Method and system for car-following optimization of narrow road network based on unmanned aerial vehicle building sheltering quantification

Abstract

The present application proposes a narrow road network vehicle following optimization method and system based on unmanned aerial vehicle building sheltering quantification, belonging to the field of intelligent transportation and vehicle-road cooperative safety control technology, the steps are: building an unmanned aerial vehicle global cruise and low-altitude follow-up dual-mode collection mechanism, collecting multi-dimensional perception data and completing preprocessing; calculating the sheltered visibility reduction and the comprehensive sheltering coefficient, correcting the effective visibility, the passing width and the braking response time; building a multi-objective following optimization model, taking the guided driving speed and the guided following distance as the decision variables; using the improved NSGA-III algorithm to solve the optimal following parameters, guiding the solution set to converge to high safety margin through the adaptive reference point and survival selection mechanism, and issuing instructions through 5G-V2X after verification; establishing a shelter risk index driven hierarchical closed-loop re-optimization mechanism to dynamically update the following parameters. The present application solves the problems of multiple perception blind areas and poor following control adaptability in the building sheltering scene, effectively improving the safety and passing efficiency of narrow road network following.

Description

Technical Field

[0001] This invention belongs to the field of intelligent transportation and vehicle-road cooperative safety control technology, specifically relating to a method and system for optimizing vehicle following in narrow road networks based on the quantification of building occlusion by unmanned aerial vehicles. Background Technology

[0002] As the density of urban built-up areas continues to increase, narrow road networks such as branch roads in old urban areas, alleys, rural narrow roads, and roads within scenic areas generally suffer from dense buildings on both sides and limited roadside space. Building walls and cantilevered structures significantly reduce drivers' visibility and encroach on effective passage width, resulting in delayed perception of forward risks and insufficient lateral safety margins when following other vehicles. These areas are high-incidence areas for rear-end collisions, side scrapes, and other traffic accidents. How to achieve refined vehicle following optimization in scenarios where buildings obstruct traffic has become a key and challenging aspect of urban road traffic safety management.

[0003] Currently, traffic safety management in narrow road networks largely relies on fixed checkpoint monitoring and static speed limit signs. Following-the-road optimization parameters are set based on open road conditions, failing to incorporate the dynamic impact of building obstructions into the optimization model. This leads to a mismatch between the optimization threshold and actual driving risks, resulting in both insufficient safety redundancy and wasted traffic efficiency. Some vehicle-road cooperative solutions rely on roadside sensing devices to collect traffic data; however, building obstructions in narrow road networks also limit the detection range of roadside devices, creating numerous blind spots and making it impossible to fully grasp the distribution of obstructions and the vehicle's operating status.

[0004] In recent years, drones, with their advantages of maneuverability and global vision, have been gradually applied to low-altitude collaborative sensing and emergency management scenarios. However, existing applications mainly focus on illegal violation detection, macro-level congestion monitoring, and accident investigation, and have not yet formed a refined car-following optimization system for building-obstructed sections, encompassing "obstruction quantification, risk assessment, dynamic guidance, and closed-loop optimization." On the one hand, the lack of multi-dimensional building obstruction quantification methods and unified classification standards makes it impossible to transform the impact of obstruction on line of sight, passage space, and driver response characteristics into calculable optimization parameters. On the other hand, existing car-following optimization algorithms do not couple dynamic corrections to the impact of obstruction, making it difficult to adapt to the spatiotemporal changes in the degree of obstruction in narrow road networks and failing to achieve differentiated safety guidance based on different levels.

[0005] In summary, existing technologies have not yet developed a vehicle following optimization technology solution for vehicle-road cooperative safety control that is adapted to narrow road network building occlusion scenarios. They cannot solve the pain points of multiple perception blind spots, difficulty in risk quantification, and low optimization accuracy under building occlusion. There is an urgent need to propose a vehicle following optimization method based on building occlusion quantification based on full-domain perception to provide technical support for traffic safety management in narrow road networks. Summary of the Invention

[0006] The purpose of this invention is to propose a vehicle following optimization method and system for narrow road networks based on UAV building occlusion quantification. The method achieves full-domain perception of occlusion through dual-mode UAV data acquisition, quantifies the multi-dimensional impact of building occlusion and corrects optimization parameters, constructs a following safety constraint system coupled with occlusion impact, uses an improved NSGA-Ⅲ algorithm to solve for the optimal guidance parameters, and establishes a hierarchical closed-loop re-optimization mechanism to dynamically adjust the guidance strategy, effectively improving the following safety and traffic efficiency in building occlusion sections of narrow road networks.

[0007] To achieve the above objectives, the technical solution of the present invention is as follows:

[0008] A vehicle following optimization method for narrow road networks based on UAV building occlusion quantification includes the following steps:

[0009] S1. Based on the dual-mode collaborative acquisition mechanism of global cruise and low-altitude follow-fly of UAVs, multi-dimensional perception data is collected in real time in narrow road network sections. After preprocessing, standardized scene data is obtained, including target vehicle information, global traffic status information, road environment and building occlusion parameters, and vehicle interaction data in narrow road network sections.

[0010] S2. Calculate the reduction in sight distance due to occlusion and the comprehensive building occlusion impact coefficient based on standardized scene data; and correct the effective sight distance, effective passage width and braking response time based on the reduction in sight distance due to occlusion and the comprehensive building occlusion impact coefficient; if the corrected effective sight distance is lower than the preset minimum safe sight distance, or the corrected effective passage width is lower than the preset minimum safe passage width, or the lateral safety margin is lower than the preset threshold, then terminate the subsequent optimization steps and trigger the safety response command; otherwise, execute step S3.

[0011] S3. Construct a multi-objective car-following optimization model, using guided driving speed and guided following distance as decision variables; wherein, the guided driving speed is the candidate driving speed to be recommended to the target vehicle, and the guided following distance is the candidate following distance to be recommended to the target vehicle; the multi-objective car-following optimization model includes multi-dimensional car-following safety constraints and a three-dimensional target vector;

[0012] The multi-dimensional car-following safety constraints are constructed based on the comprehensive building obstruction influence coefficient, the corrected braking response time, and the effective passage width, including road condition constraints, vehicle dynamics constraints, and safety interaction constraints.

[0013] The three-dimensional target vector consists of three sub-objectives: lowest energy consumption, highest traffic efficiency, and optimal safe distance, and each objective function is coupled with a comprehensive building occlusion influence coefficient.

[0014] S4. The improved NSGA-Ⅲ algorithm is used to solve the multi-objective car-following optimization model to obtain the Pareto optimal solution set. Based on the comprehensive evaluation function, the optimal guidance parameter set is selected from the Pareto optimal solution set. After validity verification, the parameter set is sent to the vehicle guidance device of the target vehicle through the vehicle-road cooperative network to push safe speed and following distance prompts to the driver.

[0015] S5. Establish a hierarchical closed-loop re-optimization mechanism based on the occlusion risk index: Calculate the normalized occlusion risk index to classify risk levels. When the preset hierarchical triggering conditions are met, trigger the re-optimization process of the corresponding priority. And periodically adjust the weight of the comprehensive objective function according to the spacing execution error.

[0016] Preferably, the specific contents of the target vehicle information, global traffic status information, road environment and building occlusion parameters, and vehicle interaction data in the narrow road network section are as follows:

[0017] The target vehicle information includes the following parameters of the guided target vehicle traveling in a narrow road network section with obstructing buildings: the vehicle's measured speed. (Low-altitude follow-flying drones generate measured speed sequences through continuous frame target localization), vehicle width Vehicle braking performance parameters, vehicle design maximum speed Rated maximum acceleration and rated maximum deceleration The vehicle braking performance parameters include the standard driver braking response time. and the vehicle's maximum braking deceleration ;

[0018] The global traffic status information includes the following parameters for narrow road network sections: total length of the narrow road network section. Traffic flow Vehicle density Maximum traffic density of the section Average driving speed Traffic congestion index Length of continuous obstruction The target vehicle speed obtained from coarse-grained identification The longitudinal distance between the target vehicle and the vehicle in front obtained by coarse-grained identification and the benchmark theoretical safe following distance ;

[0019] The road environment and building obstruction parameters include the following parameters for narrow road network sections: theoretical lane width of the section. Road curvature Road slope Road surface adhesion coefficient Static obstacle position coordinates Minimum speed limit for the section Maximum speed limit for the section The shortest line of sight in the section theory Coordinates of the building's obstruction location Building height Effective lateral width of building obstruction Area of ​​obstructed view Total monitoring field of view area Angle of view from the driver ;

[0020] The vehicle interaction data includes the following parameters in narrow road network sections: real-time longitudinal distance between the target vehicle and the vehicle in front. Speed ​​of the vehicle in front Lateral center distance between the target vehicle and vehicles in adjacent lanes Adjacent lane vehicle width Probability of interaction conflict with the current vehicle The It is calculated based on the relative position, relative speed and preset prediction time window of the target vehicle with the vehicle in front and the vehicles in adjacent lanes;

[0021] The target vehicle information, global traffic status information, road environment and building occlusion parameters, and vehicle interaction data of the narrow road network section are standardized scene data after spatiotemporal coordinate calibration, noise reduction (moving average filtering), and outlier removal.

[0022] Preferably, the dual-mode collaborative data acquisition mechanism of global drone cruise and low-altitude follow-fly is as follows:

[0023] In global cruise mode, the drone covers narrow road network sections along a preset flight path, completing the collection of global traffic status information and road environment and building occlusion parameters; the preset flight path is set according to the effective monitoring area of ​​the drone, so that the coverage of the narrow road network section by the preset flight path is not less than a preset threshold.

[0024] In low-altitude follow-fly mode, the drone maintains synchronous flight with the guided target vehicle, completing the collection of target vehicle information and vehicle interaction data; wherein, the relative position control error during follow-flying satisfies:

[0025]

[0026] In the formula, This refers to the error between the actual longitudinal position of the drone and its preset longitudinal tracking position relative to the target vehicle. This represents the error between the actual lateral position of the drone and the preset lateral following position. This represents the error between the actual flight altitude of the drone and the preset follow-up altitude. The preset maximum follow-fly position error threshold is used;

[0027] In the global patrol mode, the drone prioritizes the detection of target vehicles. Intelligent switching:

[0028] In the formula, For preset coefficients, This is a correction factor for the following distance risk item, and ; The target vehicle speed is obtained through coarse-grained identification in global cruise mode. This refers to the longitudinal distance between the target vehicle and the vehicle in front, obtained through coarse-grained identification in global cruise mode. This is the maximum speed limit for this section. The baseline theoretical safe following distance;

[0029] when When the preset switching threshold is reached, the drone switches to the low-altitude follow-fly mode corresponding to the target vehicle;

[0030] In low-altitude follow-fly mode, the actual vehicle speed is used. Real-time longitudinal distance between the target vehicle and the vehicle in front Replace the coarse-grained recognition and The drone recalculates the target vehicle monitoring priority; when the target vehicle leaves the obstructed area, or when the recalculated target vehicle monitoring priority is continuously lower than the preset exit threshold, the drone returns to global cruise mode.

