all simulated or historical fault conditions. A higher percentile value in the value distribution, such as the 90th percentile value, or a higher value set by expert judgment that can stably distinguish between normal conditions and serious fault conditions to ensure that serious system health problems can be reliably detected and ensure that Significantly higher than ;

[0119] Among them, the critical threshold is higher than the warning threshold, that is, The mode decision module continuously receives the real-time system risk index calculated by the third processing module , and execute the following three-branch judgment logic:

[0120] When the system risk index Below the preset risk warning threshold hour, , the system operation mode is determined to be the optimal efficiency mode; this situation indicates that the system data quality is high, the internal decision-making is stable, and it is in the healthiest operating state;

[0121] When the system risk index Not less than the risk warning threshold But below the preset risk threshold hour, , the system operation mode is determined to be cautious balance mode; this situation corresponds to a certain degree of deterioration in the system risk index, and data or decision stability begins to show signs of risk, but has not yet reached a serious level;

[0122] When the system risk index Not less than the critical risk threshold hour, , the system operation mode is determined to be high-confidence mode; this situation indicates that the system risk index has seriously deteriorated, and there is a high risk of data contamination or internal logic confusion;

[0123] This clear set of threshold-based state transition rules makes the system's mode switching behavior highly predictable and stable. It ensures that the system's response to risk is graded and gradual, rather than oscillating randomly between different strategies. This structured decision-making logic enables the system to increase its operational conservatism and transparency proportionally to the deterioration of its health, thereby achieving a dynamic and reasonable balance between risk management and performance pursuit.

[0124] The present invention builds a multi-objective dynamic trade-off framework of efficiency, safety and transparency. The traditional traffic signal system takes optimizing traffic efficiency as its single goal, and its decision-making process is often an opaque black box. This solution uses the mode decision module to make decisions based on the system risk index. Switching between three preset operating modes enables dynamic optimization of multiple objectives; When the risk warning threshold is reached, the system is in the optimal efficiency mode, fully optimizing the efficiency cost of traffic. ;exist When the risk threshold is reached, the system enters a cautious trade-off mode and optimizes At the same time, it also takes into account higher decision explainability ; and in When This design makes the system's operating strategy no longer static, but dynamically coupled with its own credibility, presenting the most appropriate system characteristics in different scenarios, which is not available in existing technologies.

[0125] Example 5

[0126] The mode decision module is also used to:

[0127] When it is detected that the traffic efficiency cost exceeds the preset cost upper limit for a continuous preset period of time, the system operation mode is forced to be determined as the cautious trade-off mode regardless of the current system risk index value;

[0128] In this embodiment, a mandatory intervention rule is added to the mode decision module based on the mode decision logic; the rule is designed to deal with extreme traffic congestion and ensure the ultimate bottom-line function of the system;

[0129] In addition to the systemic risk index In addition to making mode decisions, the mode decision module also monitors the traffic efficiency cost calculated by the first processing module in parallel. ; The system has a preset cost cap and a duration The cost cap refers to the maximum acceptable level of congestion delay, which defines the bottom line of the traffic service level and can be set by the traffic management department. The duration refers to the time window for determining whether the congestion state is solidified, which filters out instantaneous traffic fluctuations and can be set based on statistical analysis of the time when historical traffic congestion occurred.

[0130] The enhanced logic of the module is implemented as follows: when the traffic efficiency cost is detected Consistently exceeding pre-set cost caps Reaching the preset time Regardless of the current system risk index For any value, the system operation mode will be forced to be determined as a cautious trade-off mode;

[0131] Rules have the highest priority; for example, even if the system risk index Very high, In theory, it should enter the high confidence mode, but if there is serious congestion at the intersection, continued For a long time, the rules will override the rules based on health, forcing the system to switch to a cautious trade-off mode; set up control strategies designed to break the traffic lock state caused by the overly conservative high-confidence mode, for example, using fixed-cycle signal timing; once the congestion is relieved, Fall back to Below, the forced intervention is lifted and the system will return to its original state based on health Three-branch decision logic;

[0132] In addition, this rule also provides the necessary redundancy guarantee for the optimal efficiency mode; high-performance decision-making models under this mode, such as deep learning models, although they perform well under normal conditions, may fail to make decisions or become unstable when encountering extreme traffic events that their training data does not cover, which in turn aggravates congestion; therefore, when the traffic efficiency cost is high, Continuously exceeding the upper limit also indicates that the model is no longer competent for the current situation. In this case, forcibly switching to a cautious trade-off mode with a simpler algorithm and more robust behavior is a rational degradation strategy designed to ensure that the intersection has basic traffic relief capabilities under all circumstances and avoid functional paralysis of the system due to model failure.

[0133] This enhanced rule adds a crucial congestion relief mechanism to the system, prioritizing the protection of citizens' basic right of passage and the maintenance of basic connectivity of the urban transportation network in extreme situations. This circuit breaker effectively prevents the system from becoming functionally paralyzed due to excessive pursuit of a single metric, such as absolute safety or explainability. This ensures the system maintains the most basic ability to alleviate traffic congestion in all scenarios, greatly enhancing its real-world applicability and robustness.

