Intelligent traffic signal control method and system based on traffic flow prediction

Abstract

The invention relates to an intelligent traffic signal control method and system based on traffic flow prediction, and belongs to the technical field of intelligent traffic. Traffic vehicle video streams and traffic flow data are collected through a high-definition camera and a traffic flow sensor group, different vehicles and speeds, directions and movement tracks thereof are identified through the traffic vehicle video streams, special vehicles are judged, traffic violation behaviors and road abnormal conditions are monitored according to the traffic vehicle video streams and the traffic flow data, and the traffic violation behaviors and the road abnormal conditions are monitored. A traffic prediction model is established according to traffic vehicle video streams and traffic flow historical record data, and traffic flow, vehicle positions, vehicle categories, lane line information, vehicle movement tracks, traffic violation, road abnormal conditions, and traffic flow values and vehicle speeds in future time periods of all lanes at all intersections are obtained from a server in real time. And different signal lamp control schemes are generated according to different vehicles. According to the invention, a signal control scheme is designed in combination with vehicle types, traffic violation, road abnormity and other conditions, and accurate traffic congestion dispersion is realized.

Description

Technical Field

[0001] The present invention relates to an intelligent traffic signal control method and system based on traffic flow prediction, belonging to the technical field of intelligent traffic management. Background Art

[0002] With the acceleration of urbanization and the rapid growth of motor vehicle ownership, traffic congestion has become a major challenge for urban operations. To address this issue, an increasing number of technologies are being applied to the transportation sector. However, traditional traffic signal control systems also have the following problems: fixed signal cycles make it difficult to respond to dynamic changes in traffic flow in real time, leading to increased congestion during peak hours and wasted resources during off-peak hours; the use of single magnetic induction coil data cannot effectively utilize video and other data for traffic situation identification; and the single method for handling special vehicles cannot effectively coordinate traffic conditions across the entire local area. With the increasing maturity and widespread penetration of artificial intelligence technology, it is urgent to explore intelligent traffic signal control methods and systems that, through the deep integration of data-driven and intelligent decision-making technologies, can overcome the core challenges of traditional traffic signal control, such as poor dynamic adaptability and insufficient coordination at multiple intersections, to achieve efficient traffic flow management and optimal resource allocation. Summary of the Invention

[0003] The purpose of the present invention is to overcome the deficiencies of the prior art and provide an intelligent traffic signal control method based on traffic flow prediction, which can effectively alleviate peak congestion.

[0004] The technical solution adopted by the present invention is:

[0005] The intelligent traffic signal control method based on traffic flow prediction includes the following steps:

[0006] S1 receives traffic vehicle video streams and traffic flow data through high-definition cameras and traffic flow sensor groups, and pre-processes the received data;

[0007] S2. Identify different vehicles and their speed, direction, and movement trajectory through traffic vehicle video streams, and determine whether they are special vehicles. Send the vehicle type, speed, direction, and movement trajectory information detected at each intersection to the server;

[0008] The Yolov8 model is used to detect targets in each frame of the video, obtaining vehicle position and category information, and the U-Net model is used for lane line detection. Calculating vehicle speed begins by mapping pixel coordinates to actual distance based on the video's shooting parameters and the actual scene's calibration information. The displacement between each vehicle's consecutive frames is then calculated and converted into actual speed based on the video's frame rate. The direction of motion is determined based on the change in the vehicle's center point position in consecutive frames. The vehicle's trajectory is determined by integrating the trained Yolov8 model with the DeepSORT algorithm to detect and track dynamic targets.

[0009] S3. Monitor traffic violations and road anomalies based on traffic vehicle video streams and traffic flow data. Traffic violations include running red lights and speeding, and road anomalies include sudden traffic incidents and road construction. Traffic violations and road anomalies are sent to the server, which records and saves sudden traffic incidents and road construction on each road, and records and sends penalty information to traffic violators.

[0010] Red light running detection first detects the status of the traffic light (red / green / yellow). Secondly, it determines whether the vehicle enters the stop line area during the red light period based on the detected vehicle trajectory. Finally, it determines whether the vehicle enters the stop line area based on the red light status and the vehicle trajectory.

[0011] Speeding detection is to combine the measured vehicle speed with the speed limit threshold in the traffic flow data to determine whether the vehicle is speeding;

[0012] The detection of sudden traffic incidents includes sudden braking detection and reverse driving detection. The sudden braking detection is performed by calculating the trajectory acceleration change rate (such as the speed change rate> 5m / s 2 ) judgment, the reverse detection is to determine whether the angle between the trajectory direction and the lane line direction is greater than 90°;

[0013] Road construction detection uses the method of identifying static obstacle detection and construction area characteristics to make judgments;

[0014] S4. Build a traffic prediction model based on the traffic vehicle video stream and traffic flow history data. Use the traffic prediction model to predict traffic flow values and vehicle speeds for future time periods based on real-time data and send the predictions to the server.

