Automatic driving recognition, signal timing and highway ramp control method

Abstract

The present application relates to the technical fields of automatic driving recognition and mixed traffic management, and particularly relates to an automatic driving recognition, signal machine timing and expressway ramp control method. The present application first acquires traffic monitoring data, and identifies the car-following behavior index and lane-changing behavior index of a vehicle through video image technology; calculates the car-following score of the vehicle based on the car-following behavior index, and if the car-following score is greater than the set car-following threshold, the vehicle is determined to be an automatic driving vehicle; if the car-following score is less than or equal to the car-following threshold, the lane-changing score of the vehicle is calculated based on the lane-changing behavior index; if the lane-changing score is greater than the set lane-changing threshold, the vehicle is determined to be an automatic driving vehicle; otherwise, the vehicle is determined to be a manually driving vehicle. The CAV recognition is based on the group behavior characteristics of the vehicle, does not need to rely on the vehicle-mounted equipment, is applicable to all vehicles, and improves the universality and accuracy of the CAV recognition; and overcomes the defects in the prior art that the CAV recognition relies on the vehicle-mounted equipment and is difficult to guarantee the recognition accuracy.

Description

Technical Field

[0001] This invention relates to the fields of autonomous driving recognition and hybrid traffic management technology, and in particular to an autonomous driving recognition, signal timing, and highway ramp control method. Background Technology

[0002] Urban traffic congestion is increasingly becoming a problem that restricts urban development and affects the quality of life for residents. As a key node in traffic flow, the efficiency of intersections directly determines the overall operation of the road network. Traffic signal control is a core technical means to improve intersection efficiency, and its development has mainly gone through stages such as fixed timing, inductive control, and in recent years, vehicle-road cooperative control.

[0003] Traditional fixed-time traffic signal allocation schemes rely on historical traffic flow data to set parameters such as cycles and green light ratios, lacking real-time performance and flexibility, and struggling to cope with dynamic changes in traffic flow. Inductive control, through the installation of detectors such as coils or radar, senses vehicle arrival information in real time, dynamically adjusting signal phase and duration, significantly improving responsiveness, and has become the mainstream application. However, both fixed-time and inductive control rely on the core logic of detecting the homogeneous unit of "vehicle," failing to distinguish the inherent differences in driving behavior characteristics. They provide a uniform right-of-way allocation strategy for all vehicles, failing to provide fine-grained control based on the driving characteristics of different vehicle types.

[0004] With the rapid development of autonomous driving technology, the composition of vehicles on the road is gradually transforming into a mixed traffic flow consisting of manually driven vehicles (HDVs) and autonomous vehicles (CAVs). Research shows that CAVs have unique advantages over HDVs in car-following and lane-changing behaviors. These differences mean that CAVs have the ability to drive in tighter formations, faster response times, and more precise coordination potential in response to changes in traffic light status.

[0005] Existing signal control systems fail to effectively recognize and utilize the advantages of CAVs. On the one hand, the inability to identify vehicle types prevents the provision of more efficient traffic strategies for CAV fleets; on the other hand, the system still needs to accommodate the conservative driving characteristics of HDVs, limiting further improvements in overall traffic efficiency. Furthermore, while solutions relying on dedicated communication equipment installed in vehicles can acquire CAV information, their low penetration rate, high retrofit costs, and communication reliability make large-scale deployment difficult in the short term.

[0006] The challenges of autonomous driving recognition also limit traffic optimization based on autonomous driving, such as traffic light timing and high-speed power ramp control. Summary of the Invention

[0007] To overcome the shortcomings of existing technologies where CAV recognition relies on in-vehicle equipment and is difficult to guarantee accuracy, this invention proposes an image recognition-based autonomous driving recognition method. This method recognizes CAV based on the group behavior characteristics of vehicles, does not rely on in-vehicle equipment, is applicable to all vehicles, and improves the universality and accuracy of CAV recognition.

[0008] The present invention proposes an autonomous driving identification method based on image recognition. First, traffic monitoring data is acquired, and the following behavior indicators and lane-changing behavior indicators of vehicles are identified through video image technology. The following behavior indicators are used to calculate the following score of the vehicle. If the following score is greater than the set following threshold, the vehicle is determined to be an autonomous driving vehicle.

[0009] If the car-following score is less than or equal to the car-following threshold, the vehicle's lane-changing score is calculated based on the lane-changing behavior index; if the lane-changing score is greater than the set lane-changing threshold, the vehicle is determined to be an autonomous vehicle; otherwise, the vehicle is determined to be a manually driven vehicle.

[0010] Preferred indicators of car-following behavior include: average vehicle spacing. Standard deviation of following distance Acceleration fluctuation rate and the average collision time Lane-changing behavior indicators include: turn signal usage rate before lane changing. lateral acceleration peak Lane changing safety distance ratio Number of lane change conflicts .

[0011] Preferably, the method for calculating the car-following score of a vehicle based on the car-following behavior index is as follows: first, normalize each car-following behavior index, and then perform a weighted summation of the normalized values ​​of each car-following behavior index.

[0012] The method for calculating a vehicle's lane-changing score based on lane-changing behavior indicators is as follows: normalize each lane-changing behavior indicator, and then sum the normalized values ​​of each lane-changing behavior indicator by weight.

[0013] The preferred method for normalizing the car-following index is as follows:

[0014] The normalized formulas for each car-following behavior indicator are as follows:

[0015] ;

[0016] ;

[0017] ;

[0018] ;

[0019] ;

[0020] in, , , and Average vehicle spacing Standard deviation of following distance Acceleration fluctuation rate and the average collision time The normalized value; To maintain a safe distance for vehicles, The set standard deviation threshold for following distance. The acceleration volatility threshold; To observe the vehicle's current speed, For vehicle driving reaction time; The maximum acceleration for vehicle deceleration. The speed of the car in front. , and All of these are empirical constants.

[0021] The preferred normalization formulas for each lane-changing behavior indicator are as follows:

[0022] ;

[0023] ;

[0024] ;

[0025] ;

[0026] in, , , and The usage rate of turn signals before lane change lateral acceleration peak Lane changing safety distance ratio Number of lane change conflicts The normalized value;

[0027] This is the baseline value for lateral acceleration; This is the conflict sensitivity coefficient; , and All of these are empirical constants.