[0134] The present invention introduces a mandatory intervention mechanism to ensure the basic functions of the system, solving the problem of system failure under extreme congestion. In some cases, the existing technology may aggravate congestion due to the conservatism or misjudgment of the algorithm. The mode decision module in this solution has a rule that goes beyond conventional logic: when the traffic efficiency cost is Consistently exceeding pre-set cost caps , regardless of the current health value, it is forced to switch to cautious trade-off mode. This is to deal with a special situation: the system may be in an extremely safe high-confidence mode, such as using inefficient fixed timing, which may cause traffic to be completely locked. The forced intervention rule ensures the realization of the highest priority goal of maintaining basic access, adding an insurance against functional paralysis to the system, greatly improving the practicality and reliability of the solution in the real world.

[0135] Example 6

[0136] When the system operation mode is the optimal efficiency mode, the control execution module is used to:

[0137] Activate the optimal performance decision model preset in the decision model set and execute the control strategy with the goal of minimizing the cost of traffic efficiency;

[0138] When the system operation mode is determined to be the optimal efficiency mode, the control execution module activates the optimal performance decision model ; The objective function of this model is not only to minimize the cost of traffic efficiency , and also expanded into a multi-objective optimization function that includes lighting energy consumption, achieving refined adaptive control of lighting while ensuring traffic efficiency;

[0139] The lighting control strategy at this time is as follows:

[0140] On-demand zoned lighting: Dynamically adjust intersection lighting based on the locations and movement of vehicles, pedestrians, and non-motorized vehicles tracked in real time from the global traffic and environmental status view. For example, only illuminate the area where vehicles or pedestrians are about to arrive, turn on lighting on the guidance path in advance based on their speed, and maintain lighting at a lower energy-saving level in areas without traffic activity.

[0141] Continuous Brightness Adjustment: Based on data from the ambient light sensor, the system smoothly and continuously reverses the lighting intensity. At dusk or dawn, the system gradually increases or decreases the output of artificial lighting to ensure that the total road illumination is always within the optimal safety threshold, preventing sudden changes in brightness from affecting the driver's visual adaptation.

[0142] This mode enables the system to fully utilize its most advanced perception and algorithm capabilities in a healthy state to achieve dual optimization of traffic efficiency and energy efficiency;

[0143] When the system operation mode is the cautious trade-off mode, the control execution module is used to:

[0144] Activate the decision model with the best complexity and explainability among the preset decision model set, and execute the control strategy with the goal of optimizing the cost of traffic efficiency while taking into account the explainability of the decision;

[0145] When the system operation mode is switched to the cautious trade-off mode, the control execution module activates a decision model with moderate complexity and interpretability. ; Its control goal is to optimize the cost of traffic efficiency At the same time, take into account and improve decision explainability The lighting control strategy is now simplified accordingly, enabling adaptation in a more understandable and auditable manner:

[0146] Tiered pre-set lighting: Instead of continuous adjustment, the system selects one of a set of pre-set lighting profiles, such as energy-saving mode, standard mode, and enhanced mode, based on ambient light and overall traffic flow. For example, if it detects nighttime and low traffic flow, it switches to energy-saving mode; if pedestrians or inclement weather are detected, it forces a switch to standard or enhanced mode.

[0147] Linked signal lighting: Lighting strategies are clearly linked to traffic signals; for example, when the pedestrian crossing signal is green, the lighting intensity of the crosswalk and waiting area is proactively increased to enhance pedestrian protection;

[0148] This mode provides a graceful performance degradation mechanism. When the system faces moderate risk, it switches to a carefully selected intermediate state, improving decision safety and reliability while still maintaining a considerable level of adaptive traffic and lighting optimization capabilities.

[0149] When the system is in high-reliability mode, the control execution module is used to:

[0150] Activate the default rule-based simplest decision model in the decision model set to maximize decision explainability, and isolate the identified contaminated data sources based on the composition of the system risk index to execute the default basic assurance control plan;

[0151] When the system operation mode is forced to enter the high-confidence mode, the control execution module executes a set of dimensionality reduction operation strategies with absolute security and transparency as the primary principle; it activates the preset rule-based simplest decision model in the decision model set. The model is based entirely on deterministic rules, and its behavior can be fully predicted and audited, making decisions explainable. Reaching the maximum value ; The control execution module will be based on the system risk index The specific composition of the system is to actively isolate the contaminated data source; for example, if the analysis shows that The increase is mainly due to the cross-source conflict In this mode, the system temporarily ignores all V2X data inputs and relies solely on visual perception data for decision-making. In this mode, traffic signaling control will fall back to a pre-set basic guarantee control scheme, such as a fixed-cycle timing scheme that ensures safe passage under all circumstances. Lighting control will implement the following most basic and reliable adaptive rules:

[0152] Binary switching based on reliable sensing: Lighting control relies solely on the most reliable sensor—the ambient light meter. If and only if the illuminance falls below a preset safety threshold, such as during nighttime, the system turns on all intersection lighting to 100% of the maximum design brightness. If the illuminance rises above the threshold, the lighting is turned off. This is a deterministic, fully predictable, and auditable binary adaptive logic.