[0031] Preferably, the calculation of the occlusion sight distance reduction and the comprehensive building occlusion impact coefficient based on standardized scene data is as follows:

[0032] Calculate the reduction in sight distance due to occlusion :

[0033]

[0034] In the formula, This is the shortest sight distance in the section theory. For building height, The driver's line of sight angle;

[0035] Calculate the line-of-sight occlusion coefficient :

[0036]

[0037] Calculate the lateral occlusion coefficient :

[0038]

[0039] In the formula, The effective lateral width for building shading. This refers to the theoretical lane width of the section.

[0040] Calculate the longitudinal occlusion coefficient :

[0041]

[0042] In the formula, To cover the continuous length of the road section, This refers to the total length of the narrow road network section;

[0043] Calculate the area shading coefficient :

[0044]

[0045] In the formula, To block the area of ​​vision, Total monitored field of view area;

[0046] Calculate the initial value of the comprehensive building shading impact coefficient :

[0047]

[0048] In the formula, These are the fusion weights for the line-of-sight occlusion coefficient, lateral occlusion coefficient, longitudinal occlusion coefficient, and area occlusion coefficient, respectively. ;

[0049] The initial value of the comprehensive building shading impact coefficient is normalized and truncated to obtain the comprehensive building shading impact coefficient. .

[0050] Preferably, the correction of effective sight distance, effective passage width, and braking response time based on the reduction in sight distance due to occlusion and the comprehensive building occlusion impact coefficient is as follows:

[0051] Based on the comprehensive building shading impact coefficient Corrected braking response time:

[0052] In the formula, This is the standard driver braking response time when there is no obstruction. The corrected braking response time, The gain coefficient representing the effect of building obstruction on driver braking response time;

[0053] Based on the reduction in viewing distance due to occlusion Corrected effective line of sight:

[0054]

[0055] In the formula, This is the shortest sight distance in the section theory. The traffic congestion index. The corrected effective line of sight. The coefficient representing the impact of traffic congestion on the effective sight distance correction.

[0056] Based on the comprehensive building shading impact coefficient Adjust the effective passage width:

[0057]

[0058]

[0059]

[0060] In the formula, This refers to the theoretical lane width of the section. The reduction in passage width caused by building obstruction. The correction factor for the effective passage width due to building obstruction; For vehicle model correction items, This is a correction factor for vehicle width. The reference vehicle width; The target vehicle width, This is the corrected effective passage width.

[0061] Preferably, the multi-dimensional car-following safety constraints are as follows:

[0062] Based on the corrected effective passage width and the comprehensive building shading impact coefficient, road condition constraints are constructed, including maximum allowable guiding speed constraints and slope constraints.

[0063] The maximum allowable guiding speed constraint is:

[0064] In the formula, To guide driving speed, To calculate the overall building shading impact coefficient, This is the corrected effective passage width. It is the acceleration due to gravity. The road surface adhesion coefficient, For road curvature, The correction factor for the maximum permissible guiding speed due to building obstruction;

[0065] The slope constraint is:

[0066]

[0067] In the formula, To guide the rate of change of driving speed, For road slope, The correction factor for the slope constraint caused by building shading;

[0068] The vehicle dynamics constraints include maximum speed constraints and acceleration constraints:

[0069]

[0070]

[0071] In the formula, Design the vehicle for its maximum speed. For the rated maximum acceleration, This is the rated maximum deceleration;

[0072] Based on the corrected braking response time and the comprehensive building occlusion impact coefficient, safety interaction constraints are constructed, including dynamic safe following distance constraints and predicted conflict probability constraints.

[0073] The dynamic safe following distance constraint is:

[0074]

[0075] Among them, the occlusion correction theoretical safe following distance corresponding to the candidate guidance speed for:

[0076]

[0077] In the formula, To guide following distance, The corrected braking response time, This is the vehicle's maximum braking deceleration. This is the safe distance correction amount corresponding to traffic congestion. To correct for the occlusion safety distance, The baseline theoretical safe following distance;

[0078] The constraint on the predicted conflict probability is:

[0079]

[0080] In the formula, Based on candidate guided driving speed Candidate following distance And the probability of vehicle interaction conflict predicted from the current motion state of neighboring vehicles; This represents the baseline maximum probability of vehicle interaction conflicts allowed under unobstructed conditions. This is a correction factor for the risk of occlusion.

[0081] Preferably, the three-dimensional target vector is as follows:

[0082] Introducing a comprehensive building shading impact coefficient as an energy consumption penalty term, an energy consumption calculation model incorporating driving resistance, acceleration resistance, and slope resistance is constructed. The minimum energy consumption target is then expressed as: :

[0083] In the formula, The equivalent energy consumption coefficient of driving resistance corresponding to the square of the guiding driving speed. Let be the equivalent energy consumption coefficient for acceleration and deceleration corresponding to the rate of change of the guiding driving speed. The three coefficients are pre-calibrated based on the target vehicle type and are equivalent to the slope resistance energy consumption coefficient corresponding to the road slope term. To calculate the overall building shading impact coefficient, To guide the rate of change of driving speed, To guide driving speed;

[0084] The calculation of travel time is differentiated between obstructed and unobstructed road sections. The travel speed on obstructed road sections is then adjusted based on a comprehensive building shading impact coefficient. The target for maximum traffic efficiency is then expressed as follows: :

[0085]

[0086] In the formula, Total travel time The total length of the narrow road network section To cover the continuous length of the road section, The correction factor for the travel time target caused by building obstruction;

[0087] To guide following distance Safe following distance with occlusion correction theory Taking the minimum deviation as the objective, and introducing a comprehensive building shading influence coefficient as a safety weight, the optimal safety distance objective is expressed as: :

[0088]

[0089] In the formula, The theoretical safe following distance is determined by the occlusion correction for the candidate guidance speed. The weighting correction factor for building occupancy on the safe distance target;

[0090] The overall objective vector is then expressed as: .

[0091] Preferably, step S4 is as follows:

[0092] S41. Based on multi-dimensional car-following safety constraints, determine the feasible range of values ​​for guiding driving speed and guiding following distance, and randomly generate an initial population within the feasible region;

[0093] S42. Based on the comprehensive building occlusion influence coefficient, dynamically adjust the distribution of reference points in the 3D target space in the NSGA-Ⅲ algorithm: the larger the comprehensive building occlusion influence coefficient, the higher the proportion of reference points in the target dimension with the optimal safe distance. The larger the value, the smaller the proportion of reference points in the target dimension with the lowest energy consumption and highest traffic efficiency; an adaptive 3D reference point set is generated based on the adjusted proportion. ;in The calculation formula is:

[0094]

[0095] In the formula, The percentage of reference points representing the target with the optimal safe distance when there is no obstruction. The upper limit of the proportion of the target reference point at the optimal safe distance. The overall building shading impact coefficient;

[0096] S43. Calculate the constraint violation degree of each individual in the current population regarding the dynamic safe following distance constraint. Apply adaptive penalties to individuals that violate both the dynamic safe following distance constraint and the lateral safe distance constraint. The penalty term is applied to the comprehensive target vector as components. Based on the penalized comprehensive target vector... Perform a non-dominated ranking on the individuals to obtain the non-dominated hierarchy of each individual:

[0097]

[0098] In the formula, To synthesize the target vector, The penalty gain coefficient, The degree of violation of safety constraints is the ratio of the amount of constraint exceedance to the constraint threshold; the larger the comprehensive building shading influence coefficient, the greater the penalty for violating safety constraints.

[0099] S44. Combine the non-dominated ranking results with the individual occlusion fitness to conduct tournament selection, and give priority to individuals with high non-dominated levels and high occlusion fitness to enter the offspring population.

[0100] The formula for calculating the individual occlusion fit is:

[0101]

[0102] In the formula, For individual occlusion adaptation, The corresponding guiding following distance for each individual vehicle. The theoretical safe following distance is adjusted for occlusion for each individual vehicle; a higher fit indicates a better safety margin for the individual vehicle in the current occlusion scenario.

[0103] S45. Normalize the target value of the candidate population formed by non-dominated sorting, and associate each candidate individual with the nearest adaptive 3D reference point. When the number of candidate individuals in the current non-dominated layer exceeds the remaining population capacity, perform survival selection based on reference point according to the number of individuals already associated with each reference point and the vertical distance of the candidate individual to the reference point, and prioritize retaining the candidate individuals corresponding to the reference point with the optimal target dimension of the safe distance.

[0104] S46. When the number of iterations reaches the preset maximum value or the change in the optimal solution over multiple consecutive generations is less than the preset convergence threshold, the iteration is terminated and the complete Pareto optimal solution set is output.

[0105] S47, Based on the comprehensive evaluation function Select the optimal compromise solution from the Pareto optimal solution set:

[0106]

[0107] , , The weighting coefficients for the corresponding targets are dynamically adjusted based on the comprehensive building shading impact coefficient, and The weight of the target with the optimal safe distance increases progressively with the increase of the comprehensive building shading influence coefficient, while the weight of the target with the highest traffic efficiency decreases progressively with the increase of the shading level. , , This serves as a reference baseline value for the corresponding target;

[0108] The Pareto solution with the minimum comprehensive evaluation function value is selected as the optimal set of guiding parameters;

[0109] S48. Based on the current road conditions and obstruction status, the roadside unit verifies the validity of the optimal driving speed and optimal following distance in the optimal guidance parameter set.

[0110] The speed validity condition is:

[0111]

[0112] in, To verify the target following speed, In the formula, The dynamic permissible safe speed is calculated based on road curvature, pavement adhesion coefficient, effective traffic width, and building obstruction status, which is the right-hand value of the maximum permissible guiding speed constraint; To achieve the optimal driving speed, This is the minimum speed limit for the section. This is the maximum speed limit for this section. Average driving speed The speed verification amplification factor is the one corresponding to the average vehicle speed, and ;

[0113] The spacing validity calibration condition is:

[0114]

[0115]

[0116] in, To verify the target following distance, , For optimal following distance, Based on the basic safety distance verification coefficient, To obscure the correction factor for the safety distance check, This is the spacing verification coefficient corresponding to the sight distance, and ; This is the corrected effective line-of-sight distance;

[0117] S49. If the validity verification passes, the verified guidance parameters are sent to the vehicle guidance device of the target vehicle through the vehicle-road cooperative network to push safe speed and following distance prompts to the driver; if the verification fails, return to step S41.