[0015] The vehicle motion trajectories in step S2 and the daily traffic flow data in step S1 are timestamped to form sequence data. The sequence data is used as input, the timestamp, day of the week, and holidays are used as input signals, and the traffic flow and speed are used as output. Then, a long short-term memory network is used to build a model, and the network is trained using a loss function to generate a traffic prediction model.

[0016] S5. The server obtains real-time traffic flow, vehicle location, vehicle type, lane marking information, vehicle movement trajectory, traffic violations, road anomalies, and traffic flow and speed values for each lane at each intersection. It generates customized traffic light control plans based on the specific vehicles and sends commands to each intersection for dynamic traffic light control.

[0017] According to the vehicle category determination in step S2, if there is a special vehicle, the following solution is executed:

[0018] (1) Calculate the weight of each road section based on the traffic state parameters of each road section in the road network, and then use graph theory to abstract the road network connectivity graph to establish the optimal path selection model. Use the discrete particle swarm algorithm to discretize the particle optimization process, solve the optimal path with the goal of minimizing the time it takes for the vehicle to reach the destination, and finally plan the preset driving route with the shortest travel time;

[0019] (2) Based on the determined planned path, all traffic light nodes passed through are associated, and the switching time of the traffic lights is calculated according to the path travel time. The traffic lights are also fed back according to the real-time position of the special vehicle, so that the special vehicle can get green lights all the way;

[0020] (3) Receive new obstacle events and traffic flow information from the server in real time, update the traffic status parameters of each road section in real time, dynamically adjust the driving path, and update the traffic light control plan in real time according to the new path;

[0021] According to the vehicle type determination in step S2, if there is no special vehicle, the following solution is executed:

[0022] A multi-objective optimization algorithm is used to calculate the optimal green light time and phase switching timing. The multi-objectives include minimizing average waiting time, maximizing road capacity and reducing environmental pollution. The optimization parameters are dynamically adjusted according to real-time traffic flow and road anomalies.

[0023] The data preprocessing in step S1 of the above method is to perform frame processing and information labeling on the traffic vehicle video stream data, use the mean filter algorithm to remove noise from the traffic flow data, and use the median absolute deviation algorithm to identify and eliminate outliers;

[0024] The special vehicles described in step S2 include ambulances, police cars, fire trucks, and rescue vehicles. The specific process of calculating the speed in step S2 is:

[0025] (1) First, the traffic video stream is converted into images frame by frame, and the Yolov8 model is trained to output the category and bounding box coordinates of each vehicle. The trained Yolov8 model is integrated with the DeepSORT algorithm to give each vehicle in the video stream a unique ID, and the U-Net model is used for lane line detection.

[0026] (2) Based on the video shooting parameters and by shooting an object of known size, the ratio of pixels to actual distance is calculated, and the camera intrinsic parameter matrix is used to convert the pixel coordinates into three-dimensional world coordinates, thereby establishing a mapping relationship between pixel coordinates and actual distance;

[0027] (3) For consecutive frames with the same vehicle ID, the pixel displacement of the center point of the bounding box of each consecutive frame of the vehicle is calculated. The pixel displacement is converted into actual displacement using the mapping relationship in (2). Combined with the frame rate, the speed is calculated: speed = actual displacement × frame rate.

[0028] In step S3, a multi-target tracking algorithm is introduced to ensure that the behavior of each vehicle can be accurately distinguished when multiple vehicles appear in the picture at the same time, and the identified behavior and traffic anomalies are sent to the server, which makes corresponding decisions based on the information. For the detection of sudden traffic incidents, for each vehicle trajectory, the velocity change rate (acceleration) a(t) between consecutive frames is calculated:

[0029]

[0030] If |a(t)|>5m / s 2 , output the sudden braking event record, including vehicle ID, time, and acceleration value;

[0031] Extract the direction vector of each vehicle in step S2 and lane direction vector Calculate the angle θ between the trajectory direction and the lane line direction:

[0032]

[0033] If θ>90°, the wrong-way event record is output, including vehicle ID, time, and angle value.