[0028] The present invention proposes a traffic signal timing method, which first uses the image recognition-based autonomous driving identification method to identify autonomous vehicles, and when an autonomous vehicle is detected, switches the traffic signal timing mechanism from a non-mixed traffic flow timing mechanism to a mixed traffic flow timing mechanism.

[0029] The hybrid traffic flow timing mechanism constructs an objective function by minimizing the sum of squares of the maximum waiting time while maximizing vehicle speed. It sets constraints based on actual safety and physical conditions, solves the objective function by combining the constraints, and obtains the traffic light control signals and the acceleration commands for autonomous vehicles.

[0030] Preferably, the objective function J is:

[0031]

[0032] Where N is the prediction time step length, and i is the prediction time step. Let be the speed of vehicle u at the predicted time i|k, where k is the sampling time point and the predicted time i|k is the time step i after time k. For vehicles at traffic light j at time step i; This represents the maximum waiting time among all vehicles at time step i; and The weighting coefficients are set.

[0033] The constraints include:

[0034] Queue decision constraints:

[0035] ;

[0036] in, This represents the distance of vehicle u from the stop line at the predicted time i|k, where M and m are respectively... upper and lower boundaries, For a set small positive number; Let i be the queuing state of vehicle u at the predicted time i|k; =0 means no need to queue, and the opposite means you need to queue;

[0037] Vehicle collision constraints: If both the observed vehicle and the following vehicle are autonomous vehicles, and both are set to constant deceleration to prevent a collision when they reach the parking position; if at least one of the observed vehicle and the following vehicle is manually driven, the constraint formula is:

[0038] ;

[0039] Indicates the vehicle following vehicle u. , vehicles The distance from the parking line and the vehicle speed at the predicted time N|k; T is the sampling time interval; , These represent the distance and speed of vehicle u from the stop line at the predicted time N|k, respectively. Let u be the length of the vehicle.

[0040] Startup delay constraint:

[0041] In the same lane, the acceleration of the second vehicle during the subsequent manual restart delay of the vehicle in the adjacent lane. It remains at zero for 2 seconds;

[0042] If both the observed vehicle u and the following vehicle f(u) are manually driven, then at time... l =i,i+1,…,min{i+ On the range -1,N}, the acceleration of the observed vehicle u remains at 0.

[0043] Preferably, solving the objective function includes the following steps:

[0044] St51, represent the vehicle state and traffic light state in the next N steps as decision variables, including: , , and ; , and These are binary numbers representing the green, yellow, and red light states of traffic light j at prediction time i|k, where 1 indicates the light is on and 0 indicates the light is off.

[0045] St52. The original optimization problem, consisting of the objective function and constraints, is transformed into a standard mixed-integer quadratic programming problem. Then, the MIQP solver is used to obtain all predicted values ​​and state variables. Predicted values ​​include the velocity, acceleration, and distance to the parking line in state i|k. State variables include: , , and ;

[0046] St53, Extracting the signal light control solution , and And the acceleration commands for autonomous vehicles are executed.

[0047] The present invention proposes a traffic signal timing system based on automatic driving recognition, comprising a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the traffic signal timing method.

[0048] This invention proposes a highway ramp control method for controlling the steering actions of autonomous vehicles on highways. When the method is executed, the autonomous vehicle identification method based on image recognition is first used to identify autonomous vehicles in the traffic network. Then, the autonomous vehicle is used as an agent, and a machine learning algorithm is used to train the agent to generate the steering actions of the autonomous vehicle.

[0049] This invention proposes a highway ramp control method for controlling the steering actions of autonomous vehicles on highways. The steps are as follows: First, autonomous vehicles in the same traffic network are used as agents. Markov models are used to train the agents and generate the steering actions of the autonomous vehicles.

[0050] The state space F and action space A of the Markov model are set as follows:

[0051] F=[W] n ;

[0052] W=[v,y, l ];

[0053] ;

[0054] Where W represents the feature vector used to describe the vehicle state, n is the agent, and v, y, and l These represent the vehicle's speed, longitudinal coordinate in the world coordinate system, and current lane, respectively. This indicates changing lanes to the left. The driver will continue driving in the current lane. This indicates a lane change to the right.

[0055] Reward function of agent n for:

[0056] ;

[0057] ;

[0058] in, The set average speed bonus coefficient; This represents the total speed of vehicle n relative to the speed limit; The set speed weight; This is the penalty coefficient; Let be a binary collision weighting factor. When vehicle n collides, then... =1, otherwise =0; The value represents the distance between the front and rear of the vehicle, which is the ratio of the distance between the front of the vehicle and the rear of the vehicle to the speed of the vehicle behind. This refers to the safety factor for the distance between the vehicle's front and rear. This indicates a penalty point for changing lanes. Lane change factor; The number of all intelligent agents in the transportation network. Let n be the velocity of the agent. This is the maximum permissible speed.

[0059] Preferably, the method for obtaining the longitudinal coordinates in the world coordinate system is as follows: determine the coordinate mapping relationship from image coordinates to world coordinates based on the parameters of the camera device, then convert the image coordinates of the autonomous vehicle as an intelligent agent on the video frame into world coordinates through the coordinate mapping relationship, and extract the longitudinal coordinates in the world coordinate system.

[0060] The advantages of this invention are:

[0061] (1) The image recognition-based autonomous driving identification method proposed in this invention enables real-time and accurate identification of autonomous vehicles on the road without relying on onboard communication equipment. This invention utilizes existing road cameras to collect video data, eliminating the need for onboard communication equipment and avoiding the high cost of vehicle modification. By dynamically analyzing vehicle behavior characteristics through image recognition technology, it achieves accurate differentiation between autonomous vehicles and manually driven vehicles, and is compatible with mixed traffic scenarios with different penetration rates.

[0062] (2) This invention achieves efficient and accurate identification of autonomous vehicle modules, solves the problem of autonomous vehicle identification in mixed traffic flow, and breaks through the bottleneck of mixed traffic flow perception.