[0118] Preferably, the hierarchical closed-loop re-optimization mechanism specifically includes:

[0119] S51. Calculate the occlusion risk index at time t. :

[0120] In the formula, , , , The weight coefficients are non-negative and satisfy the following conditions: And the weight of building obstruction Not lower than the preset occlusion weight lower limit ; The comprehensive building shading influence coefficient at time t; Let be the vehicle density at time t. This represents the maximum traffic density for the section. The real-time longitudinal distance between the target vehicle and the vehicle in front at time t; The baseline theoretical safe following distance; Let t be the actual measured speed of the vehicle at time t;

[0121] The occlusion risk index is normalized, and the results are based on the normalized occlusion risk index. The values ​​are categorized into three risk levels: low risk. Medium risk High risk ,in , All are preset risk thresholds;

[0122] S52. Match the corresponding triggering conditions and response priorities according to the risk level. All risk level re-optimizations return to step S2 for execution, the difference being the re-optimization start time limit:

[0123] Level 1 High Priority Response: Triggered when any of the following conditions are met. Upon triggering, the current boot cycle is immediately interrupted, and the process returns to step S2 to perform immediate re-optimization and issue adjustment instructions:

[0124] (1) ;

[0125] (2) Change in building shading influence coefficient ;

[0126] (3) A new static obstacle is detected, and the current vehicle interaction conflict probability is high. ;

[0127] in This represents the maximum allowed collision probability under the current occlusion state.

[0128] Level 2 Medium Priority Response: Triggered when any of the following conditions are met, and after triggering, the response will be re-optimized at a preset medium risk interval. Returning to step S2 completes the re-optimization, and the current safe boot parameters remain unchanged until new results are generated:

[0129] (1) ;

[0130] (2) ;

[0131] (3) The percentage decrease in the corrected effective passage width exceeds the preset threshold;

[0132] Level 3 low priority response: Satisfied However, if traffic flow fluctuations exceed a preset ratio and the real-time longitudinal distance between the target vehicle and the vehicle in front consistently fails to meet safety requirements, the dynamic minimum re-optimization interval will be triggered. In the next update cycle, return to step S2 for re-optimization;

[0133] in All are threshold values ​​for changes in occlusion coefficient, and ;

[0134] Minimum re-optimization interval period The calculation formula is:

[0135] In the formula, This is the lower bound of the minimum allowable re-optimization interval period. This represents the maximum re-optimization interval under low-risk conditions.

[0136] S53. Calculate the interval execution error of the current control cycle. :

[0137] In the formula, This is the updated measured longitudinal following distance after the guidance parameters for the current control cycle are issued. The target following distance is issued after verification in the current control cycle;

[0138] Based on the spacing execution error and the comprehensive building occlusion influence coefficient, the weight coefficients of the optimal safe distance target in the comprehensive evaluation function are corrected:

[0139] In the formula, The weighting coefficients for the optimal target at the corrected safe distance; To correct the gain coefficient, the three weights are then normalized.

[0140] After each fixed period, the weight coefficients of the target with the optimal safe distance in the comprehensive evaluation function are calibrated offline based on the execution error, forming a complete closed loop of monitoring-evaluation-optimization-execution-correction.

[0141] A vehicle following optimization system for narrow road networks based on UAV building occlusion quantification includes a processor, a memory, and a computer program stored in the memory. When the processor executes the computer program, it specifically performs any of the steps in the aforementioned vehicle following optimization method for narrow road networks.

[0142] Compared with the prior art, the present invention has the following beneficial effects:

[0143] (1) Construct a dual-mode collaborative data collection mechanism of global patrol and low-altitude follow-flying of UAVs. Through the complementary perception of full coverage and refined follow-flying, it breaks through the bottleneck of large blind spots and insufficient data dimensions of fixed monitoring equipment under the obstruction of narrow road network buildings. It realizes the synchronous collection of multi-source data of road environment, building obstruction, traffic flow and vehicle interaction, and can form a complete and effective perception coverage of the target narrow road network area, providing a complete perception foundation for refined vehicle follow-flying optimization.

[0144] (2) A multi-dimensional building obstruction quantification method and grading standard are proposed. The obstruction impact coefficient is calculated and weighted from three dimensions: sight distance, horizontal and vertical. A normalized truncation and safety boundary processing mechanism is provided to further correct the effective sight distance, effective passage width and driver braking response time. The qualitative safety impact of building obstruction is transformed into a quantifiable optimization parameter correction item, providing a precise quantitative basis for the construction of car-following safety constraints.

[0145] (3) Construct a multi-dimensional car-following safety constraint system that couples the full-dimensional influence of building occlusion. Embed the occlusion influence coefficient into road condition constraints, vehicle dynamics constraints and safety interaction constraints, so that the safety optimization threshold can be dynamically adjusted according to the degree of occlusion. This effectively improves the car-following safety redundancy in occlusion scenarios and solves the problem of mismatch between static optimization standards and actual risks.

[0146] (4) The NSGA-Ⅲ algorithm is adapted to the building occlusion optimization scenario. The algorithm input and output boundaries are clarified. Through feasible region population initialization, adaptive reference point distribution with upper limit, safety constraint adaptive penalty and occlusion adaptability selection mechanism, the algorithm is guided to converge to a feasible region with high safety margin. The balanced optimization of the three conflicting goals of energy consumption, traffic efficiency and safety is achieved, and the car-following optimization needs of different occlusion levels are adapted.

[0147] (5) Establish a hierarchical closed-loop re-optimization mechanism based on the normalized occlusion risk index. Combine the dynamic changes of occlusion with the risk level to match the differentiated response priority and optimization update interval. This ensures rapid safety warnings when occlusion changes suddenly occur, while avoiding driver operation interference caused by frequent guidance. At the same time, the model self-correction is achieved through execution error feedback, continuously improving optimization adaptability and accuracy. Attached Figure Description

[0148] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0149] The following is in conjunction with the appendix Figure 1 The technical solution of the present invention will be described in detail below.

[0150] This invention proposes a vehicle following optimization method for narrow road networks based on UAV building occlusion quantification, comprising the following steps:

[0151] S1. Based on the dual-mode collaborative acquisition mechanism of global cruise and low-altitude follow-fly of UAVs, multi-dimensional perception data is collected in real time in narrow road network sections. After preprocessing, standardized scene data is obtained, including target vehicle information, global traffic status information, road environment and building occlusion parameters, and vehicle interaction data in narrow road network sections.

[0152] S2. Calculate the reduction in sight distance due to occlusion and the comprehensive building occlusion impact coefficient based on standardized scene data; and correct the effective sight distance, effective passage width, and braking response time based on the reduction in sight distance due to occlusion and the comprehensive building occlusion impact coefficient; if the corrected effective sight distance is lower than the preset minimum safe sight distance, or the corrected effective passage width is lower than the preset minimum safe passage width, or the lateral safety margin is lower than the preset threshold, then terminate the subsequent optimization steps and trigger a safety response command (deceleration, stopping, avoidance, or manual intervention prompt); otherwise, proceed to step S3.

[0153] S3. Construct a multi-objective car-following optimization model, using guided driving speed and guided following distance as decision variables; wherein, the guided driving speed is the candidate driving speed to be recommended to the target vehicle, and the guided following distance is the candidate following distance to be recommended to the target vehicle; the multi-objective car-following optimization model includes multi-dimensional car-following safety constraints and a three-dimensional target vector;

[0154] The multi-dimensional car-following safety constraints are constructed based on the comprehensive building obstruction influence coefficient, the corrected braking response time, and the effective passage width, including road condition constraints, vehicle dynamics constraints, and safety interaction constraints.

[0155] The three-dimensional target vector consists of three sub-objectives: lowest energy consumption, highest traffic efficiency, and optimal safe distance, and each objective function is coupled with a comprehensive building occlusion influence coefficient.

[0156] S4. The improved NSGA-Ⅲ algorithm is used to solve the multi-objective car-following optimization model to obtain the Pareto optimal solution set. Based on the comprehensive evaluation function, the optimal guidance parameter set is selected from the Pareto optimal solution set. After validity verification, it is sent to the vehicle guidance device (vehicle terminal, navigation application or vehicle assisted driving control module) of the target vehicle through the 5G-V2X vehicle-road cooperative network to push safe speed and following distance prompts to the driver.

[0157] S5. Establish a hierarchical closed-loop re-optimization mechanism based on the occlusion risk index: Calculate the normalized occlusion risk index to classify risk levels. When the preset hierarchical triggering conditions are met, trigger the re-optimization process of the corresponding priority. And periodically adjust the weight of the comprehensive objective function according to the spacing execution error.

[0158] In this embodiment, the specific contents of the target vehicle information, global traffic status information, road environment and building occlusion parameters, and vehicle interaction data in the narrow road network section are as follows:

[0159] The target vehicle information includes the following parameters of the guided target vehicle traveling in a narrow road network section with obstructing buildings: the vehicle's measured speed. (Low-altitude follow-flying drones generate measured speed sequences through continuous frame target localization), vehicle width Vehicle braking performance parameters, vehicle design maximum speed Rated maximum acceleration and rated maximum deceleration The vehicle braking performance parameters include the standard driver braking response time. and the vehicle's maximum braking deceleration ;

[0160] The global traffic status information includes the following parameters for narrow road network sections: total length of the narrow road network section. Traffic flow Vehicle density Maximum traffic density of the section Average driving speed Traffic congestion index Length of continuous obstruction The target vehicle speed obtained from coarse-grained identification The longitudinal distance between the target vehicle and the vehicle in front obtained by coarse-grained identification and the benchmark theoretical safe following distance ;

[0161] The road environment and building obstruction parameters include the following parameters for narrow road network sections: theoretical lane width of the section. Road curvature Road slope Road surface adhesion coefficient Static obstacle position coordinates Minimum speed limit for the section Maximum speed limit for the section The shortest line of sight in the section theory Coordinates of the building's obstruction location Building height Effective lateral width of building obstruction Area of ​​obstructed view Total monitoring field of view area Angle of view from the driver ;

[0162] The vehicle interaction data includes the following parameters in narrow road network sections: real-time longitudinal distance between the target vehicle and the vehicle in front. Speed ​​of the vehicle in front Lateral center distance between the target vehicle and vehicles in adjacent lanes Adjacent lane vehicle width Probability of interaction conflict with the current vehicle The It is calculated based on the relative position, relative speed and preset prediction time window of the target vehicle with the vehicle in front and the vehicles in adjacent lanes;

[0163] The target vehicle information, global traffic status information, road environment and building occlusion parameters, and vehicle interaction data of the narrow road network section are standardized scene data after spatiotemporal coordinate calibration, noise reduction (moving average filtering), and outlier removal.