[0034] In step S5, a multi-objective optimization algorithm is used to calculate the optimal green light time and phase switching timing. A phase refers to the time period when a group of non-conflicting traffic flows (such as going straight, turning left, and turning right) simultaneously obtain the green light right of way. The process is as follows:

[0035] (1) Define the objective function:

[0036] Minimize average waiting time

[0037] Maximizing road capacity

[0038] Reduce environmental pollution

[0039] Where i is the vehicle number, N is the total number of vehicles, t wait,iis the waiting time of vehicle i, t free,i is the free travel time of vehicle i, t idle,i is the idling time of vehicle i, e i is the vehicle emission factor;

[0040] (2) Initialize the population and randomly generate N solutions, each of which contains the optimal green light time T g and the phase switching timing ΔT, calculate the objective function value (f1, f2, f3) of each solution, and then divide it into different levels according to the quality of the objective value, select, crossover, and mutate to generate the next generation population, and repeat the iteration until convergence outputs the optimal solution set, where 30s≤T g ≤120s, ΔT≥5s;

[0041] (3) Using the traffic prediction model in step S4 to predict traffic flow, the traffic flow in the future time period is predicted. If the traffic flow is predicted to surge, the green light time is extended in advance;

[0042] (4) If the traffic volume is detected to be greater than the traffic volume threshold (the traffic volume threshold for main roads is 1200 vehicles / hour / lane, and the traffic volume threshold for other roads is 600 vehicles / hour / lane), the green light time will be immediately extended. If no vehicles are detected during the green light period, the green light time will be shortened. If a sudden traffic incident is detected, the previous intersection will switch to red light, and the current intersection will switch to green light. Vehicles are prohibited from entering the incident area, and traffic at surrounding intersections will be prioritized.

[0043] (5) Adjust the optimized T g and ΔT are sent to the traffic signal controller.

[0044] Another object of the present invention is to provide an intelligent traffic signal control system based on traffic flow prediction for implementing the above-mentioned method, comprising monitoring equipment, a signal light group, a data processing center, and a traffic signal control service center. The monitoring equipment comprises a high-definition camera group and a traffic flow sensor group. The data processing center (corresponding to a cabinet) comprises a preprocessing module, a vehicle video stream processing module, a traffic violation and road accident detection module, a traffic prediction model construction and prediction module, and a communication module. The traffic signal control service center comprises a server and a display. The server comprises a signal control plan generation module, a control module, and a data storage and transmission module.

[0045] The monitoring equipment collects traffic vehicle video streams and traffic flow through high-definition cameras and traffic flow sensor groups; the data processing center preprocessing module preprocesses the received traffic vehicle video streams and traffic flow data; the vehicle video stream processing module identifies different vehicles and their speeds, directions, and movement trajectories based on the traffic vehicle video streams, and determines whether they are special vehicles; the traffic violation and road accident detection module monitors traffic violations and road anomalies based on traffic vehicle video streams and traffic flow data. Traffic violations include running red lights and speeding, and road anomalies include sudden traffic incidents and road construction; the traffic prediction model construction and prediction module establishes a traffic prediction model based on traffic vehicle video streams and traffic flow history data and makes real-time predictions; the communication module transmits the processed data to the server of the traffic signal control service center;

[0046] The signal control scheme generation module in the server calculates and generates the dynamic control scheme of the signal light based on the data obtained by combining and processing different vehicles; the control module controls the signal light group according to the control scheme.

[0047] The monitoring equipment and the signal light group are installed above the signal light pole. The data processing center is set in the cabinet below the signal light.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] (1) By integrating real-time traffic flow videos and traffic flow data, we can deeply mine the information value and design signal control schemes based on vehicle types, traffic violations, road anomalies, etc., to achieve precise traffic congestion relief;

[0050] (2) Based on real-time traffic flow forecast data, the traffic light switching time can be dynamically adjusted to effectively alleviate congestion during peak hours and reduce resource waste during off-peak hours;

[0051] (3) For special vehicles such as ambulances and police cars, an intelligent dispatching system is established to automatically plan the optimal route and implement dynamic signal priority control;

[0052] (4) Use advanced intelligent technology to record and analyze violations in real time, improve the effectiveness of civilized driving supervision, and build a solid line of defense for traffic safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 Flow chart of the method of the present invention;

[0054] Figure 2 Design a flow chart for the signal light control scheme of the present invention;

[0055] Figure 3 This is the external structure diagram of the system of the present invention;

[0056] Figure 4 This is a schematic diagram of the system structure of the present invention;

[0057] Figure 5 Schematic diagram of the cabinet structure;

[0058] Among them, 1. High-definition camera, 2. Traffic flow sensor group, 3. Traffic light group, 4. Traffic light pole, 5. Cabinet, 6. Countdown display, 7. Server, 8. Display. DETAILED DESCRIPTION

[0059] The present invention is further described below with reference to specific embodiments.