[0063] (3) The present invention proposes a signal timing method based on autonomous driving recognition. Based on autonomous driving recognition, it considers the group behavior characteristics of autonomous driving recognition and the mixing ratio of HDV, dynamically optimizes the signal timing strategy, fully explores the intersection passage potential in mixed traffic flow environment, and alleviates urban traffic congestion.

[0064] (4) This invention uses model-controlled prediction to dynamically optimize signal timing. Based on the mixed ratio of autonomous and manual driving, it differentiates vehicle dynamics constraints to minimize waiting time and maximize vehicle speed, generating adaptive signal cycles and green light intervals. By leveraging the collaborative potential of autonomous vehicles, it dynamically compresses green light intervals, reducing parking delays for convoys at intersections. It works in conjunction with existing intelligent transportation equipment, providing a foundation for advanced vehicle-road cooperative applications. Data fusion and feature extraction are completed on the road test server, reducing the load on the central system and improving response speed.

[0065] (5) This invention proposes a highway ramp control method based on autonomous driving recognition. This method dynamically adjusts the deflection control strategy of autonomous vehicles on highways based on recognition information to address the challenges of future mixed traffic, fully leveraging the efficiency and safety potential brought by autonomous driving technology. The multi-agent reinforcement learning component of this invention achieves dynamic and differentiated ramp control based on vehicle type, greatly improving control efficiency and traffic flow. The gate control strategy proposed in this invention optimizes the overall traffic flow on highways, synergistically improving mainline capacity and system safety. Attached Figure Description

[0066] Figure 1 This is a flowchart of an image recognition-based autonomous driving recognition method proposed in this invention;

[0067] Figure 2 This is a flowchart of the hybrid traffic flow timing mechanism proposed in this invention;

[0068] Figure 3 Flowchart of the method for solving the objective function in the timing mechanism of mixed traffic flow;

[0069] Figure 4 This is a schematic diagram illustrating the identification of CAV and HDV vehicles using video recognition as an example.

[0070] Figure 5 A schematic diagram illustrating the use of acquired traffic parameters for SUMO simulation;

[0071] Figure 6 This is a schematic diagram of a Markov model; where s represents the state. a Let r be the action, r be the reward function, and s' be the state at the next time step.

[0072] Figure 7 This is a schematic diagram of a scene at a traffic light, as shown in the example.

[0073] Figure 8 The signal light timing results are shown in the example. Detailed Implementation

[0074] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0075] like Figure 1 As shown, this embodiment proposes an image recognition-based autonomous driving recognition method, which includes the following steps.

[0076] S1. Acquire traffic monitoring data and identify vehicle following behavior indicators and lane changing behavior indicators through video image technology;

[0077] Car-following behavior indicators include: average vehicle spacing Standard deviation of following distance Acceleration fluctuation rate and the average collision time ;

[0078] Lane-changing behavior indicators include: turn signal usage rate before lane changing. lateral acceleration peak Lane changing safety distance ratio Number of lane change conflicts .

[0079] Lane change conflict refers to a situation where a vehicle changes lanes and the collision time is less than the set safe lane change time threshold. As long as a vehicle changes lanes, a lane change conflict is considered, regardless of whether the vehicle successfully changed lanes.

[0080] Average vehicle spacing , Let N' be the distance between the observed vehicle and the vehicle in front at frame t, and N' be the total number of frames of observed vehicle data. The observed vehicle data refers to the segments of traffic monitoring data containing the observed vehicle. Intelligent driving vehicles have high stability, and the average distance between vehicles is relatively stable, usually close to a fixed value.

[0081] Standard deviation of following distance The standard deviation of following distance is used to measure the degree of fluctuation in following distance. The smaller the standard deviation, the more stable the following behavior. The standard deviation of following distance of intelligent driving vehicles is often smaller than that of human driving vehicles.

[0082] Acceleration volatility ,in ; Let t be the acceleration of the vehicle observed in frame t. This represents the average acceleration of the observed vehicle data. The volatility of acceleration reflects the smoothness of acceleration and deceleration. Intelligent driving behavior is relatively stable, with generally low volatility, while manual driving often involves sudden acceleration or braking, resulting in higher volatility.

[0083] Mean collision time , This represents the collision time in the t-th frame of the observed vehicle's data. , Let t be the distance between the observed vehicle and the vehicle in front. Let t be the relative speed between the vehicle and the vehicle in front observed at frame t. , Let t be the speed of the vehicle observed in frame t. Let t represent the speed of the vehicle in front; the vehicle in front is the vehicle adjacent to and ahead of the observed vehicle. The higher the TTC (Time to Collision), the higher the safety, indicating that autonomous driving can usually maintain a high and relatively stable TTC.

[0084] Turn signal usage rate before lane change , The number of times a vehicle uses its turn signal to change lanes. The total number of lane changes for the vehicle indicates that the driving behavior of intelligent driving vehicles is highly standardized, strictly follows traffic rules, and has a high turn signal usage rate.

[0085] peak lateral acceleration , Let be the lateral acceleration in frame t, which can be calculated using the trajectory curvature, i.e. Higher lateral acceleration means a more aggressive lane-changing process. The lane-changing process of intelligent driving vehicles is precisely controlled, prioritizing safety and stability. Represents the velocity of frame t. This represents the curvature of the turning trajectory in frame t.

[0086] Lane changing safety distance ratio , and These represent the distances between the vehicle in front and behind in the target lane and the observed vehicle, respectively, when changing lanes. To observe the vehicle's current speed, the lane change safety distance ratio indicates the safe time for the vehicle to switch lanes with the vehicles in front and behind during the lane change process. Intelligent driving vehicles must meet the safety lane change requirements, and their safety lane change ratios are generally high, allowing ample time for lane changes; "min" indicates taking the minimum value during the lane change process.

[0087] Lane change conflict number This refers to the number of times a vehicle's minimum TTC (Time to Collision) with a vehicle in the target lane is less than a minimum threshold during lane changing.