[0164] In this embodiment, the dual-mode collaborative data acquisition mechanism of UAV global cruise and low-altitude follow-fly is as follows:

[0165] In global cruise mode, the drone covers narrow road network sections along a preset flight path, completing the collection of global traffic status information and road environment and building occlusion parameters; the preset flight path is set according to the effective monitoring area of ​​the drone, so that the coverage of the narrow road network section by the preset flight path is not less than a preset threshold.

[0166]

[0167] In the formula, For road network coverage, To determine the effective monitoring area of ​​the drone, The total area of ​​the target road segment;

[0168] In low-altitude follow-fly mode, the drone maintains synchronous flight with the guided target vehicle, completing the collection of target vehicle information and vehicle interaction data; wherein, the relative position control error during follow-flying satisfies:

[0169]

[0170] In the formula, This refers to the error between the actual longitudinal position of the drone and its preset longitudinal tracking position relative to the target vehicle. This represents the error between the actual lateral position of the drone and the preset lateral following position. This represents the error between the actual flight altitude of the drone and the preset follow-up altitude. The preset maximum follow-fly position error threshold is used;

[0171] In the global patrol mode, the drone prioritizes the detection of target vehicles. Intelligent switching:

[0172] In the formula, For preset coefficients, This is a correction factor for the following distance risk item, and ; The target vehicle speed is obtained through coarse-grained identification in global cruise mode. This refers to the longitudinal distance between the target vehicle and the vehicle in front, obtained through coarse-grained identification in global cruise mode. This is the maximum speed limit for this section. The baseline theoretical safe following distance;

[0173] when When the preset switching threshold is reached, the drone switches to the low-altitude follow-fly mode corresponding to the target vehicle;

[0174] In low-altitude follow-fly mode, the actual vehicle speed is used. Real-time longitudinal distance between the target vehicle and the vehicle in front Replace the coarse-grained recognition and The drone recalculates the target vehicle monitoring priority; when the target vehicle leaves the obstructed area, or when the recalculated target vehicle monitoring priority is continuously lower than the preset exit threshold, the drone returns to global cruise mode.

[0175] In this embodiment, the calculation of the occlusion viewing distance reduction and the comprehensive building occlusion impact coefficient based on standardized scene data is as follows:

[0176] Calculate the reduction in sight distance due to occlusion :

[0177]

[0178] In the formula, This is the shortest sight distance in the section theory. For building height, The driver's line of sight angle;

[0179] Calculate the line-of-sight occlusion coefficient :

[0180]

[0181] Calculate the lateral occlusion coefficient :

[0182]

[0183] In the formula, The effective lateral width for building shading. This refers to the theoretical lane width of the section.

[0184] Calculate the longitudinal occlusion coefficient :

[0185]

[0186] In the formula, To cover the continuous length of the road section, This refers to the total length of the narrow road network section;

[0187] Calculate the area shading coefficient :

[0188]

[0189] In the formula, To block the area of ​​vision, Total monitored field of view area;

[0190] Calculate the initial value of the comprehensive building shading impact coefficient :

[0191]

[0192] In the formula, These are the fusion weights for the line-of-sight occlusion coefficient, lateral occlusion coefficient, longitudinal occlusion coefficient, and area occlusion coefficient, respectively. ;

[0193] The initial value of the comprehensive building shading impact coefficient is normalized and truncated to obtain the comprehensive building shading impact coefficient. The occlusion level is divided into: no occlusion. Slight obstruction Moderate occlusion Severe occlusion .

[0194] In this embodiment, the correction of effective sight distance, effective passage width, and braking response time based on the reduction in sight distance due to occlusion and the comprehensive building occlusion influence coefficient is as follows:

[0195] Based on the comprehensive building shading impact coefficient Adjusted braking response time to reflect the impact of perception lag caused by obstruction on following safety:

[0196] In the formula, This is the standard driver braking response time when there is no obstruction. The corrected braking response time, The gain coefficient representing the effect of building obstruction on driver braking response time;

[0197] Based on the reduction in viewing distance due to occlusion Corrected effective line of sight:

[0198]

[0199] In the formula, This is the shortest sight distance in the section theory. The traffic congestion index. The corrected effective line of sight. The coefficient representing the impact of traffic congestion on the effective sight distance correction.

[0200] Based on the comprehensive building shading impact coefficient Adjust the effective passage width:

[0201]

[0202]

[0203]

[0204] In the formula, This refers to the theoretical lane width of the section. The reduction in passage width caused by building obstruction. The correction factor for the effective passage width due to building obstruction; For vehicle model correction items, This is a correction factor for vehicle width. The reference vehicle width; The target vehicle width, This is the corrected effective passage width.

[0205] In this embodiment, the multi-dimensional car-following safety constraints are specifically as follows:

[0206] Based on the corrected effective passage width and the comprehensive building shading impact coefficient, road condition constraints are constructed, including maximum allowable guiding speed constraints and slope constraints.

[0207] The maximum allowable guiding speed constraint is:

[0208] In the formula, To guide driving speed, To calculate the overall building shading impact coefficient, This is the corrected effective passage width. It is the acceleration due to gravity. The road surface adhesion coefficient, For road curvature, The correction factor for the maximum permissible guiding speed due to building obstruction;

[0209] The slope constraint is:

[0210] In the formula, To guide the rate of change of driving speed, For road slope, The correction factor for the slope constraint caused by building shading;

[0211] The vehicle dynamics constraints include maximum speed constraints and acceleration constraints:

[0212]

[0213]

[0214] In the formula, Design the vehicle for its maximum speed. For the rated maximum acceleration, This is the rated maximum deceleration;

[0215] Based on the corrected braking response time and the comprehensive building occlusion impact coefficient, safety interaction constraints are constructed, including dynamic safe following distance constraints and predicted conflict probability constraints.

[0216] The dynamic safe following distance constraint is:

[0217]

[0218] Among them, the occlusion correction theoretical safe following distance corresponding to the candidate guidance speed for:

[0219]

[0220] In the formula, To guide following distance, The corrected braking response time, This is the vehicle's maximum braking deceleration. This is the safe distance correction amount corresponding to traffic congestion. To correct for the occlusion safety distance, The baseline theoretical safe following distance;

[0221] The constraint on the predicted conflict probability is:

[0222]

[0223] In the formula, Based on candidate guided driving speed Candidate following distance And the probability of vehicle interaction conflict predicted from the current motion state of neighboring vehicles; This represents the baseline maximum probability of vehicle interaction conflicts allowed under unobstructed conditions. This is a correction factor for the risk of occlusion.

[0224] In this embodiment, the three-dimensional target vector is specifically as follows:

[0225] Introducing a comprehensive building shading impact coefficient as an energy consumption penalty term, an energy consumption calculation model incorporating driving resistance, acceleration resistance, and slope resistance is constructed. The minimum energy consumption target is then expressed as: :

[0226] In the formula, The equivalent energy consumption coefficient of driving resistance corresponding to the square of the guiding driving speed. The equivalent energy consumption coefficient of driving resistance corresponding to the square of the guiding driving speed. The three coefficients are pre-calibrated based on the target vehicle type and are equivalent to the slope resistance energy consumption coefficient corresponding to the road slope term. To calculate the overall building shading impact coefficient, To guide the rate of change of driving speed, To guide driving speed;

[0227] The calculation of travel time is differentiated between obstructed and unobstructed road sections. The travel speed on obstructed road sections is then adjusted based on a comprehensive building shading impact coefficient. The target for maximum traffic efficiency is then expressed as follows: :

[0228]

[0229] In the formula, Total travel time The total length of the narrow road network section To cover the continuous length of the road section, The correction factor for the travel time target caused by building obstruction;

[0230] To guide following distance Safe following distance with occlusion correction theory Taking the minimum deviation as the objective, and introducing a comprehensive building shading influence coefficient as a safety weight, the optimal safety distance objective is expressed as: :

[0231]

[0232] In the formula, The theoretical safe following distance is determined by the occlusion correction for the candidate guidance speed; The weighting correction factor for building occupancy on the safe distance target;

[0233] The overall objective vector is then expressed as: .

[0234] There are coupling and conflict relationships among the objectives. The non-dominated optimal solution set is solved by improving the NSGA-Ⅲ algorithm.

[0235] In this embodiment, step S4 is as follows:

[0236] S41. Based on multi-dimensional car-following safety constraints, determine the feasible range of values ​​for guiding driving speed and guiding following distance, and randomly generate an initial population within the feasible region;

[0237] S42. Based on the comprehensive building occlusion influence coefficient, dynamically adjust the distribution of reference points in the 3D target space in the NSGA-Ⅲ algorithm: the larger the comprehensive building occlusion influence coefficient, the higher the proportion of reference points in the target dimension with the optimal safe distance. The larger the value, the smaller the proportion of reference points in the target dimension with the lowest energy consumption and highest traffic efficiency; an adaptive 3D reference point set is generated based on the adjusted proportion. The higher the degree of occlusion, the more numerous and densely distributed the reference points corresponding to the safe distance target, guiding the population to converge towards the feasible region with a high safety margin; when the proportion of reference points for the safe target reaches the upper limit, the proportion of reference points for the energy consumption target and the traffic efficiency target is normalized and redistributed; among which The calculation formula is:

[0238]

[0239] In the formula, The percentage of reference points representing the target with the optimal safe distance when there is no obstruction. The upper limit of the proportion of the target reference point at the optimal safe distance. The overall building shading impact coefficient;

[0240] S43. Calculate the constraint violation degree of each individual in the current population regarding the dynamic safe following distance constraint. Apply adaptive penalties to individuals that violate both the dynamic safe following distance constraint and the lateral safe distance constraint. The penalty term is applied to the comprehensive target vector as components. Based on the penalized comprehensive target vector... Perform a non-dominated ranking on the individuals to obtain the non-dominated hierarchy of each individual:

[0241]

[0242] In the formula, To synthesize the target vector, The penalty gain coefficient, The degree of violation of safety constraints is the ratio of the amount of constraint exceedance to the constraint threshold; the larger the comprehensive building shading influence coefficient, the greater the penalty for violating safety constraints.

[0243] S44. Combine the non-dominated ranking results with the individual occlusion fitness to conduct tournament selection, and give priority to individuals with high non-dominated levels and high occlusion fitness to enter the offspring population.

[0244] The formula for calculating the individual occlusion fit is:

[0245]

[0246] In the formula, For individual occlusion adaptation, The corresponding guiding following distance for each individual vehicle. The theoretical safe following distance is adjusted for occlusion for each individual vehicle; a higher fit indicates a better safety margin for the individual vehicle in the current occlusion scenario.

[0247] S45. Normalize the target value of the candidate population formed by non-dominated sorting, and associate each candidate individual with the nearest adaptive 3D reference point. When the number of candidate individuals in the current non-dominated layer exceeds the remaining population capacity, survival selection based on reference points is performed according to the number of individuals associated with each reference point and the vertical distance from the candidate individual to the reference point, prioritizing the retention of candidate individuals corresponding to the reference point with the optimal target dimension of safe distance;

[0248] S46. When the number of iterations reaches the preset maximum value or the change in the optimal solution over multiple consecutive generations is less than the preset convergence threshold, the iteration is terminated and the complete Pareto optimal solution set is output.