[0060] Example 1 An intelligent traffic signal control method based on traffic flow prediction includes the following steps:

[0061] S1. Receive vehicle video streams and traffic flow data through a high-definition camera and traffic flow sensor group, and preprocess the received data:

[0062] The traffic vehicle video stream data is frame processed and information labeled, where the information labeling includes vehicle information, lane lines, and signal light area labeling; the traffic flow data is subjected to noise and outlier removal, and the mean filtering algorithm is used to remove noise from the traffic flow data, and the median absolute deviation algorithm is used to identify and eliminate outliers in the traffic flow data.

[0063] S2. Identify different vehicles and their speed, direction, and trajectory through the traffic video stream, determine whether they are special vehicles, and send the vehicle type, speed, direction, and trajectory information detected at each intersection to the server:

[0064] (1) First, the traffic video stream is converted into images frame by frame, and the images are scaled to 640×640. The U-Net model is used for lane line recognition, and the Yolov8 model is trained to output the category and bounding box coordinates of each vehicle.

[0065] (2) Based on the video focus, sensor size and other shooting parameters and an object of known size (such as a calibration plate), the ratio of pixels to actual distance is calculated, and the camera intrinsic parameter matrix (K) is used to convert the pixel coordinates into three-dimensional world coordinates, thereby establishing a mapping relationship between pixel coordinates and actual distance;

[0066] (3) For consecutive frames with the same vehicle ID, calculate the pixel displacement of the center point of the bounding box of each consecutive frame of the vehicle. Use the mapping relationship in step (2) to convert the pixel displacement into actual displacement, and combine it with the frame rate to calculate the speed: speed = actual displacement × frame rate;

[0067] (4) The direction of movement is determined based on the change in the center point position of the vehicle in consecutive frames. The center point of the vehicle is projected onto the nearest lane line, and the vehicle direction is corrected based on the direction of the lane line to further determine the driving direction.

[0068] (5) The bounding box coordinates, category and other information of each target obtained by the Yolov8 model, as well as the recorded motion trajectory information of each target, are input into DeepSORT. In the first frame of the video, all vehicles detected by YOLOv8 are initialized as new tracking targets and assigned unique IDs. The ReID (re-identification) model is used to extract the appearance features of each vehicle to distinguish different vehicles. The Kalman filter is then used to predict the motion trajectory of the target and the filter is used to track the target. If the target is occluded or crossed, DeepSORT makes corrections based on historical information to ensure the accuracy of tracking, thereby realizing the detection and tracking of dynamic targets and predicting the vehicle motion trajectory.

[0069] S3. Monitor traffic violations and road anomalies based on vehicle video streams and traffic flow data. Traffic violations include running red lights and speeding, and road anomalies include sudden traffic incidents and road construction. These violations and road anomalies are sent to the server, which records and saves sudden traffic incidents and road construction on each road. The server also records and sends penalty information for vehicles violating traffic regulations.

[0070] (1) Use the YOLOv8-tiny target detection model to locate the traffic light area, then perform color segmentation on the traffic light area and output the red / yellow / green status. The motion trajectory of each vehicle obtained by the above method is used to extract the stop line position through the U-net model. For each vehicle trajectory, the red light running condition is judged.

[0071] Case 1: The traffic light is red at time t, the center coordinate of the vehicle is behind the stop line at time t, and it crosses the stop line at time t+1, and the traffic light is red at time t+1, which is considered a red light violation;

[0072] Case 2: At time t, the traffic light is red, the center point of the vehicle is in front of the stop line, the traffic light is red at time t+1, the vehicle continues to move through the intersection at time t+1, and if there are no special vehicles, it is judged as running a red light;

[0073] Case 3: At time t, the traffic light is red, and the center coordinates of the vehicle are located in front of the stop line. At time t+1, the traffic light is red, and the center coordinates of the vehicle are located behind the stop line. This is considered a red light violation.

[0074] Finally, the red light running event record is output, including vehicle ID, time, and location;

[0075] (2) Extract the instantaneous speed of each vehicle calculated above, extract the speed limit of the current road section from the traffic flow data, and for each vehicle, determine whether there is a time point t that satisfies: vehicle speed (t) > speed limit. If so, output the speeding event record, including vehicle ID, time, and speed value;

[0076] (3) For each vehicle’s trajectory, calculate the velocity change rate (acceleration) a(t) between consecutive frames:

[0077]

[0078] If |a(t)|>5m / s 2 , output the sudden braking event record, including vehicle ID, time, and acceleration value;

[0079] Extract the direction vector of each vehicle calculated above and lane direction vector Calculate the angle θ between the trajectory direction and the lane line direction:

[0080]

[0081] If θ>90°, the wrong-way event record is output, including vehicle ID, time, and angle value.