[0088] ;

[0089] Where the parameter t is the time frame index, that is, the discrete time frame from the start of the track change to the end of the track change; The time frame at the start of the lane change. The time frame for the end of a lane change; TTC lead (t) represents the collision time between the observed vehicle and the vehicle in front at time frame t, TTC. lag (t) represents the collision time between the observed vehicle and the following vehicle at time frame t. For the safe collision time threshold, This represents an indicator function; it is 1 if the condition within the parentheses is true, and 0 otherwise.

[0090] ;

[0091] ;

[0092] t represents the t-th frame. (t) represents the observed vehicle speed in frame t. (t) represents the speed of the vehicle in front in frame t. (t) represents the speed of the following vehicle in frame t. (t) represents the distance between the observed vehicle and the vehicle in front on frame t. The distance between the observed vehicle and the following vehicle is t frames.

[0093] In practice, target detection algorithms such as the YOLOv series can be used to perform image recognition processing on camera frames, thereby automatically identifying CAV vehicles and obtaining behavioral characteristics, such as distance from the parking line, vehicle speed, and acceleration.

[0094] S2. Normalize the various car-following behavior indicators, including the average vehicle spacing. Standard deviation of following distance Acceleration fluctuation rate and the average collision time The normalized values ​​are denoted as follows: , , , ;

[0095] The normalized formulas for each car-following behavior indicator are as follows:

[0096] ;

[0097] ;

[0098] ;

[0099] ;

[0100] ;

[0101] in, To maintain a safe distance for vehicles, To observe the vehicle's current speed, For vehicle driving reaction time; The maximum acceleration for vehicle deceleration. The speed of the vehicle in front; The standard deviation of following distance. The standard deviation threshold for following distance. For acceleration volatility, For acceleration volatility threshold, This represents the average collision time. 3, 7, 10, and 1 are empirical constants.

[0102] S3. Calculate the vehicle's following score F1; determine if F1 is greater than the set following threshold. If yes, then determine that the vehicle is an autonomous vehicle; otherwise, proceed to step S4.

[0103] ;

[0104] in, , , and Average vehicle spacing Standard deviation of following distance Acceleration fluctuation rate and the average collision time The weights;

[0105] In practical implementation, the following settings can be configured: =0.3, =0.2, =0.2, =0.3.

[0106] S4. Normalize the various lane-changing behavior indicators, including the turn signal usage rate before lane changes. lateral acceleration peak Lane changing safety distance ratio Number of lane change conflicts The normalized values ​​are denoted as follows: , , and ;

[0107] The normalization formulas for each lane-changing behavior indicator are as follows:

[0108] ;

[0109] ;

[0110] ;

[0111] ;

[0112] in, To increase the usage rate of turn signals before changing lanes; The peak value of lateral acceleration. This is the baseline value for lateral acceleration; The safe distance ratio for lane changing; This is the conflict sensitivity coefficient, used to control the penalty intensity of the number of conflicts on the final score; 0.7, 0.8, and 1.5 are all empirical values.

[0113] S5. Calculate the lane-changing score F2 for the vehicle; determine if F2 is greater than the set lane-changing threshold. If yes, the vehicle is determined to be an autonomous vehicle; otherwise, it is determined to be a manually driven vehicle.

[0114] ;

[0115] in, , , and The usage rate of turn signals before lane change lateral acceleration peak Lane changing safety distance ratio Number of lane change conflicts The weight.

[0116] In practical implementation, the following settings can be configured: =0.3, , =0.2, .

[0117] Follow-up threshold and lane change threshold These are all empirical values. For example, in subsequent experimental simulations, they are set to 0.8. In actual implementation, they can also be set to other values ​​according to the traffic environment. The specific value range can be set to [0.5, 1).

[0118] This embodiment also provides a traffic signal timing method based on autonomous driving recognition. The traffic indication system is configured with two sets of traffic signal timing mechanisms. The first set of timing mechanisms is suitable for traffic flow with purely manual driving, and the second set of timing mechanisms is suitable for mixed traffic flow with both manual and autonomous driving. The traffic indication system uses the first set of timing mechanisms as the default working state, and can specifically use existing traffic timing mechanisms. The traffic indication system uses the above-mentioned image recognition-based autonomous driving recognition method to detect autonomous vehicles. When an autonomous vehicle is detected, it switches to the second set of timing mechanisms, i.e., the mixed traffic flow timing mechanism.

[0119] Reference Figure 2 , Figure 3 The second timing mechanism (mixed traffic flow timing mechanism) is as follows:

[0120] Step 1: Initialize the MPC (Model Predictive Control) parameters, setting the maximum number of steps N and the sampling time interval T; configure the minimum signal times for traffic lights (yellow, green, and red), denoted as follows: , and Vehicle dynamics constraints are defined as including: maximum speed. Maximum acceleration , minimum acceleration Set the restart delay time for manually driven vehicles. ;

[0121] The purpose of this method is to predict traffic information at time points k+1, k+2, ..., k+N based on observation information at sampling time point k (hereinafter referred to as time k). The time value at sampling time point k is the sum of the start time and k×T, and the time values ​​at time points k+1, k+2, and k+N are the sums of the start time and (k+1)×T, the start time and (k+2)×T, and the start time and (k+N)×T, respectively.

[0122] St2. At the intersection, for each sampling time point k=1, 2, ..., vehicle status information is obtained through monitoring cameras, including: the distance of vehicle u from the stop line. ,speed And the vehicle driving type (whether it is an autonomous vehicle), update the lane vehicle set. , For vehicles that arrive at traffic light j at time k, Let k be the vehicle that leaves traffic light j. Let K be the set of vehicles at traffic light j at time k.

[0123] St3. Construct the objective function by minimizing the sum of squares of the maximum waiting time while maximizing the vehicle speed, and set constraints based on actual safety and physical conditions.

[0124] The objective function J is defined as follows:

[0125] ;

[0126] Where N is the prediction time step length, and i is the prediction time step. Let i be the speed of vehicle u at the predicted time i|k; For vehicles at traffic light j at time step i; This represents the maximum waiting time among all vehicles at time step i; and The weighting coefficients are set. The prediction time i|k is the time i after time k. Assuming the starting time is T0, the time value of sampling time point k is T0+kT, and the time value of prediction time i|k is T0+(k+i)T.