[0153] Emergency warning function: In this mode, if the system risk index If the composition of the lights indicates a serious cross-source conflict, which may mean an accident or serious failure, the system can superimpose a warning flashing function in a specific area on the above basic lighting to alert passing traffic participants to the danger;

[0154] This model provides the system with the ultimate baseline safe operating state. When facing serious risks, the system decisively abandons the pursuit of efficiency and instead ensures the absolute transparency, auditability and security of the decision-making process. By activating the simplest model and executing the most basic and reliable adaptive lighting rules, the model effectively blocks the further transmission of risks, ensures the most basic safety order at the intersection, and buys time and space for subsequent manual intervention or system repairs, reflecting the highest safety standards in the design of critical infrastructure.

[0155] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, and all of these should be included in the scope of protection of the present invention.

Claims

1. A smart intersection adaptive lighting and traffic signaling system that integrates vehicle-road collaboration and visual perception, characterized by: include: The data fusion module is used to collect vehicle-road cooperative data and visual perception data, and align and fuse the vehicle-road cooperative data and visual perception data in time and space to generate a unified global traffic status view; A first processing module is configured to determine a traffic efficiency cost based on the average vehicle delay time and average queue length included in the unified global traffic status view, combined with a preset reference delay time, reference queue length, and a weight coefficient; A second processing module is configured to determine the decision interpretability based on the decision model currently activated by the system and based on the interpretability score preset and bound to the decision model; a third processing module for determining a system risk index based on a unified global traffic status view and control instructions output by the system; The mode decision module is used to determine the system operation mode based on the system risk index and the preset risk threshold, combined with the traffic efficiency cost and the preset cost upper limit; A control execution module, configured to activate a corresponding decision model in response to a system operation mode to adjust the control strategy of traffic signals and lighting; The third processing module determines the system risk index, including: Map the visual perception data and vehicle-road collaboration data into a visual density matrix and a collaboration density matrix respectively; Based on the difference between the visual density matrix and the collaborative density matrix, the cross-source conflict degree is determined; and the system risk index is determined by combining the time average value of the cross-source conflict degree with the time series stability of the system output control instructions.

2. The intelligent intersection adaptive lighting and traffic signaling system integrating vehicle-road collaboration and visual perception according to claim 1 is characterized in that: The second processing module determines decision explainability, including: A set of preset decision models, where each model in the set is arranged in ascending order of complexity and is associated with a monotonically decreasing preset interpretability score; The decision interpretability is determined as a preset interpretability score corresponding to the decision model currently activated by the control execution module.

3. The intelligent intersection adaptive lighting and traffic signaling system integrating vehicle-road collaboration and visual perception according to claim 1 is characterized in that: The mode decision module determines the system operation mode, including: When the system risk index is lower than the preset risk warning threshold, the system operation mode is determined to be the optimal efficiency mode; When the system risk index is not lower than the risk warning threshold but lower than the preset risk critical threshold, the system operation mode is determined to be the cautious balance mode; When the system risk index is not lower than the critical risk threshold, the system operation mode is determined to be a high-confidence mode.

4. The intelligent intersection adaptive lighting and traffic signaling system integrating vehicle-road collaboration and visual perception according to claim 3 is characterized in that: The mode decision module is further configured to: When it is detected that the traffic efficiency cost exceeds the preset cost upper limit for a continuous preset period of time, the system operation mode is forced to be determined as the cautious trade-off mode regardless of the current system risk index value.

5. The intelligent intersection adaptive lighting and traffic signaling system integrating vehicle-road collaboration and visual perception according to claim 3 is characterized in that: When the system operation mode is the optimal efficiency mode, the control execution module is used to: Activate the optimal performance decision model preset in the decision model set and execute the control strategy with the goal of minimizing the traffic efficiency cost.

6. The intelligent intersection adaptive lighting and traffic signaling system integrating vehicle-road collaboration and visual perception according to claim 3 is characterized in that: When the system operation mode is the cautious balance mode, the control execution module is used to: Activate the decision model with medium complexity and explainability in the preset decision model set, and execute the control strategy with the goal of optimizing the cost of traffic efficiency while taking into account the explainability of the decision.

7. The intelligent intersection adaptive lighting and traffic signaling system integrating vehicle-road collaboration and visual perception according to claim 3 is characterized in that: When the system operation mode is the high-reliability mode, the control execution module is used to: Activate the preset rule-based simplest decision model in the decision model set to maximize decision explainability, and isolate the determined contaminated data source based on the composition of the system risk index to execute the preset basic assurance control plan.

Images