[0249] S47, Based on the comprehensive evaluation function Select the optimal compromise solution from the Pareto optimal solution set:

[0250]

[0251] In the formula, , , The weighting coefficients for the corresponding targets are dynamically adjusted based on the comprehensive building shading impact coefficient, and The weight of the target with the optimal safe distance increases progressively with the increase of the comprehensive building shading influence coefficient, while the weight of the target with the highest traffic efficiency decreases progressively with the increase of the shading level. , , This serves as a reference baseline value for the corresponding target;

[0252] Optionally, the weighting coefficient is dynamically adjusted based on the shading level determined by the comprehensive building shading impact coefficient: when there is no shading. =0.35、 =0.4、 =0.25, under mild shading =0.35、 =0.35、 =0.3, when there is moderate shading =0.3、 =0.25、 =0.45, under severe occlusion =0.3、 =0.2、 =0.5; the weight of the safety target increases progressively with the level of occlusion, while the weight of the traffic efficiency target decreases progressively with the level of occlusion.

[0253] The Pareto solution with the minimum comprehensive evaluation function value is selected as the optimal set of guiding parameters;

[0254] S48. Based on the current road conditions and obstruction status, the roadside unit verifies the validity of the optimal driving speed and optimal following distance in the optimal guidance parameter set.

[0255] The speed validity condition is:

[0256]

[0257] in, To verify the target following speed, In the formula, The dynamic permissible safe speed is calculated based on road curvature, pavement adhesion coefficient, effective traffic width, and building obstruction status, which is the right-hand value of the maximum permissible guiding speed constraint; To achieve the optimal driving speed, This is the minimum speed limit for the section. This is the maximum speed limit for this section. Average driving speed The speed verification amplification factor is the one corresponding to the average vehicle speed, and ;

[0258] The spacing validity calibration condition is:

[0259]

[0260]

[0261] in, To verify the target following distance, , For optimal following distance, Based on the basic safety distance verification coefficient, Correction factor for occlusion verification of safe distance; This is the spacing verification coefficient corresponding to the sight distance, and ; This is the corrected effective line-of-sight distance;

[0262] S49. If the validity verification passes, the verified guidance parameters are sent to the vehicle guidance device of the target vehicle through the vehicle-road cooperative network to push safe speed and following distance prompts to the driver; if the verification fails, return to step S41.

[0263] In this embodiment, the hierarchical closed-loop re-optimization mechanism specifically includes:

[0264] S51. Calculate the occlusion risk index at time t. :

[0265] In the formula, , , , The weight coefficients are non-negative and satisfy the following conditions: And the weight of building obstruction Not lower than the preset occlusion weight lower limit ; The comprehensive building shading influence coefficient at time t; Let be the vehicle density at time t. This represents the maximum traffic density for the section. The real-time longitudinal distance between the target vehicle and the vehicle in front at time t; The baseline theoretical safe following distance; Let t be the actual measured speed of the vehicle at time t;

[0266] The occlusion risk index is normalized, and the results are based on the normalized occlusion risk index. The values ​​are categorized into three risk levels: low risk. Medium risk High risk ,in , All are preset risk thresholds;

[0267] S52. Match the corresponding triggering conditions and response priorities according to the risk level. All risk level re-optimizations return to step S2 for execution, the difference being the re-optimization start time limit:

[0268] Level 1 High Priority Response: Triggered when any of the following conditions are met. Upon triggering, the current boot cycle is immediately interrupted, and the process returns to step S2 to perform immediate re-optimization and issue adjustment instructions:

[0269] (1) ;

[0270] (2) Change in building shading influence coefficient ;

[0271] (3) A new static obstacle is detected, and the current vehicle interaction conflict probability is high. ;

[0272] in This represents the maximum allowed collision probability under the current occlusion state.

[0273] Level 2 Medium Priority Response: Triggered when any of the following conditions are met, and after triggering, the response will be re-optimized at a preset medium risk interval. Returning to step S2 completes the re-optimization, and the current safe boot parameters remain unchanged until new results are generated:

[0274] (1) ;

[0275] (2) ;

[0276] (3) The percentage decrease in the corrected effective passage width exceeds the preset threshold;

[0277] Level 3 low priority response: Satisfied However, if traffic flow fluctuations exceed a preset ratio and the real-time longitudinal distance between the target vehicle and the vehicle in front consistently fails to meet safety requirements, the dynamic minimum re-optimization interval will be triggered. In the next update cycle, return to step S2 for re-optimization;

[0278] in All are threshold values ​​for changes in occlusion coefficient, and ;

[0279] Minimum re-optimization interval period To suppress vehicle control oscillations caused by frequent re-optimization, the calculation formula is as follows:

[0280] In the formula, This is the lower bound of the minimum allowable re-optimization interval period. This represents the maximum re-optimization interval under low-risk conditions; the higher the occlusion risk index, the shorter the minimum re-optimization interval and the more frequent the re-optimization response.

[0281] S53. Calculate the interval execution error of the current control cycle. :

[0282] In the formula, After issuing the guidance parameters for the current control cycle The target following distance is issued after verification in the current control cycle;

[0283] Based on the spacing execution error and the comprehensive building occlusion influence coefficient, the weight coefficients of the optimal safe distance target in the comprehensive evaluation function are corrected:

[0284] In the formula, The weighting coefficients for the optimal target at the corrected safe distance; To correct the gain coefficient, the three weights are then normalized.

[0285] After each fixed period, the weight coefficients of the target with the optimal safe distance in the comprehensive evaluation function are calibrated offline based on the execution error, forming a complete closed loop of monitoring-evaluation-optimization-execution-correction.

[0286] A vehicle following optimization system for narrow road networks based on UAV building occlusion quantification includes a processor, a memory, and a computer program stored in the memory. When the processor executes the computer program, it specifically performs any of the steps in the aforementioned vehicle following optimization method for narrow road networks.

[0287] The following are specific examples:

[0288] Step S1: Dual-mode data acquisition and preprocessing:

[0289] A dual-mode collaborative data collection mechanism of global drone patrol and low-altitude follow-fly was constructed to simultaneously collect target vehicle information, global traffic status information, road environment and building obstruction parameters, and vehicle interaction data in narrow road networks, and preprocess the collected data.

[0290] In this embodiment, the UAV platform is equipped with a high-definition visible light camera, LiDAR, and a BeiDou high-precision positioning module, deployed at a roadside base station, and equipped with a 5G-V2X communication module to realize data transmission and command issuance. The core hardware and operating parameters are as follows:

[0291] Global cruise flight altitude: 30m–80m, traversing the target's narrow road network along a preset gridded flight path, collecting road geometry, building distribution and global traffic flow data, with road network coverage controlled at over 98%;

[0292] Low-altitude follow-flying altitude: 10m–30m, maintaining a longitudinal follow-flying distance of 50–80m from the target vehicle, and collecting real-time data on the target vehicle's operating status, the interaction status of the vehicle in front, and details of partial obstruction.

[0293] Image acquisition frame rate: no less than 25fps;

[0294] LiDAR point cloud frequency: not less than 10Hz;

[0295] BeiDou positioning accuracy: decimeter level;

[0296] 5G-V2X end-to-end communication latency: no higher than 100ms;

[0297] Maximum position error threshold for low-altitude follow-flying: ≤2m in the horizontal direction and ≤1m in the vertical direction.

[0298] Dual-mode switching is based on optimization target priority. Drive, when When a target is identified as a high-priority optimization target, the drone switches to low-altitude follow-fly mode to conduct refined monitoring; when During this time, the drone maintains a global cruise mode. All collected data is uniformly calibrated in spatiotemporal coordinates through the BeiDou positioning system, missing values ​​are filled using third-order spline interpolation, outliers are removed using the quartile method, and after data preprocessing, it is uploaded to the roadside edge computing node.

[0299] Based on the probability of current vehicle interaction conflicts in vehicle interaction data The joint probability method based on time-distance risk is used to calculate the risk, which integrates two types of risks: longitudinal following conflict and lateral conflict. The correlation and calculation method between these risks and relative position, relative speed, and preset prediction time window are as follows:

[0300] Parameter association logic:

[0301] Relative spacing with longitudinal direction A negative correlation exists: the smaller the longitudinal distance between the target vehicle and the vehicle in front, the less space there is for collision buffering, and the higher the probability of conflict.

[0302] relative velocity with longitudinal direction There is a positive correlation: the faster the following vehicle is relative to the vehicle in front, the faster the risk of a rear-end collision increases and the higher the probability of conflict.

[0303] Remaining lateral safety distance A negative correlation exists: the smaller the lateral distance between adjacent vehicles, the higher the risk of lateral scraping;

[0304] It is positively correlated with lateral relative velocity: the greater the lateral deviation velocity of the vehicle, the shorter the time of lateral collision.

[0305] With preset prediction time window Positive correlation: The longer the prediction time domain and the wider the range of scenarios included in the risk assessment, the higher the calculated probability of conflict.

[0306] Calculation formula:

[0307] ① Calculate the longitudinal collision time :

[0308]

[0309] In the formula, This represents the real-time longitudinal distance between the target vehicle and the vehicle in front. The actual driving speed of the target vehicle was measured. The speed of the vehicle in front; when At that time, there is no risk of longitudinal collision. Take infinity.

[0310] ② Calculate the lateral collision time :

[0311]

[0312] In the formula, The lateral center distance between the target vehicle and vehicles in the adjacent lane. , These are the target vehicle width and the adjacent lane vehicle width, respectively. The lateral relative velocity, , They are respectively , The lateral component; when the lateral relative velocity is 0 and the lateral margin is sufficient, there is no risk of lateral collision. Take infinity.

[0313] ③ Combine with preset prediction time window Calculate the joint conflict probability:

[0314]

[0315] In the formula, For unobstructed baseline collision probability, This is a preset prediction time window.

[0316] Calculation Example

[0317] Using the parameters from the subsequent moderate occlusion scenario: the actual speed of the target vehicle. Speed ​​of the vehicle in front Vertical real-time spacing Lateral center-to-center distance between vehicles in adjacent lanes Target vehicle width Adjacent lane width The lateral relative velocity is Take the baseline conflict probability Preset prediction time window .

[0318] The calculation yields:

[0319] Longitudinal collision time There is no immediate conflict in the vertical direction;

[0320] Lateral remaining safety distance Lateral collision time ;

[0321] Final probability of current vehicle interaction conflict .