[0082] (4) The Yolov8 target detection model is used to locate obstacles on the road. Then, the system identifies whether there are orange or yellow highlighted areas, identifies specific shapes such as cones and water barriers, and determines that if the obstacles are densely distributed and form a closed area, it is determined to be a construction area. Then, the confidence of the construction area is generated by combining the obstacle type, color, and spatial distribution. If the confidence exceeds the threshold (such as 0.8), it is output as a construction event record, which includes the location of the construction area and the confidence.

[0083] (5) In order to ensure that the behaviors can be accurately distinguished when multiple vehicles appear at the same time, each vehicle is assigned a unique ID and cross-frame association is performed. The vehicle's appearance characteristics such as color and model and its movement characteristics such as speed and direction are combined to improve tracking stability. The trajectory and behavior characteristics of each vehicle are classified as: running a red light, speeding, sudden braking, driving against traffic, and normal driving. The detected behaviors and anomalies are then sent to the server via the HTTP protocol.

[0084] S4. Build a traffic prediction model based on the traffic vehicle video stream and traffic flow history data. Use the traffic prediction model to predict traffic flow values and vehicle speeds for future time periods based on real-time data and send the predictions to the server.

[0085] (1) Data processing: Align the calculated vehicle trajectory data and daily traffic flow data according to the timestamps so that the data at the same time point can correspond;

[0086] (2) Model construction: The vehicle movement trajectory in S2 and the daily traffic flow data in step S1 are aligned with the timestamps to form sequence data. The sequence data is used as input, the timestamp, day of the week, and holidays are used as input signals, and the traffic flow and speed are used as output. The LSTM network is selected for model construction. The multi-layer LSTM unit captures the long-term dependency in the input time series. The LSTM output is connected to the fully connected layer for feature integration and prediction, and the output is the traffic flow and speed.

[0087] (3) Model training: The data was divided into training set, validation set and test set according to the ratio of 8:1:1. The Adam optimizer was used, and the learning rate was set to 0.001. The model was trained using the training set, and the model parameters were updated through the back propagation algorithm. During the training process, the validation set was used to monitor the model loss value and accuracy. The test set was used to calculate the error index between the predicted value and the true value. The error index used was the mean square error and the mean absolute value.

[0088] (4) Real-time prediction and deployment: The real-time traffic video stream and historical record data are input into the trained model, and the predicted traffic flow value and vehicle speed are output. These are used as input for the subsequent dynamic adjustment of traffic lights. The prediction results are stored in the server. The server compares the predicted vehicle speed with the lane speed limit and issues a warning if the vehicle is about to exceed the speed limit.

[0089] (5) Model optimization: Based on the real-time prediction performance and feedback, the model structure is adjusted, features are added, and hyperparameters are adjusted. The model is also updated regularly and trained with new data to adapt to changes in traffic conditions.

[0090] S5. The server obtains real-time traffic flow, vehicle location, vehicle type, lane marking information, vehicle movement trajectory, traffic violations, road anomalies, and traffic flow and speed values for each lane at each intersection. It generates customized traffic light control plans based on the specific vehicles and sends commands to each intersection for dynamic traffic light control.

[0091] Determine whether there is special vehicle information based on the vehicle type. If so, continue to execute the following plan:

[0092] a. Calculate the weight of each road section based on the traffic state parameters of each road section in the road network, and then use the graph theory to abstract the road network connectivity graph to establish the optimal path selection model. Use the discrete particle swarm algorithm to discretize the particle optimization process, solve the optimal path with the goal of minimizing the time it takes for the vehicle to reach the destination, and finally plan the preset driving route with the shortest time ([1] Li Xue. Optimization and Implementation of Green Channel for Special Authorized Vehicles [D]. Nanjing University of Information Science and Technology, 2021);

[0093] b. Based on the determined planned path, all traffic light nodes are associated, and the traffic light switching time is calculated based on the path travel time. The traffic light is then fed back based on the real-time location of the special vehicle, ensuring that the special vehicle has green lights all the way.

[0094] c. Receive new obstacle events and traffic flow information from the server in real time, update the traffic status parameters of each road section in real time, dynamically adjust the driving path, and update the traffic light control plan in real time based on the new path.