[0127] The constraints include: traffic light control constraints, vehicle dynamics constraints, traffic light behavior constraints, vehicle collision constraints, and start-up delay constraints.

[0128] Traffic light control constraints include:

[0129] Constraint 1: Each traffic light displays only one signal at any given time.

[0130] Constraint 2: Signals in conflicting directions cannot be green or yellow simultaneously.

[0131] Constraint 3: Signal switching sequence is restricted. A green light cannot directly switch to a red light, a red light cannot directly switch to a yellow light, and a yellow light cannot directly switch to a green light.

[0132] Constraint 4: Yellow, green, and red lights must meet the minimum duration requirement, defined as follows: , , Let i|k represent the durations of the yellow, green, and red lights at traffic light j at prediction time i. The duration constraints for different colored traffic lights are as follows:

[0133] ; ;

[0134] ; ;

[0135] ; ;

[0136] in, To predict the duration of the yellow light on traffic light j at time i+1|k, To predict the duration of the yellow light on traffic light j at time i|k; The minimum duration of the yellow light for traffic light j;

[0137] To predict the duration of the green light on traffic light j at time i+1|k, To predict the duration of the green light on traffic light j at time i|k; The minimum green light duration for signal light j;

[0138] To predict the duration of the red light on traffic light j at time i+1|k, To predict the duration of the red light on traffic light j at time i|k; The minimum duration of the red light for traffic light j;

[0139] , , , , and All are binary numbers;

[0140] If the traffic light j is yellow at the predicted time i|k, then =1, =0, =0; when the traffic light j is green at the predicted time i|k, then =0, =1, =0; when the traffic light j is red at the predicted time i|k, then =0, =0, =1;

[0141] If the traffic light j is yellow at the predicted time i+1|k, then =1, =0, =0; when the traffic light j is green at the predicted time i+1|k, then =0, =1, =0; when the traffic light j is red at the predicted time i+1|k, then =0, =0, =1.

[0142] Vehicle dynamics constraints:

[0143] Constraint 1: Velocity and acceleration are within a certain range, that is:

[0144] ;

[0145] ;

[0146] in, and These are the upper limits of vehicle u's speed and acceleration, respectively; This is the lower limit of the acceleration of vehicle u, which is less than 0, and is essentially the maximum deceleration; and Let i and k be the velocity and acceleration of vehicle u at the predicted time i|k, respectively.

[0147] Constraint 2: The updates to the vehicle's position and velocity conform to the kinematic equations:

[0148] ;

[0149] ;

[0150] in, Let be the speed of vehicle u at the predicted time i+1|k, where the predicted time i+1|k is time i+1 steps after time k. Let i be the speed of vehicle u at the predicted time i|k; The acceleration of vehicle u at the predicted time i|k; Let be the distance of vehicle u from the stop line at the predicted time i+1|k. Let be the distance between vehicle u and the stop line at the predicted time i|k; T is the sampling time interval, i.e., the cycle length of the traffic light.

[0151] Behavioral constraints under traffic lights:

[0152] Constraint 1: Vehicles that can stop under a yellow light must stop, meaning that vehicles can stop before crossing the intersection and pass through with constant-force braking. When the traffic light is currently yellow, the following constraint must be followed:

[0153] ;

[0154] ;

[0155] ;

[0156] ;

[0157] in, It is the set of non-negative integers. To set a positive integer; The distance of vehicle u at the traffic light from the stop line at time k. Let be the speed of vehicle u at the traffic light at time k; Let N|k be the distance between vehicle u and the stop line at the predicted time. Let N be the vehicle speed of vehicle u at the predicted time N|k; the predicted time N|k is the time N steps after time k; D is the setting constant for the relaxation of the control logic conditions. It is a binary number; =1 indicates that traffic light j is yellow at time k. =0 indicates that traffic light j is not yellow at time k; It is a positive integer; Let be the minimum acceleration of vehicle u; To set a constant, a sufficiently large value is chosen based on experience to ensure that... When =0, the above equation holds true;

[0158] Constraint 2: Vehicles in a queue must not proceed when the traffic light is red.

[0159] ;

[0160] Let i be the queuing state of vehicle u at the predicted time i|k; =0 means no need to queue, and the opposite means you need to queue;

[0161] Let i be the queuing state of vehicle u at the predicted time i+1|k; =0 means no need to queue, and the opposite means you need to queue;

[0162] =1 indicates that the traffic light j is red when vehicle u arrives at the predicted time i|k. =0 indicates that the traffic light j that vehicle u arrives at at the predicted time i|k is not red.

[0163] Queue decision constraint: Determine whether vehicle u has not yet passed the traffic light at the predicted time i|k; the predicted time i|k is the time i extended from time k; define M and m as... The upper and lower limits of the value, For a given infinitesimal positive number, the constraints are expressed as follows:

[0164] ;

[0165] ;

[0166] This represents the distance of vehicle u from the stop line at the predicted time i|k; Let {0, 1} be the binary variable representing the parking state of vehicle u at the predicted time i|k. A value of 1 indicates that vehicle v has not yet passed the traffic light at the predicted time i|k, while a value of 0 indicates that vehicle v has completely passed the traffic light at the predicted time i|k; that is, Let i be the queuing state of vehicle u at the predicted time i|k; =0 means no need to queue, and the opposite means you need to queue.

[0167] Vehicle collision constraints:

[0168] Constraint 1: When both vehicles are CAVs (Autonomous Vehicles), both vehicles choose constant deceleration to avoid a collision at the stopping position; specifically, this can be set by optimizing variables. and Choose a constant deceleration for each vehicle, such that the two vehicles maintain a safe distance after stopping while satisfying the minimum deceleration constraint; the formula is expressed as:

[0169] ;

[0170] ;

[0171] ;

[0172] ;

[0173] in, Let f(u) and f(u) represent the discrete time steps required for vehicles u and f(u) to come to a complete stop starting from time N|k, respectively. Let u be the length of the vehicle. Indicates the vehicle following vehicle u. , vehicles The distance from the stop line and the vehicle speed at the predicted time N|k; , These represent the distance and speed of vehicle u from the stop line at the predicted time N|k, respectively.