[0322] Step S2: Quantification and optimization parameter correction of building occlusion:

[0323] Based on standardized scene data, a multi-dimensional occlusion coefficient is generated and a comprehensive building occlusion impact coefficient is calculated. This coefficient is then coupled to the correction process of road capacity parameters and driver response parameters to obtain the corrected effective sight distance, effective passage width, and braking response time.

[0324] In this embodiment, the building shading level is divided into four levels according to the comprehensive influence coefficient, and the classification criteria are shown in the table below:

[0325]

[0326] Based on standardized scene data, the reduction in sight distance due to occlusion, the sight distance occlusion coefficient, the lateral occlusion coefficient, and the longitudinal occlusion coefficient are calculated sequentially and weighted to obtain the comprehensive building occlusion impact coefficient. All occlusion coefficients are normalized to 0–1 to avoid extreme values ​​that could cause subsequent calculation anomalies, and are matched with the corresponding building occlusion level. On this basis, the driver's braking response time, effective sight distance, and effective passage width are corrected to complete the occlusion adaptation correction of the optimized parameters. When the corrected effective sight distance is lower than the preset minimum safe sight distance, or the effective passage width is lower than the preset minimum safe passage width, an extreme scenario safety warning is triggered, and no normal guidance parameters are output.

[0327] Numerical examples:

[0328] A narrow road section is 200m long, with a lane width of 3.2m, a building height of 8m, a continuous obstruction length of 60m, an effective lateral width of obstruction of 0.6m, a theoretical minimum sight distance of 80m, a traffic congestion index of 0.4, a target vehicle width of 1.8m, a driver's eye elevation angle of 15°, and an unobstructed standard braking response time of 0.8s.

[0329] Based on visible light images and lidar point clouds acquired by the UAV at the corresponding monitoring time, the building outline is extracted after spatiotemporal registration, and the building-occupied area is projected onto the preset road monitoring plane; the area corresponding to the building-occupied area is calculated to obtain the obstructed field of view area. The total field of view of the drone is obtained by projecting the effective field of view onto the road monitoring plane. .

[0330] The calculation process is as follows:

[0331] 1. Reduction in viewing distance due to occlusion:

[0332] 2. View distance occlusion coefficient:

[0333] 3. Lateral shading coefficient:

[0334] 4. Vertical shading coefficient:

[0335] 5. Area shading coefficient:

[0336] 6. Overall shading coefficient: After truncation, the value is 0.403, indicating moderate occlusion.

[0337] 7. Corrected braking response time:

[0338] 8. Corrected effective line of sight:

[0339] 9. Adjust passage width:

[0340] Step S3: Construction of a multi-dimensional car-following safety constraint system:

[0341] Using guided driving speed and guided following distance as decision variables, the comprehensive building shading influence coefficient, the corrected braking response time, and the effective passage width are introduced into various constraints to construct a multi-dimensional following safety constraint system that couples the full-dimensional influence of building shading. This system includes three types of constraints: road condition constraints, vehicle dynamics constraints, and safety interaction constraints. Among them, road condition constraints include maximum permissible guided speed constraints, lateral safety offset constraints, and slope constraints; vehicle dynamics constraints include maximum speed constraints and acceleration constraints; and safety interaction constraints include dynamic safety following distance constraints, lateral safety distance constraints, and conflict probability constraints. All constraints are coupled with the comprehensive building shading influence coefficient to achieve dynamic adjustment of the optimization threshold according to the degree of shading.

[0342] (1) Road conditions constraints

[0343] The maximum permissible guiding speed constraint expression is:

[0344]

[0345] The right-hand side of the formula represents the dynamic permissible safe speed. .

[0346] Dynamically allowed safe speed The positive and negative correlations with each influencing parameter are as follows:

[0347] ① With road surface adhesion coefficient A positive correlation exists: the higher the road surface adhesion coefficient, the stronger the vehicle's anti-skid ability and braking performance, and the higher the upper limit of the driving speed that the curve can withstand.

[0348] ② With the corrected effective passage width A positive correlation exists: the wider the effective passage width, the more sufficient the safety margin for lateral vehicle offset, and the higher the safe driving speed allowed on curves;

[0349] ③ With road curvature Negative correlation: The greater the road curvature (corresponding to the smaller the curve radius), the greater the centrifugal force generated by the vehicle, the higher the risk of rollover and skidding, and the lower the permissible safe speed;

[0350] ④ Combined with the building's shading influence coefficient A negative correlation exists: the greater the building obstruction, the more limited the driver's visibility and the less lateral maneuvering space, necessitating further reductions in permissible speeds to allow for reaction redundancy and safety margins. Calculation example:

[0351] Using the same parameters for the moderately occluded scene as described above, we take the acceleration due to gravity. Road surface adhesion coefficient (Dry asphalt pavement), road curvature (Corresponding to a curve radius of 80m), corrected effective passage width Comprehensive building shading impact coefficient Correction factor Substitute into the formula to calculate:

[0352]

[0353] This value is higher than the maximum speed limit in typical narrow road network sections, so the actual speed constraint is controlled by the maximum speed limit of the section. The lateral safety offset constraint limits the allowable offset of the vehicle's lateral travel position. The higher the degree of occlusion, the smaller the allowable lateral offset, avoiding the risk of collision with buildings. The slope constraint limits the acceleration and deceleration amplitude in slope scenarios. The threshold is tightened simultaneously in occluded scenarios to ensure the stability of car following on slopes.

[0354] (2) Vehicle dynamics constraints limit the guiding speed to no more than the vehicle's maximum design speed and the guiding acceleration and deceleration to no more than the vehicle's rated power limit, ensuring that the guiding parameters match the vehicle's physical performance and avoiding the output of unexecutable guiding commands.

[0355] (3) In the safety interaction constraints, the dynamic safety following distance constraint limits the guiding following distance to no less than the theoretical safe distance after the occlusion correction, so as to ensure longitudinal following safety;

[0356] The expression for the probability constraint of predicting conflicts is:

[0357]

[0358] Vehicle interaction conflict probability under candidate guidance parameters The positive and negative correlations with each influencing parameter are as follows:

[0359] ① and candidate guiding driving speed A positive correlation exists: the higher the guide speed, the longer the vehicle braking distance, the shorter the longitudinal collision time, and the higher the risk of lateral deviation in curves, thus increasing the probability of conflict.

[0360] ② Following distance from the candidate guide vehicle A negative correlation exists: the larger the following distance, the more sufficient the longitudinal safety margin, the longer the longitudinal collision time, and the lower the probability of conflict.

[0361] ③ Positively correlated with the relative speed difference between neighboring vehicles: The greater the speed difference between the target vehicle and the neighboring vehicle, the stronger the relative approach trend and the higher the risk of conflict;

[0362] ④ Lateral spacing with adjacent vehicles A negative correlation exists: the closer the lateral distance between adjacent vehicles, the smaller the lateral safety margin, and the higher the probability of lateral scraping and collision;

[0363] ⑤ Combined with the building's shading influence coefficient A positive correlation exists: the higher the degree of obstruction, the slower the driver's perception, the more limited the operating space, and the higher the actual risk of conflict under the same motion conditions. The calculation expression is:

[0364]

[0365] In the formula, To guide the lateral component of velocity, To preset the prediction time window, They are respectively The other parameters of the horizontal component are defined as before.

[0366] Calculation example:

[0367] Using the aforementioned moderate occlusion scenario, candidate guidance speeds are selected. Candidate following distance Lateral center-to-center distance of adjacent lanes The width of each vehicle is 1.8m, and the lateral relative speed is... Baseline conflict probability Predicted time window The calculation yields:

[0368] Longitudinal time interval Vertical risk items ;

[0369] Lateral remaining safety distance Lateral collision time Horizontal risk items ;

[0370] Occlusion Correction ;

[0371] Final prediction of conflict probability Satisfying no greater than The constraints and requirements.

[0372] Step S4: Construction of multi-objective car-following optimization model:

[0373] A multi-objective car-following optimization model is constructed, with the three independent optimization objectives of minimizing vehicle driving energy consumption, maximizing section traffic efficiency, and minimizing safe distance deviation, forming a three-dimensional objective vector.

[0374] Energy consumption calculation model, travel time calculation model and safety distance deviation calculation model are constructed respectively. The comprehensive building shading influence coefficient is coupled into each objective function to reflect the differentiated impact of shading on energy consumption, traffic efficiency and safety requirements. The three objectives are coupled and conflict with each other. The non-dominated optimal solution set is solved by multi-objective optimization algorithm.

[0375] Step S5: Improve the NSGA-Ⅲ algorithm to solve for the optimal guiding parameters:

[0376] An improved NSGA-Ⅲ algorithm is used to solve the multi-objective car-following optimization model, obtaining the Pareto optimal solution set. The optimal compromise solution is selected and verified before being sent out for guidance.

[0377] The improved NSGA-Ⅲ algorithm execution process includes feasible region population initialization, occlusion level adaptive reference point generation, safety constraint adaptive penalty processing, occlusion fitness tournament selection, iteration termination and solution set output; among them, an upper limit is set on the proportion of safe target reference points, and when the upper limit is reached, the remaining target reference points are normalized and redistributed to ensure a reasonable distribution of reference points.

[0378] After outputting the Pareto optimal solution set, a normalized weighted comprehensive evaluation function is used to select the optimal compromise solution, with the weight coefficients dynamically adjusted according to the occlusion level. The selected guiding driving speed and guiding following distance are validated. If the validation passes, the information is sent to the vehicle terminal or navigation application through the 5G-V2X vehicle-road cooperative network to push safe speed and following distance prompts to the driver. If the validation fails, the immediate optimization parameters are recalculated.

[0379] Step S6, Hierarchical closed-loop re-optimization mechanism:

[0380] Establish a hierarchical closed-loop re-optimization mechanism based on the occlusion risk index to dynamically update guidance parameters and guidance strategies.

[0381] The system calculates a real-time occlusion risk index, with the following safety margin term processed as non-negative. The final risk index is normalized to 0–1 and divided into three risk levels: low, medium, and high. Differentiated triggering conditions and response priorities are matched according to the risk level, with immediate response for high-risk scenarios and periodic updates for low-risk scenarios. The update interval is dynamically adjusted and optimized based on the occlusion risk index to avoid frequent guidance interference with driver operation. The model weights are dynamically corrected through feedback of guidance parameter execution errors, and the model self-optimization is achieved by combining periodic offline calibration, forming a complete optimization closed loop of "monitoring-evaluation-optimization-guidance-correction".

[0382] Comparative verification of the effectiveness of occlusion quantification:

[0383] To verify the effectiveness of the building occupancy quantification method proposed in this invention, a simulation comparison test was conducted in a typical narrow road network scenario.