[0095] Determine whether there is special vehicle information based on the vehicle type. If not, execute the following solution:

[0096] (1) Define the objective function:

[0097] Minimize average waiting time

[0098] Maximizing road capacity

[0099] Reduce environmental pollution

[0100] Where i is the vehicle number, N is the total number of vehicles, t wait,i is the waiting time of vehicle i, t free,i is the free travel time of vehicle i, t idle,i is the idling time of vehicle i, e i is the vehicle emission factor; (the objective function is calculated for each intersection. The vehicle number can be obtained directly, and the total number of vehicles can also be calculated based on the number. The waiting time can be calculated based on the time spent at the intersection. The free travel time is the travel time calculated when there is no traffic jam and the traffic lights are normal. The idling time is the time spent waiting at red lights and the time spent stopping in congestion. The vehicle emission factor is a specific parameter and is queried from the database).

[0101] (2) Initialize the population and randomly generate N solutions, each of which contains the optimal green light time T g and the phase switching timing ΔT, calculate the objective function value (f1, f2, f3) of each solution, and then divide it into different levels according to the quality of the objective value, select, crossover, and mutate to generate the next generation population, and repeat the iteration until convergence outputs the optimal solution set, where 30s≤T g ≤120s, ΔT≥5s;

[0102] (3) Using the traffic prediction model in step S4 to predict traffic flow, the traffic flow in the future time period is predicted. If the traffic flow is predicted to surge, the green light time is extended in advance;

[0103] (4) If the traffic volume is detected to be greater than the traffic volume threshold (the traffic volume threshold for main roads is 1200 vehicles / hour / lane, and the traffic volume threshold for other roads is 600 vehicles / hour / lane), the green light time will be immediately extended. If no vehicles are detected during the green light period, the green light time will be shortened. If a sudden traffic incident is detected, the previous intersection will switch to red light, and the current intersection will switch to green light. Vehicles are prohibited from entering the incident area, and traffic at surrounding intersections will be prioritized.

[0104] (5) The optimized T g And ΔT are sent to the traffic signal controller, such as direction A: the green light time is adjusted from 60 seconds to 75 seconds, direction B: the green light time is adjusted from 40 seconds to 30 seconds;

[0105] Real-time detection of average waiting time, traffic capacity, and emissions: a 5% reduction. If performance does not meet the standards, parameter fine-tuning is triggered, and data collection, optimization, and control are re-executed every minute to accumulate historical data and regularly update noise filter models and optimization algorithm parameters.

[0106] Example 2: Intelligent traffic signal control system based on traffic flow prediction (see Figure 3-Figure 5 ) includes monitoring equipment, a signal light group 3, a data processing center, and a traffic signal control service center. The monitoring equipment includes a high-definition camera group and a traffic flow sensor group. The data processing center (corresponding to the cabinet) includes a pre-processing module, a vehicle video stream processing module, a traffic violation and road accident detection module, a traffic prediction model construction and prediction module, and a communication module. The traffic signal control service center includes a server 7 and a display 8. The server includes a signal control plan generation module, a control module, and a data storage and transmission module.

[0107] The monitoring equipment collects traffic vehicle video streams and traffic flow through a high-definition camera 1 and a traffic flow sensor group 2. The data processing center preprocesses the received traffic vehicle video streams and traffic flow data. The vehicle video stream processing module identifies different vehicles and their speeds, directions, and movement trajectories based on the traffic vehicle video streams, and determines whether they are special vehicles. The traffic violation and road accident detection module monitors traffic violations and road anomalies based on the traffic vehicle video streams and traffic flow data. Traffic violations include running red lights and speeding, and road anomalies include sudden traffic incidents and road construction. The traffic prediction model construction and prediction module establishes a traffic prediction model based on the traffic vehicle video streams and traffic flow history data and makes real-time predictions. The communication module transmits the processed data to the server of the traffic signal control service center.

[0108] The signal control scheme generation module in the server calculates and generates the dynamic control scheme of the signal light based on the data obtained by combining and processing different vehicles; the control module controls the signal light group according to the control scheme.

[0109] The monitoring equipment and the signal light group are installed above the signal light pole 4. The data processing center is arranged in a cabinet 5 below the signal light. A countdown display 6 is provided on the signal light pole 4.

[0110] The above is a further description of the present invention in conjunction with the embodiments, and the protection scope of the present invention is not limited thereto.