[0174] Constraint 2: If one or more of the two vehicles are HDVs (manually driven vehicles), a worst-case scenario must be included to ensure that a collision can be avoided even if the leading vehicle brakes at maximum force.

[0175] ;

[0176] Startup delay constraint:

[0177] Constraint 1: HDVs stopped in the queue will cause restart latency, leading to a decrease in intersection performance. That is, if two vehicles are at zero speed, the acceleration of the following vehicle will be delayed. Keep it at zero for seconds. Delay for restarting manually driven vehicles: The formula is as follows:

[0178] ;

[0179] ;

[0180] Indicates the vehicle following vehicle u; Indicates that vehicle u is at the predicted time. The speed of the car on the road, Indicates vehicle At the predicted time The speed of the vehicle on the road; They are vehicles u, Maximum vehicle speed; It is a binary number, at the prediction time. Vehicle u and vehicle If all predicted states are stopped, then =0, otherwise =1; Represents a small constant;

[0181] =0, and vehicle u and vehicle When all driving is done manually, constraint 2 must also be satisfied: acceleration must be forced to zero, expressed by the formula:

[0182] ;

[0183] Acceleration forced zeroing constraint (i.e., the above formula) effective time l =i,i+1,…,min{i+ {-1, N}, where a is the acceleration, and the other parameters are the same as below; This indicates that the vehicle following vehicle u is at the predicted time. acceleration on the surface, This represents the minimum acceleration of the vehicle following vehicle u. This represents the maximum acceleration of the vehicle following vehicle u; This refers to the restart delay time, which is the reaction time of the human driver in manual driving.

[0184] Step 4: Based on the identified vehicle type, switch the constraints between manually driven and autonomous vehicles. That is: if vehicle u is a manually driven vehicle, execute: traffic light control constraints, vehicle dynamics constraints, traffic light behavior constraints, vehicle collision constraint 2, and start-up delay constraints;

[0185] If vehicle u is an autonomous vehicle, execute: traffic light control constraints, vehicle dynamics constraints, traffic light behavior constraints, and vehicle collision constraints 1;

[0186] Step 5: Optimize intersection traffic conditions by minimizing the objective function, and set traffic light control signals and CAV acceleration commands; the specific steps are as follows:

[0187] St51, represent the vehicle state and traffic light state in the next N steps as decision variables.

[0188] St52, transform the original optimization problem consisting of the objective function and constraints into a standard mixed-integer quadratic programming (MIQP) problem.

[0189] St53. Use a MIQP solver, such as the Gurobi solver, to calculate the solution to the problem; obtain all predicted values ​​and state variables; predicted values ​​include velocity, acceleration, and distance to the parking line in state i|k; state variables include: , , , .

[0190] St54, Extracting the signal light control solution , , And the acceleration command for the CAV vehicle is issued and executed.

[0191] This embodiment also proposes a highway ramp control method based on autonomous driving recognition, used to control the steering actions of autonomous vehicles (CAVs) on highways.

[0192] In this implementation method, a CAV vehicle is used as the intelligent agent, and a Markov model is employed to generate steering actions. Specifically, the state space F and action space A of the Markov model are set as follows:

[0193] F=[W] n ;

[0194] W=[v,y, l ];

[0195] ;

[0196] Where W represents the feature vector used to describe the vehicle state, n is the agent, and v, y, and l These represent the vehicle's speed, longitudinal coordinate in the world coordinate system, and current lane, respectively. This indicates changing lanes to the left. The driver will continue driving in the current lane. This indicates a lane change to the right.

[0197] The method for obtaining the vertical coordinates in the world coordinate system is as follows: determine the coordinate mapping relationship from image coordinates to world coordinates based on the parameters of the camera device, then convert the image coordinates of the CAV vehicle, which acts as an intelligent agent, on the video frame into world coordinates through the coordinate mapping relationship, and extract the vertical coordinates in the world coordinate system.

[0198] Specifically, the feature vector of agent n can be denoted as W(n) = {v n ,y n , l n The action can be recorded as} ; where v n y n , l n These represent the speed, longitudinal coordinate in the world coordinate system, and current lane of CAV vehicle n, respectively. These represent CAV vehicle n changing lanes to the left, not changing lanes, and changing lanes to the right, respectively.

[0199] The reward function is set as follows: In multi-lane ramp merging scenarios, the main objective is to improve the merging efficiency of the entire traffic network, ensure safety, and penalize frequent lane changes. The reward function formula is:

[0200] ;

[0201] ;

[0202] in, The average speed reward coefficient is set to encourage driving behaviors that improve CAV system efficiency, and represents the difference between the system's target speed and the actual speed. This represents the total speed of vehicle n relative to the speed limit; The set speed weight;

[0203] A penalty coefficient to ensure system security is not affected by conflicts; Let be a binary collision weighting factor. When vehicle n collides, then... =1, otherwise =0; By adding a penalty mechanism, the system prioritizes security and strives to reduce the occurrence of conflicts, thereby ensuring safer and more reliable operation;

[0204] The value represents the distance between the front and rear of the vehicle. Safety factor for vehicle frontage; vehicle frontage The formula for calculating is the ratio of the distance between the front of the vehicle and the rear of the vehicle to the speed of the vehicle behind.

[0205] This indicates penalties for changing lanes. Frequent lane changes can lead to unnecessary rapid acceleration and deceleration of many vehicles; a lane-change factor has been incorporated into the reward system. It encourages autonomous vehicles to minimize lane changes during training to reduce acceleration and deceleration.

[0206] Penalty coefficient Distance between the front and rear of the vehicle Lane change penalty points All are set to negative values.

[0207] The number of all intelligent agents in the transportation network (i.e., the number of CAV vehicles); Let n be the velocity of the agent. The maximum permissible speed;

[0208] Specifically, it can be set to 20m / s. Set it to 0.1.

[0209] like Figure 4 As shown, multi-view cameras deployed at ramp entrances and main road merging areas are used to acquire traffic videos on the highway. The YOLO-v7 object detection algorithm is then used to dynamically track vehicles and extract their positions. ),speed Parameters such as these are calculated for each vehicle using camera measurement technology. With the car behind actual distance ,judge Whether the vehicle is driven by a human or not, the identified data is transmitted to the Sumo software to construct the scenario.