[0384] Experimental conditions:

[0385] Experiment type: Simulation experiment based on VISSIM and Python;

[0386] Test section: A narrow one-way single-lane side road with a total length of 320m, an original lane width of 3.5m, including a circular curve with a radius of 80m, and a longitudinal slope of ≤2%;

[0387] Building parameters: The height of buildings on both sides ranges from 6 to 24m. Three characteristic shading sections with light, moderate and heavy shading are set along the road, with shading duration lengths of 70m, 80m and 90m respectively.

[0388] Traffic scenario: Off-peak free-flow condition, test vehicle is a small passenger car, initial speed range 20-40km / h, a total of 12 vehicles following in sync;

[0389] Experimental setup: Each working condition was tested 30 times. The data in the table is the average of multiple tests.

[0390] Comparison benchmark (traditional scheme): a conventional following optimization scheme that uses a fixed speed limit and a fixed safe following distance, without correcting for the effective sight distance, effective passage width, and braking response time caused by building obstruction.

[0391] The traditional car-following optimization method without occlusion quantization and the car-following optimization method considering occlusion quantization of the present invention were tested in three occlusion levels. The core performance indicators are compared in the table below:

[0392]

[0393] The comparison results show that:

[0394] 1. Significantly improved safety accuracy: After introducing occlusion quantization, the relative error of safe distance decreased from a maximum of 15.2% to within 2.3%, the safety constraint satisfaction rate increased from a minimum of 63.5% to over 97.6%, and the risk rate of following-car conflict was greatly reduced. This verifies that occlusion quantization can effectively correct and optimize thresholds, match actual driving risks, and significantly improve following-car safety.

[0395] 2. Collaborative optimization of operational efficiency: Under the premise of ensuring safety, the optimization scheme considering occlusion quantification achieves two-way optimization of energy consumption and traffic efficiency. In the case of heavy occlusion, energy consumption is reduced by 9.1% and traffic efficiency is improved by 7.3%, avoiding the waste of traffic efficiency and unnecessary energy loss caused by the traditional "one-size-fits-all" optimization.

[0396] 3. Enhanced scene adaptability: As the occlusion level increases, the performance of traditional solutions degrades significantly, while the performance indicators of the solution of this invention remain stable, proving that the occlusion quantification mechanism can effectively adapt to different levels of building occlusion scenarios and maintain high accuracy and high reliability optimization effects even in high occlusion risk areas.

[0397] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0398] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0399] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0400] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0401] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A vehicle following optimization method for narrow road networks based on UAV building occlusion quantification, characterized in that, Includes the following steps: S1. Based on the dual-mode collaborative acquisition mechanism of global cruise and low-altitude follow-fly of UAVs, multi-dimensional perception data is collected in real time in narrow road network sections. After preprocessing, standardized scene data is obtained, including target vehicle information, global traffic status information, road environment and building occlusion parameters, and vehicle interaction data in narrow road network sections. S2. Calculate the reduction in viewing distance due to occlusion and the comprehensive building occlusion impact coefficient based on standardized scene data; The effective sight distance, effective passage width, and braking response time are corrected based on the reduction in sight distance due to occlusion and the comprehensive building occlusion impact coefficient. If the corrected effective sight distance is lower than the preset minimum safe sight distance, or the corrected effective passage width is lower than the preset minimum safe passage width, or the lateral safety margin is lower than the preset threshold, the subsequent optimization steps are terminated and a safety response command is triggered; otherwise, step S3 is executed. S3. Construct a multi-objective car-following optimization model, with guiding driving speed and guiding following distance as decision variables; The multi-objective car-following optimization model includes multi-dimensional car-following safety constraints and a three-dimensional objective vector; The multi-dimensional car-following safety constraints are constructed based on the comprehensive building obstruction influence coefficient, the corrected braking response time, and the effective passage width, including road condition constraints, vehicle dynamics constraints, and safety interaction constraints. The three-dimensional target vector consists of three sub-objectives: lowest energy consumption, highest traffic efficiency, and optimal safe distance, and each objective function is coupled with a comprehensive building occlusion influence coefficient. S4. The improved NSGA-Ⅲ algorithm is used to solve the multi-objective car-following optimization model to obtain the Pareto optimal solution set. Based on the comprehensive evaluation function, the optimal guidance parameter set is selected from the Pareto optimal solution set. After validity verification, the parameter set is sent to the vehicle guidance device of the target vehicle through the vehicle-road cooperative network to push safe speed and following distance prompts to the driver. S5. Establish a hierarchical closed-loop re-optimization mechanism based on the occlusion risk index: Calculate the normalized occlusion risk index to classify risk levels. When the preset hierarchical triggering conditions are met, the corresponding priority re-optimization process is triggered. The weights of the comprehensive objective function are periodically adjusted based on the spacing error.

2. The method for optimizing vehicle following in narrow road networks based on UAV building occlusion quantification according to claim 1, characterized in that, The specific content of the target vehicle information, global traffic status information, road environment and building occlusion parameters, and vehicle interaction data for the narrow road network section is as follows: The target vehicle information includes the following parameters of the guided target vehicle traveling in a narrow road network section with obstructing buildings: the vehicle's measured speed. Vehicle width Vehicle braking performance parameters, vehicle design maximum speed Rated maximum acceleration and rated maximum deceleration The vehicle braking performance parameters include the standard driver braking response time. and the vehicle's maximum braking deceleration ; The global traffic status information includes the following parameters for narrow road network sections: total length of the narrow road network section. Traffic flow Vehicle density Maximum traffic density of the section Average driving speed Traffic congestion index Length of continuous obstruction The target vehicle speed obtained from coarse-grained identification The longitudinal distance between the target vehicle and the vehicle in front obtained by coarse-grained identification and the benchmark theoretical safe following distance ; The road environment and building obstruction parameters include the following parameters for narrow road network sections: theoretical lane width of the section. Road curvature Road slope Road surface adhesion coefficient Static obstacle position coordinates Minimum speed limit for the section Maximum speed limit for the section The shortest line of sight in the section theory Coordinates of the building's obstruction location Building height Effective lateral width of building obstruction Area of ​​obstructed view Total monitoring field of view area and the driver's line of sight upward angle ; The vehicle interaction data includes the following parameters in narrow road network sections: real-time longitudinal distance between the target vehicle and the vehicle in front. Speed ​​of the vehicle in front Lateral center distance between the target vehicle and vehicles in adjacent lanes Adjacent lane vehicle width Probability of interaction conflict with the current vehicle The It is calculated based on the relative position, relative speed and preset prediction time window of the target vehicle with the vehicle in front and the vehicles in adjacent lanes; The target vehicle information, global traffic status information, road environment and building occlusion parameters, and vehicle interaction data in the narrow road network section are standardized scene data that have undergone spatiotemporal coordinate calibration, noise reduction, and outlier removal.

3. The method for optimizing vehicle following in narrow road networks based on UAV building occlusion quantification according to claim 1, characterized in that, The specific details of the dual-mode collaborative data collection mechanism of global cruise and low-altitude follow-fly for the UAV are as follows: In global cruise mode, the drone covers narrow road network sections along a preset flight path, completing the collection of global traffic status information and road environment and building occlusion parameters; the preset flight path is set according to the effective monitoring area of ​​the drone, so that the coverage of the narrow road network section by the preset flight path is not less than a preset threshold. In low-altitude follow-fly mode, the drone maintains synchronous flight with the guided target vehicle, completing the collection of target vehicle information and vehicle interaction data; wherein, the relative position control error during follow-flying satisfies: In the formula, This refers to the error between the actual longitudinal position of the drone and its preset longitudinal tracking position relative to the target vehicle. This represents the error between the actual lateral position of the drone and the preset lateral following position. This represents the error between the actual flight altitude of the drone and the preset follow-up altitude. The preset maximum follow-fly position error threshold is used; In the global patrol mode, the drone prioritizes the detection of target vehicles. Intelligent switching: In the formula, For preset coefficients, This is a correction factor for the following distance risk item, and ; The target vehicle speed is obtained through coarse-grained identification in global cruise mode. This refers to the longitudinal distance between the target vehicle and the vehicle in front, obtained through coarse-grained identification in global cruise mode. This is the maximum speed limit for this section. The baseline theoretical safe following distance; when When the preset switching threshold is reached, the drone switches to the low-altitude follow-fly mode corresponding to the target vehicle; In low-altitude follow-fly mode, the actual vehicle speed is used. Real-time longitudinal distance between the target vehicle and the vehicle in front Replace the coarse-grained recognition and The drone recalculates the target vehicle monitoring priority; when the target vehicle leaves the obstructed area, or when the recalculated target vehicle monitoring priority is continuously lower than the preset exit threshold, the drone returns to global cruise mode.

4. The method for optimizing vehicle following in narrow road networks based on UAV building occlusion quantification according to claim 1, characterized in that, The calculation of the reduction in viewing distance due to occlusion and the comprehensive building occlusion impact coefficient based on standardized scene data is as follows: Calculate the reduction in sight distance due to occlusion : In the formula, This is the shortest sight distance in the section theory. For building height, The driver's line of sight angle; Calculate the line-of-sight occlusion coefficient : Calculate the lateral occlusion coefficient : In the formula, The effective lateral width for building shading. This refers to the theoretical lane width of the section. Calculate the longitudinal occlusion coefficient : In the formula, To cover the continuous length of the road section, This refers to the total length of the narrow road network section; Calculate the area shading coefficient : In the formula, To block the field of vision, Total monitored field of view area; Calculate the initial value of the comprehensive building shading impact coefficient : In the formula, These are the fusion weights for the line-of-sight occlusion coefficient, lateral occlusion coefficient, longitudinal occlusion coefficient, and area occlusion coefficient, respectively. ; The initial value of the comprehensive building shading impact coefficient is normalized and truncated to obtain the comprehensive building shading impact coefficient. .

5. The method for optimizing vehicle following in narrow road networks based on UAV building occlusion quantification according to claim 1, characterized in that, The correction of effective sight distance, effective passage width, and braking response time based on the reduction in sight distance due to occlusion and the comprehensive building occlusion impact coefficient is as follows: Based on the comprehensive building shading impact coefficient Corrected braking response time: In the formula, This is the standard driver braking response time when there is no obstruction. This is the corrected braking response time; The gain coefficient representing the effect of building obstruction on driver braking response time; Based on the reduction in viewing distance due to occlusion Corrected effective line of sight: In the formula, This is the shortest sight distance in the section theory. The traffic congestion index. This is the corrected effective line-of-sight distance; The coefficient representing the impact of traffic congestion on the effective sight distance correction. Based on the comprehensive building shading impact coefficient Adjust the effective passage width: In the formula, This refers to the theoretical lane width of the section. The reduction in passage width caused by building obstruction. The correction factor for the effective passage width due to building obstruction; For vehicle model correction items, This is a correction factor for vehicle width. The reference vehicle width; The target vehicle width, This is the corrected effective passage width.