Claims

1. An intelligent traffic signal control method based on traffic flow prediction is characterized by: The steps are as follows: S1 receives traffic vehicle video streams and traffic flow data through high-definition cameras and traffic flow sensor groups, and pre-processes the received data; S2. Identify different vehicles and their speed, direction, and movement trajectory through traffic vehicle video streams, and determine whether they are special vehicles. Send the vehicle type, speed, direction, and movement trajectory information detected at each intersection to the server; S3. Monitor traffic violations and road anomalies based on traffic vehicle video streams and traffic flow data. Traffic violations include running red lights and speeding, and road anomalies include sudden traffic incidents and road construction. Traffic violations and road anomalies are sent to the server, which records and saves sudden traffic incidents and road construction on each road, and records and sends penalty information to traffic violators. S4. Build a traffic prediction model based on the traffic vehicle video stream and traffic flow history data. Use the traffic prediction model to predict traffic flow values and vehicle speeds for future time periods based on real-time data and send the predictions to the server. S5. The server obtains real-time traffic flow, vehicle location, vehicle type, lane marking information, vehicle movement trajectory, traffic violations, road anomalies, and traffic flow and speed values for each lane at each intersection. It generates customized traffic light control plans based on the specific vehicles and sends commands to each intersection for dynamic traffic light control. According to the vehicle category determination in step S2, if there is a special vehicle, the following solution is executed: (1) The weight of each road section is calculated based on the traffic state parameters of each road section in the road network. Then, the optimal path selection model is established on the road network connectivity graph abstracted by graph theory. The particle optimization process is discretized using the discrete particle swarm algorithm. The optimal path is solved with the goal of minimizing the time it takes for the vehicle to reach the destination, and finally a preset driving route with the shortest travel time is planned; (2) Based on the determined planned path, all traffic light nodes passed through are associated, and the switching time of the traffic lights is calculated according to the path travel time. The traffic lights are also fed back according to the real-time position of the special vehicle, so that the special vehicle can get green lights all the way; (3) Receive new obstacle events and traffic flow information from the server in real time, update the traffic status parameters of each road section in real time, dynamically adjust the driving path, and update the traffic light control plan in real time according to the new path; According to the vehicle type determination in step S2, if there is no special vehicle, the following solution is executed: A multi-objective optimization algorithm is used to calculate the optimal green light time and phase switching timing. The multi-objectives include minimizing average waiting time, maximizing road capacity and reducing environmental pollution. The optimization parameters are dynamically adjusted according to real-time traffic flow and road anomalies.

2. The intelligent traffic signal control method based on traffic flow prediction according to claim 1 is characterized in that: In step S2, the Yolov8 model is used to detect targets in each frame of the video to obtain vehicle position and category information, and the U-Net model is used to detect lane lines. The vehicle speed is calculated by first establishing a mapping relationship between pixel coordinates and actual distance based on the video shooting parameters and the calibration information of the actual scene. Then, the displacement between consecutive frames of each vehicle is calculated, and combined with the frame rate of the video, the displacement is finally converted into actual speed. The direction of movement is determined by calculating the change in the center point position of the vehicle in consecutive frames. The vehicle motion trajectory is achieved by integrating the trained Yolov8 model with the DeepSORT algorithm to achieve dynamic target detection and tracking.

3. The intelligent traffic signal control method based on traffic flow prediction according to claim 2 is characterized in that: The specific process of calculating the speed in step S2 is: (1) First, the traffic video stream is converted into images frame by frame, and the Yolov8 model is trained to output the category and bounding box coordinates of each vehicle. The trained Yolov8 model is integrated with the DeepSORT algorithm to give each vehicle in the video stream a unique ID, and the U-Net model is used for lane line detection. (2) Based on the video shooting parameters and by shooting an object of known size, the ratio of pixels to actual distance is calculated, and the camera intrinsic parameter matrix is used to convert the pixel coordinates into three-dimensional world coordinates, thereby establishing a mapping relationship between pixel coordinates and actual distance; (3) For consecutive frames with the same vehicle ID, the pixel displacement of the center point of the bounding box of each consecutive frame of the vehicle is calculated. The pixel displacement is converted into actual displacement using the mapping relationship in (2). Combined with the frame rate, the speed is calculated: speed = actual displacement × frame rate.

4. The intelligent traffic signal control method based on traffic flow prediction according to claim 1 is characterized in that: The special vehicles described in step S2 include ambulances, police cars, fire trucks, and rescue vehicles.

5. The intelligent traffic signal control method based on traffic flow prediction according to claim 1 is characterized in that: In step S3, the red light running behavior detection first detects the signal light status, then determines whether the vehicle enters the stop line area during the red light period based on the detected vehicle movement trajectory, and finally determines whether the vehicle enters the stop line area based on the red light status and the vehicle trajectory; Speeding detection is to combine the measured vehicle speed with the speed limit threshold in the traffic flow data to determine whether the vehicle is speeding; Traffic emergency detection includes sudden braking detection and wrong-way detection. The sudden braking detection is determined by calculating the trajectory acceleration change rate, and the wrong-way detection is determined by determining whether the angle between the trajectory direction and the lane line direction is greater than 90°. Road construction detection uses the method of identifying static obstacle detection and construction area characteristics to make judgments.