[0210] Using SUMO software to construct highway ramp scenarios, such as Figure 5 As shown, the main road length at the ramp entrance is 200 meters, the merging lane entrance section is 120 meters long, and the ramp area is 50 meters long. In a multi-lane environment, the main road has three lanes.

[0211] Multi-agent reinforcement learning training is performed in the SUMO environment. After convergence, multi-agent decisions can be made to control highway gates, as shown in Table 1.

[0212] Table 1: Steering strategies of each intelligent agent (autonomous vehicle)

[0213] ;

[0214] This verification demonstrates the effectiveness of the ramp control method proposed in this invention.

[0215] The above signal timing method is verified in conjunction with specific embodiments below.

[0216] In this embodiment, system deployment includes the following steps:

[0217] (1) Hardware deployment: The perception layer includes electronic police cameras and checkpoint cameras. The computing layer consists of edge servers. The control layer includes intelligent traffic signal controllers and industrial Ethernet switches. The communication layer includes 5G RSUs and vehicle-mounted OBUs.

[0218] (2) Software deployment: YOLOv5 and DeepSORT were used for vehicle tracking, GurobiOptimizer was used for MPC optimization, DDS architecture was used for communication middleware, and C-V2X ASN.1 codec and SPAT / MAP message generator were used for the protocol stack.

[0219] In this embodiment, traffic lights at a certain intersection need to be controlled. The origin is the intersection of the lane and the median strip of the road, the X-axis is the lane, and the Y-axis is the road median strip. Four reference points are selected to solve for the C-coefficient: the lane start point, the lane end point, the lane left boundary reference point, and the lane right boundary reference point. The correspondence between their world coordinates (X,Y) and image coordinates (u,v) is shown in the table below:

[0220] ;

[0221] According to the expression

[0222] ;

[0223] ;

[0224] Substituting the four control points, we obtain eight equations, and finally solve for the C coefficient matrix:

[0225] C= ;

[0226] The pixel coordinates of the image can then be used to solve for the real-world coordinates using a transformation expression.

[0227] In this implementation example, after obtaining the vehicle's true coordinates through the above transformation process, the vehicle's speed, distance, acceleration, and other states are calculated. This allows for the calculation of various characteristic indicators related to the vehicle's car-following and lane-changing maneuvers. A weighted summation of the comprehensive score determines whether the vehicle is an autonomous vehicle. A table of judgment indicator scores for some vehicles is provided below.

[0228] Table 2 Vehicle Identification Data in Target Video

[0229] ;

[0230] Taking vehicle 1 as an example, the normalization calculation process for its four indicators is as follows:

[0231] 1.00;

[0232] ;

[0233] ;

[0234] ;

[0235] The overall score is calculated as follows:

[0236] 0.8596;

[0237] Based on its overall score being greater than the threshold of 0.8, it is determined to be a CAV vehicle.

[0238] Taking vehicle 5 as an example, the normalization calculation process for its four indicators is as follows:

[0239] ;

[0240] ;

[0241] ;

[0242] ;

[0243] The overall score is calculated as follows:

[0244] 0.4746;

[0245] Since its overall score is less than the threshold of 0.8, it is judged as HDV. At this point, it is necessary to further calculate the lane change score F2. The normalization process of its four lane change indicators is as follows:

[0246] ;

[0247] ;

[0248] ;

[0249] ;

[0250] The overall scores are calculated as follows:

[0251] 0.2187;

[0252] Based on its overall score being less than the threshold of 0.8, it is determined to be HDV.

[0253] Table 3: Judgment Results Based on the Carrying Indicator

[0254] ;

[0255] In this implementation example, consider the following: Figure 7 The traffic scenario shown depicts a main crossroads, consisting of east-west arterial roads and north-south secondary roads. A southwest-extending road connects to a roundabout, which in turn connects two entrances. Five traffic lights are installed. Figure 7 The diagram shows TL1, TL2, TL3, TL4, and TL5. TL1 controls the east entrance for eastbound straight traffic, TL2 controls the west entrance for westbound straight traffic, TL3 controls the north entrance for northbound straight traffic, TL4 controls the southwest entrance from the roundabout to the southwest of the intersection, and TL5 controls the southeast entrance from the roundabout to the southeast of the intersection. It's important to note that traffic lights TL4 and TL5 are mounted on the same signal pole (physically overlapping), so in later diagrams, TL4 and TL5 appear to be at the same point, but logically they control different traffic flows.

[0256] The parameters are set as follows: sampling time is 1 second, prediction time domain is 20 steps, minimum yellow light duration is 3 seconds, maximum vehicle speed is 12 m / s, and maximum acceleration is 3 m / s. The minimum acceleration is -3m / The HDV restart delay is 3 seconds. Vehicle states are initialized at k=0, with a total of 20 vehicles randomly distributed across 5 queues. The CAV penetration rate is 40%. The initial vehicle states can be obtained from the video data, and it is assumed that the vehicle arrival rate follows a Poisson distribution. The arrival rates at each traffic light (the ratio of vehicles at the traffic light to the total number of vehicles) are shown in Table 4 below:

[0257] Table 4: Arrival Rate at Traffic Lights

[0258] ;

[0259] As can be seen from the table above, the vehicle arrival rate at each traffic light is relatively even, indicating that the present invention effectively alleviates traffic congestion.

[0260] Based on the above conditions, construct an MPC optimization problem and define decision variables and signal control. , , Following step St3, constraints are applied to construct an objective function that minimizes waiting time and maximizes vehicle speed. The MIQP problem is then solved using the Gurobi solver. The final signal light phase results for the first 60 seconds of each time interval are as follows: Figure 8 As shown.

[0261] Of course, those skilled in the art will recognize that the present invention is not limited to the details of the exemplary embodiments described above, but also includes the same or similar structures that can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0262] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

[0263] The technologies, shapes, and structures not described in detail in this invention are all known technologies.