6. The method for optimizing vehicle following in narrow road networks based on UAV building occlusion quantification according to claim 1, characterized in that, The specific multi-dimensional car-following safety constraints are as follows: Based on the corrected effective passage width and the comprehensive building shading impact coefficient, road condition constraints are constructed, including maximum allowable guiding speed constraints and slope constraints. The maximum allowable guiding speed constraint is: In the formula, To guide driving speed, To calculate the overall building shading impact coefficient, This is the corrected effective passage width. It is the acceleration due to gravity. The road surface adhesion coefficient, For road curvature; The correction factor for the maximum permissible guiding speed due to building obstruction; The slope constraint is: In the formula, To guide the rate of change of driving speed, Road slope; The correction factor for the slope constraint caused by building shading; The vehicle dynamics constraints include maximum speed constraints and acceleration constraints: In the formula, Design the vehicle for its maximum speed. For the rated maximum acceleration, This is the rated maximum deceleration; Based on the corrected braking response time and the comprehensive building occlusion impact coefficient, safety interaction constraints are constructed, including dynamic safe following distance constraints and predicted conflict probability constraints. The dynamic safe following distance constraint is: Among them, the occlusion correction theoretical safe following distance corresponding to the candidate guidance speed for: In the formula, To guide following distance, The corrected braking response time, This is the vehicle's maximum braking deceleration. This is the safe distance correction amount corresponding to traffic congestion. To correct for the occlusion safety distance, The baseline theoretical safe following distance; The constraint on the predicted conflict probability is: In the formula, Based on candidate guided driving speed Candidate following distance And the probability of vehicle interaction conflict predicted from the current motion state of neighboring vehicles; This represents the baseline maximum probability of vehicle interaction conflicts allowed under unobstructed conditions. This is a correction factor for the risk of occlusion.

7. The method for optimizing vehicle following in narrow road networks based on UAV building occlusion quantification according to claim 1, characterized in that, The specific three-dimensional target vector is as follows: Introducing a comprehensive building shading impact coefficient as an energy consumption penalty term, an energy consumption calculation model incorporating driving resistance, acceleration resistance, and slope resistance is constructed. The minimum energy consumption target is then expressed as: : In the formula, The equivalent energy consumption coefficient of driving resistance corresponding to the square of the guiding driving speed. The equivalent energy consumption coefficient for acceleration and deceleration corresponding to the rate of change of the guiding driving speed. The three coefficients are pre-calibrated based on the target vehicle type and are equivalent to the slope resistance energy consumption coefficient corresponding to the road slope term. To calculate the overall building shading impact coefficient, To guide the rate of change of driving speed, To guide driving speed; The calculation of travel time is differentiated between obstructed and unobstructed road sections. The travel speed on obstructed road sections is then adjusted based on a comprehensive building shading impact coefficient. The target for maximum traffic efficiency is then expressed as follows: : In the formula, Total travel time The total length of the narrow road network section, To cover the continuous length of the road section; The correction factor for the travel time target caused by building obstruction; To guide following distance Safe following distance with occlusion correction theory Taking the minimum deviation as the objective, and introducing a comprehensive building shading influence coefficient as a safety weight, the optimal safety distance objective is expressed as: : In the formula, The theoretical safe following distance is determined by the occlusion correction for the candidate guidance speed; The weighting correction factor for building occupancy on the safe distance target; The overall objective vector is then expressed as: .

8. The method for optimizing vehicle following in narrow road networks based on UAV building occlusion quantification according to claim 6, characterized in that, Step S4 is as follows: S41. Based on multi-dimensional car-following safety constraints, determine the feasible range of values ​​for guiding driving speed and guiding following distance, and randomly generate an initial population within the feasible region; S42. Based on the comprehensive building occlusion influence coefficient, dynamically adjust the distribution of reference points in the 3D target space in the NSGA-Ⅲ algorithm: the larger the comprehensive building occlusion influence coefficient, the higher the proportion of reference points in the target dimension with the optimal safe distance. The larger the value, the smaller the proportion of reference points in the target dimension with the lowest energy consumption and highest traffic efficiency; an adaptive 3D reference point set is generated based on the adjusted proportion. ;in The calculation formula is: In the formula, The percentage of reference points representing the target with the optimal safe distance when there is no obstruction. The upper limit of the proportion of the target reference point at the optimal safe distance. The overall building shading impact coefficient; S43. Calculate the constraint violation degree of each individual in the current population regarding the dynamic safe following distance constraint. Apply adaptive penalties to individuals that violate both the dynamic safe following distance constraint and the lateral safe distance constraint. The penalty term is applied to the comprehensive target vector as components. Based on the penalized comprehensive target vector... Perform a non-dominated ranking on the individuals to obtain the non-dominated hierarchy of each individual: In the formula, To synthesize the target vector, The penalty gain coefficient, The degree of violation of safety constraints is determined by the ratio of the amount of constraint violation to the constraint threshold. The greater the overall building shading impact coefficient, the stronger the penalty for violating safety constraints; S44. Combine the non-dominated ranking results with the individual occlusion fitness to conduct tournament selection, and give priority to individuals with high non-dominated levels and high occlusion fitness to enter the offspring population. The formula for calculating the individual occlusion fit is: In the formula, For individual occlusion adaptation, The corresponding guiding following distance for each individual vehicle. The theoretical safe following distance is adjusted for occlusion for each individual vehicle; a higher fit indicates a better safety margin for the individual vehicle in the current occlusion scenario. S45. Normalize the target value of the candidate population formed by non-dominated sorting, and associate each candidate individual with the nearest adaptive 3D reference point. When the number of candidate individuals in the current non-dominated layer exceeds the remaining population capacity, perform survival selection based on reference point according to the number of individuals already associated with each reference point and the vertical distance of the candidate individual to the reference point, and prioritize retaining the candidate individuals corresponding to the reference point with the optimal target dimension of the safe distance. S46. When the number of iterations reaches the preset maximum value or the change in the optimal solution over multiple consecutive generations is less than the preset convergence threshold, the iteration is terminated and the complete Pareto optimal solution set is output. S47, Based on the comprehensive evaluation function Select the optimal compromise solution from the Pareto optimal solution set: In the formula, , , The weighting coefficients for the corresponding targets are dynamically adjusted based on the comprehensive building shading impact coefficient, and The weight of the target with the optimal safe distance increases progressively with the increase of the comprehensive building shading influence coefficient, while the weight of the target with the highest traffic efficiency decreases progressively with the increase of the shading level. , , This serves as a reference baseline value for the corresponding target; The Pareto solution with the minimum comprehensive evaluation function value is selected as the optimal set of guiding parameters; S48. Based on the current road conditions and obstruction status, the roadside unit verifies the validity of the optimal driving speed and optimal following distance in the optimal guidance parameter set. The speed validity condition is: in, To verify the target following speed, ; The dynamic permissible safe speed is calculated based on road curvature, pavement adhesion coefficient, effective traffic width, and building obstruction status, which is the right-hand value of the maximum permissible guiding speed constraint; To achieve the optimal driving speed, This is the minimum speed limit for the section. This is the maximum speed limit for this section. Average driving speed The speed verification amplification factor is the one corresponding to the average vehicle speed, and ; The spacing validity calibration condition is: in, To verify the target following distance, , For optimal following distance, Based on the basic safety distance verification coefficient, To obscure the correction factor for the safety distance check, This is the spacing verification coefficient corresponding to the sight distance, and ; This is the corrected effective line-of-sight distance; S49. If the validity verification passes, the verified guidance parameters are sent to the on-board guidance device of the target vehicle through the vehicle-road cooperative network to push safe speed and following distance prompts to the driver; if the verification fails, return to step S41.

9. The vehicle following optimization method for narrow road networks based on UAV building occlusion quantification according to claim 8, characterized in that, The hierarchical closed-loop re-optimization mechanism specifically includes: S51. Calculate the occlusion risk index at time t. : In the formula, , , , The weight coefficients are non-negative and satisfy the following conditions: And the weight of building obstruction Not lower than the preset occlusion weight lower limit ; The comprehensive building shading influence coefficient at time t; Let be the vehicle density at time t. This represents the maximum traffic density for the section. The real-time longitudinal distance between the target vehicle and the vehicle in front at time t; The baseline theoretical safe following distance; Let t be the actual measured speed of the vehicle at time t; The occlusion risk index is normalized, and the results are based on the normalized occlusion risk index. The values ​​are categorized into three risk levels: low risk. Medium risk High risk ,in , All are preset risk thresholds; S52. Match the corresponding triggering conditions and response priorities according to the risk level. All risk level re-optimizations return to step S2 for execution, the difference being the re-optimization start time limit: Level 1 High Priority Response: Triggered when any of the following conditions are met. Upon triggering, the current boot cycle is immediately interrupted, and the process returns to step S2 to perform immediate re-optimization and issue adjustment instructions: （1） ； (2) Change in building shading influence coefficient ; (3) A new static obstacle is detected, and the current vehicle interaction conflict probability is high. ; in This represents the maximum allowed collision probability under the current occlusion state. Level 2 Medium Priority Response: Triggered when any of the following conditions are met, and after triggering, the response will be re-optimized at a preset medium risk interval. Returning to step S2 completes the re-optimization, and the current safe boot parameters remain unchanged until new results are generated: （1） ； （2） ； (3) The percentage decrease in the corrected effective passage width exceeds the preset threshold; Level 3 low priority response: Satisfied However, if traffic flow fluctuations exceed a preset ratio and the real-time longitudinal distance between the target vehicle and the vehicle in front consistently fails to meet safety requirements, the dynamic minimum re-optimization interval will be triggered. In the next update cycle, return to step S2 for re-optimization; in All are threshold values ​​for changes in occlusion coefficient, and ; Minimum re-optimization interval period The calculation formula is: In the formula, This is the lower bound of the minimum allowable re-optimization interval period. This represents the maximum re-optimization interval under low-risk conditions. S53. Calculate the interval execution error of the current control cycle. : In the formula, This is the updated measured longitudinal following distance after the guidance parameters for the current control cycle are issued. The target following distance is issued after verification in the current control cycle; Based on the spacing execution error and the comprehensive building occlusion influence coefficient, the weight coefficients of the optimal safe distance target in the comprehensive evaluation function are corrected: In the formula, The weighting coefficients for the optimal target at the corrected safe distance; To correct the gain coefficient, the three weights are then normalized. After each fixed period, the weight coefficients of the target with the optimal safe distance in the comprehensive evaluation function are calibrated offline based on the execution error, forming a complete closed loop of monitoring-evaluation-optimization-execution-correction.

10. A vehicle following optimization system for narrow road networks based on UAV building occlusion quantification, characterized in that, It includes a processor, a memory, and a computer program stored in the memory. When the processor executes the computer program, it specifically performs the steps in the narrow road network vehicle following optimization method as described in any one of claims 1-9.

Images