6. The intelligent traffic signal control method based on traffic flow prediction according to claim 5 is characterized in that: For the detection of sudden traffic incidents, for each vehicle’s trajectory, the speed change rate a(t) between consecutive frames is calculated: If |a(t)|>5m / s 2 , output the sudden braking event record, including vehicle ID, time, and acceleration value; Extract the direction vector of each vehicle in step S2 and lane direction vector Calculate the angle θ between the trajectory direction and the lane line direction: If θ>90°, the wrong-way event record is output, including vehicle ID, time, and angle value.

7. The intelligent traffic signal control method based on traffic flow prediction according to claim 1 is characterized in that: The construction of the traffic prediction model in step S4 is to align the vehicle movement trajectory in step S2 and the daily traffic flow data in step S1 with the timestamps to form sequence data, with the sequence data as input, the timestamp, week number, and holiday as input signals, and the vehicle flow and vehicle speed as output. Then, the long short-term memory network is used to build the model, and the network is trained through the loss function to generate the traffic prediction model.

8. The intelligent traffic signal control method based on traffic flow prediction according to claim 1 is characterized in that: In step S5, the multi-objective optimization algorithm is used to calculate the optimal green light time and phase switching timing. The process is as follows: (1) Define the objective function: Minimize average waiting time Maximizing road capacity Reduce environmental pollution Where i is the vehicle number, N is the total number of vehicles, t wait,i is the waiting time of vehicle i, t free,i is the free travel time of vehicle i, t idle,i is the idling time of vehicle i, e i is the vehicle emission factor; (2) Initialize the population and randomly generate N solutions, each of which contains the optimal green light time T g and the phase switching timing ΔT, calculate the objective function value (f1, f2, f3) of each solution, and then divide it into different levels according to the quality of the objective value, select, crossover, and mutate to generate the next generation population, and repeat the iteration until convergence outputs the optimal solution set, where 30s≤T g ≤120s, ΔT≥5s; (3) Using the traffic prediction model in step S4 to predict traffic flow, the traffic flow in the future time period is predicted. If the traffic flow is predicted to surge, the green light time is extended in advance; (4) If the traffic volume is detected to be greater than the traffic volume threshold, the green light time is immediately extended. If no vehicles are detected during the green light period, the green light time is shortened. If a sudden traffic incident is detected, the previous intersection is switched to red light, and the current intersection is switched to green light. Vehicles are prohibited from entering the incident area, and traffic at surrounding intersections is prioritized. (5) Adjust the optimized T g and ΔT are sent to the traffic signal controller.

9. Intelligent traffic signal control system based on traffic flow prediction, characterized by: It includes monitoring equipment, signal light groups, data processing center and traffic signal control service center. The monitoring equipment includes high-definition camera groups and traffic flow sensor groups. The data processing center includes pre-processing module, vehicle video stream processing module, traffic violation and road accident detection module, traffic prediction model construction and prediction module, communication module. The traffic signal control service center includes servers and displays. The server includes signal control plan generation module, control module, data storage and transmission module. The monitoring equipment collects traffic vehicle video streams and traffic flow through high-definition cameras and traffic flow sensor groups; the data processing center preprocessing module preprocesses the received traffic vehicle video streams and traffic flow data; the vehicle video stream processing module identifies different vehicles and their speeds, directions, and movement trajectories based on the traffic vehicle video streams, and determines whether they are special vehicles; the traffic violation and road accident detection module monitors traffic violations and road anomalies based on traffic vehicle video streams and traffic flow data. Traffic violations include running red lights and speeding, and road anomalies include sudden traffic incidents and road construction; the traffic prediction model construction and prediction module establishes a traffic prediction model based on traffic vehicle video streams and traffic flow history data and makes real-time predictions; the communication module transmits the processed data to the server of the traffic signal control service center; The signal control scheme generation module in the server calculates and generates the dynamic control scheme of the signal light based on the data obtained by combining and processing different vehicles; the control module controls the signal light group according to the control scheme.

10. The intelligent traffic signal control system based on traffic flow prediction according to claim 9 is characterized in that: The monitoring equipment and the signal light group are installed above the signal light pole, and the data processing center is arranged in a cabinet below the signal light.

Images