Claims

1. An image recognition-based automatic driving recognition method, characterized by, Firstly, traffic monitoring data is acquired, and a following behavior index and a lane changing behavior index of a vehicle are identified through video image technology; a following score of the vehicle is calculated based on the following behavior index, and if the following score is greater than a set following threshold, the vehicle is determined as an autonomous vehicle; if the following score is less than or equal to the following threshold, a lane changing score of the vehicle is calculated based on the lane changing behavior index; if the lane changing score is greater than a set lane changing threshold, the vehicle is determined as an autonomous vehicle; otherwise, the vehicle is determined as a manual driving vehicle. 2.The image recognition-based automatic driving recognition method of claim 1, wherein, The follow-up behavior indicators include: average inter-vehicle distance , follow-up distance standard deviation , acceleration fluctuation rate , and collision time mean ; The lane changing behavior indicators include: the use rate of the turn signal before lane changing , lateral acceleration peak , lane changing safety distance ratio , and lane changing conflict times . 3.The image recognition-based automatic driving recognition method of claim 2, wherein, The way of calculating the following score of the vehicle based on the following behavior index is that: firstly, each following behavior index is normalized, and then the normalized values of each following behavior index are weighted and summed up; The way of calculating the lane changing score of the vehicle based on the lane changing behavior index is that: each lane changing behavior index is normalized, and then the normalized values of each lane changing behavior index are weighted and summed up. 4.The image recognition-based automatic driving recognition method of claim 3, wherein, The following index normalization processing mode is as follows: The normalization formula of each following behavior index is as follows: wherein, , , and are normalized values of average inter-vehicle distance , standard deviation of following distance , acceleration fluctuation rate and mean collision time , respectively; is a safe distance of vehicle driving, is a set threshold value of standard deviation of following distance, is a threshold value of acceleration fluctuation rate; is a current speed of an observed vehicle, is a reaction time of vehicle driving; is a maximum acceleration of vehicle deceleration, is a speed of a preceding vehicle, , and are all empirical constants. 5.The image recognition-based automatic driving recognition method of claim 3, wherein, The normalization formula of each lane changing behavior index is as follows: wherein, , , and are normalized values of the turn signal usage rate , the lateral acceleration peak value , the lane-changing safety distance ratio , and the number of lane-changing conflicts , respectively. is a reference value for lateral acceleration; is a conflict sensitivity coefficient; , and are empirical constants.

6. A signal timing method employing the automatic driving recognition method based on image recognition according to any one of claims 1 to 5, characterized by Firstly, the autonomous vehicle is identified by using the image recognition-based autonomous driving identification method of any one of claims 1-5, and when the autonomous vehicle is detected, the signal light timing mechanism is switched from a non-mixed traffic flow timing mechanism to a mixed traffic flow timing mechanism; The mixed traffic flow timing mechanism constructs a target function by minimizing the square sum of the maximum waiting time and maximizing the vehicle speed, sets constraints according to actual safety conditions and physical conditions, solves the target function combined with the constraints, and obtains a signal light control signal and an autonomous vehicle acceleration instruction.

7. The method of claim 6, wherein the signal timing is determined by: ###0002### The target function J is: where N is the prediction time step length, i is the prediction time step, is the speed of vehicle u at prediction time i | k, k is the sampling time point, and prediction time i | k is the time step i ahead of time k; is the vehicle at signal light j at time step i; represents the maximum waiting time among all vehicles at time step i; and is the set weight coefficient; The constraints include: Queue judgment constraint: ； wherein, denotes the distance of the vehicle u to the stop line at the prediction time i | k, M and m are the upper and lower bounds of the distance of the vehicle u to the stop line at the prediction time i | k, M and m are the upper and lower bounds of the is a small positive number set; is the queuing status of the vehicle u at the prediction time i | k; = 0 means no queuing, otherwise means queuing. Vehicle collision constraint: if the observed vehicle and the rear vehicle are both autonomous vehicles, both vehicles are set to constant deceleration so that the two vehicles do not collide at the stop position; if at least one of the observed vehicle and the rear vehicle is a manual driving vehicle, the constraint formula is: ； denotes the rear vehicle of vehicle u, , denotes the distance and the speed of vehicle u at the prediction time N|k, respectively; denotes the distance and the speed of vehicle u at the prediction time N|k, respectively; , denotes the distance and the speed of vehicle u at the prediction time N|k, respectively; denotes the length of vehicle u; Start-up delay constraint: The acceleration of the second vehicle in the same lane remains zero for the next manual driving vehicle restart delay seconds. If both the observing vehicle u and the following vehicle f(u) are manually driven, then the observing vehicle u acceleration is kept at 0 on the time instants l = i, i+1,..., min{i+1, N}. = i, i+1,..., min{i+1, N}.

8. The traffic signal timing method as recited in claim 7, wherein, The solution of the target function includes the following steps: St51, the vehicle state and signal light state of future N steps are expressed as decision variables, including: , , and ; , and are binary numbers representing the green light state, yellow light state and red light state of signal light j at prediction time i|k, 1 represents on, and 0 represents off; is a binary number, and when the predicted states of vehicle u and the following vehicle at prediction time are both stop, then =0, otherwise =1; St52, the original optimization problem composed of objective function and constraints is converted into a standard mixed integer quadratic programming problem, and then solved by using a MIQP solver to obtain all the predicted values and state quantities; the predicted values include the speed, acceleration, and stopping line distance at state i|k; the state quantities include: , , and ; St53, extracting signal light control solution 、 and and acceleration instructions of the autonomous vehicle and executes.

9. A signal timing system based on automatic driving recognition, characterized by, A memory and a processor are included, the memory stores a computer program, and the processor is used to execute the computer program to realize the signal machine timing method of claim 6 or 7 or 8.

10. A highway ramp control method employing the automatic driving recognition method based on image recognition according to any one of claims 1 to 5, characterized by, The method is used to control the steering action of an autonomous vehicle on a highway; when the method is executed, firstly, the autonomous vehicle in the traffic network is identified by using the image recognition-based autonomous driving identification method of any one of claims 1-5; then the autonomous vehicle is taken as an agent, and a machine learning algorithm is used to train the agent to generate the steering action of the autonomous vehicle.

Images